Graphics cards / Compare

GIGABYTE AORUS GeForce RTX 5090 MASTER ICE 32G vs INNO3D GeForce RTX 5090 X3 OC

A

Product A: GIGABYTE

AORUS GeForce RTX 5090 MASTER ICE 32G

74 of 100 known

Change GIGABYTE AORUS GeForce RTX 5090 MASTER ICE 32G

Compare the INNO3D GeForce RTX 5090 X3 OC with

Search graphics cards

B

Product B: INNO3D

GeForce RTX 5090 X3 OC

63 of 100 known

Change INNO3D GeForce RTX 5090 X3 OC

Compare the GIGABYTE AORUS GeForce RTX 5090 MASTER ICE 32G with

Search graphics cards

All 100 fields

Every value links to where we found it.

Only differences

AGIGABYTE AORUS GeForce RTX 5090 MASTER ICE 32G

BINNO3D GeForce RTX 5090 X3 OC

General

3 of 10 differ

Series

AORUS Master2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Brand GIGABYTE, product family AORUS, product name AORUS GeForce RTX 5090 MASTER ICE 32G. The MASTER tier of the AORUS family is stored as the line.

X31Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Product name INNO3D GeForce RTX 5090 X3 OC. The breadcrumb shows part number N50903-32D7X-17593928. The line is X3.

Model numbersiMaker identifiers for this exact model and its listed variants. Excludes sibling models and other generations.

GV-N5090AORUSM ICE-32GD1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Spec table row Model: GV-N5090AORUSM ICE-32GD. Vietnamese retailer copy of the Gigabyte spec tab.2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Product code GV-N5090AORUSM ICE-32GD. This is a catalog datasheet (publication date 06-FEB-2026, Spanish) that a Spanish retailer hosts. It carries Gigabyte's data and Gigabyte's product name.

N50903-32D7X-175939281Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Breadcrumb on the product page: N50903-32D7X-17593928. The non-OC X3 is N50903-32D7-17593928 and has its own page. It is a separate product.4Spec databasebuildcores.com/products/GPU/n1nvzipf0/Inno3D-X3-OC-GeForce-RTX-5090-32GB-GDDR7Read 16 October 2026Part Numbers: N50903-32D7X-17593928.

GTIN / EAN

4719331355906, 8895230485972Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Barcodes on page 3: 4719331355906 (EAN) and 889523048597 (UPC, also printed as 0889523048597).

88863077001314Spec databasebuildcores.com/products/GPU/n1nvzipf0/Inno3D-X3-OC-GeForce-RTX-5090-32GB-GDDR7Read 16 October 2026EAN 8886307700131. The same EAN is on the blinQ and Eltel listings for part number N50903-32D7X-17593928. BuildCores lists no UPC.5Retailer pageshop.blinq.no/inno3d-geforce-rtx-5090-x3/cat-p/c/p1005852980Read 16 October 2026Item number N50903-32D7X-17593928, EAN 8886307700131.

AnnouncediThe maker's first announcement of this model. This is separate from first retail availability.

7 January 20257Press articletomshardware.com/pc-components/gpus/gigabyte-releases-27-new-nvidia-rtx-50-series-gpus-at-ces-2025-3-aorus-radeon-rx-9000-gpus-also-announcedRead 15 October 2026Published 2025-01-07 (CES 2025). Its table lists Aorus GeForce RTX 5090 Master Ice 32G: 28Gbps, 32 GB GDDR7, 512-bit, 16-pin x 1, PCIe 5.0, price TBD.

Not available

Status

Available4Retailer pageldlc.com/en/product/PB00662770.htmlRead 15 October 2026LDLC (France) lists the card as in stock with an Add to basket button on 2026-10-15. Its Power consumption: 600 W row is a retail feed value and is not stored. Its 2 year warranty is the seller's term.

Available3Retailer pagecoolblue.nl/en/product/963046/inno3d-geforce-rtx-5090-x3-oc-32gb.htmlRead 16 October 2026Coolblue lists the card with manufacturer code N50903-32D7X-17593928 and year introduced 2025. It is out of stock there on 2026-10-16. INNO3D's product page is live. See also Geizhals.6Spec databasegeizhals.de/inno3d-geforce-rtx-5090-x3-oc-n50903-32d7x-17593928-a3444085.htmlRead 16 October 2026On 2026-10-16 the page lists 2 offers for this card from EUR 7456. Scan marks the card End Of Life and Coolblue shows it out of stock, so it is sold from remaining stock.

PredecessoriThe model this one directly replaced, in the same line, tier and size, as the maker or a trusted review names it.

Not available

Not available

SuccessoriThe model that directly replaced this one, in the same line, tier and size. A revision such as a Slim or OLED model does not count when a next generation exists.

Not available

Not available

Launch price (US)iThe maker's US launch MSRP for the stated configuration, excluding tax. Current store prices and later discounts do not qualify.

Not available

Not available

Launch price (EU)iThe maker's euro launch price for the stated configuration, including VAT as published. Current store prices and later discounts do not qualify.

Not available

Not available

MarketsiISO country or region codes where this exact model or variant is sold. A source page's language does not establish availability.

Not available

Not available

ColoursiThe maker's own colour names

White2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Product color: Blanco (White).3Retailer pageparadigit.ie/gigabyte-aorus-geforce-rtx-5090-master-ice-32g/21757552/productRead 15 October 2026Colour of product: White.

Not available

GPUiThe graphics chip as the chip maker names it, without the card maker's name or the memory size, e.g. "NVIDIA GeForce RTX 5070 Ti", "AMD Radeon RX 9070 XT", "Intel Arc B580". The memory size is in VRAM.

NVIDIA GeForce RTX 50909Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026Specs table for the GeForce RTX 5090. The copy read here still had price and date placeholders.

NVIDIA GeForce RTX 50901Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Specification title: INNO3D GeForce RTX 5090 X3 OC. Feature text: NVIDIA GeForce RTX 5090.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026Page title and specification table: GeForce RTX 5090.

GPU brandiThe chip maker, not the card maker.

NVIDIA9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026

NVIDIA8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026

Reference launch price (US)iThe chip maker's launch price for this GPU in USD, excluding tax. It repeats in every card with this GPU and is not the card's own price. A price the card maker states goes in the launch price field.

$1,99912Press releasenvidianews.nvidia.com/news/nvidia-blackwell-geforce-rtx-50-series-opens-new-world-of-ai-computer-graphicsRead 15 October 2026The RTX 5090 is priced at $1,999 and is available starting January 30 (2025).

$1,99910Press releasenvidianews.nvidia.com/news/nvidia-blackwell-geforce-rtx-50-series-opens-new-world-of-ai-computer-graphicsRead 16 October 2026The GeForce RTX 5090 will be available on Jan. 30 at $1,999. The release is dated 2025-01-06.

GPU

0 of 14 differ

GPU architectureiAs the chip maker names it, e.g. "Blackwell", "RDNA 4", "Xe2".

Blackwell9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026Architecture: Blackwell.

Blackwell9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: GPU Architecture NVIDIA Blackwell.

GPU dieiThe die name, e.g. "GB203", "Navi 48", "BMG-G21". Cards with the same GPU name can use different dies.

GB20210Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Appendix A: Blackwell GB202 GPU. The RTX 5090 uses GB202.

GB2029Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: GPU codename GB202.

ProcessiFoundry process name as the chip maker writes it, e.g. "TSMC 4N". When the chip maker names none, a spec table in a trusted review counts, and the source note says so. Node names are marketing names, not a physical size. Write the foundry and its node name with no unit, such as "Samsung 8N", not "Samsung 8 nm".

TSMC 4N10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: TSMC 4nm 4N, NVIDIA custom process.

TSMC 4N9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: TSMC 4nm 4N, NVIDIA Custom Process.

TransistorsiTransistors on the whole die, in billions. A cut-down GPU keeps the full count of its die.

92.2billion10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 92.2 billion for the full GB202 chip.

92.2billion9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Appendix A: 92.2 billion for the whole GB202 die.

Die sizeiArea of the whole die.

750mm²10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 750 mm2.

750mm²9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Appendix A table: 750 mm2.

Compute unitsiNVIDIA streaming multiprocessors (SM), AMD compute units (CU) or Intel Xe-cores, enabled on this GPU. The blocks differ between makers, so compare within one maker.

17010Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 170 SMs.

1709Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: 170 SMs.

Shader coresiNVIDIA CUDA cores or AMD stream processors, as published. For Intel only a shader count a trusted source states, not a count of Xe Vector Engines. Makers count differently, so compare within one maker.

21,7609Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026CUDA cores: 21,760.10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 21760 CUDA cores.

21,7601Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: CUDA Cores 21760.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026NVIDIA CUDA Cores: 21760.

Ray tracing coresiNVIDIA RT cores, AMD Ray Accelerators or Intel Ray Tracing Units, enabled on this GPU.

17010Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 170 RT cores, 4th generation.

1709Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: 170 RT cores (4th Gen).

AI coresiNVIDIA Tensor cores, AMD AI Accelerators or Intel XMX engines, enabled on this GPU.

68010Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 680 Tensor cores, 5th generation.

6809Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: 680 Tensor cores (5th Gen).

ROPsiRender output units enabled on this GPU.

17610Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 176 ROPs.

1769Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: 176 ROPs.

L2 cacheiL2 cache enabled on this GPU.

96MB10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: L2 cache 98304 KB, which is 96 MB.

96MB9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: 98304 KB, which is 96 MB.

Infinity Cache (L3)iAMD Infinity Cache, a last-level cache behind L2. Left out for a GPU whose maker states no such cache.

Not available

Not available

FP32 compute (claimed)iPeak single-precision shader throughput the chip maker publishes. Makers count operations differently. AMD includes dual-issue rates. Not a game performance figure.

104.8TFLOPS10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: Peak FP32 TFLOPS (non-Tensor) 104.8, based on the reference boost clock.

104.8TFLOPS9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Peak FP32 TFLOPS (non-Tensor) 104.8, based on the 2407 MHz GPU Boost Clock.

AI performance (claimed)iPeak AI throughput the maker publishes, one row per precision and sparsity. A figure without a stated precision stays out. Compare rows with the same precision and sparsity only. NVIDIA: store the headline rows (FP8 and FP16 with FP16 accumulate, FP4 as listed) and name the accumulate precision in the source note. When a maker does not say whether a figure uses sparsity, leave the sparse key out of that row. The chip maker's figures are stored, the same in every card with that GPU. A card maker's higher figure for its own clocks, such as an OC mode, goes in the source note.

AGIGABYTE AORUS GeForce RTX 5090 MASTER ICE 32G

Precision	With sparsity	Throughput (TOPS)
FP4	No	1,676
FP4	Yes	3,352
FP8	No	838
FP8	Yes	1,676
FP16	No	419
FP16	Yes	838
INT8	No	838
INT8	Yes	1,676

Source10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: FP4 1676/3352 (FP32 accumulate), FP8 838/1676 (FP16 accumulate), FP16 419/838 (FP16 accumulate), INT8 838/1676. The second figure uses sparsity. Dense and sparse rows are stored.

BINNO3D GeForce RTX 5090 X3 OC

Precision	With sparsity	Throughput (TOPS)
FP4	No	1,676
FP4	Yes	3,352
FP8	No	838
FP8	Yes	1,676
FP16	No	419
FP16	Yes	838
INT8	No	838
INT8	Yes	1,676

Source9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3, dense and effective with sparsity. FP4 uses FP32 accumulate (1676 and 3352). FP8 and FP16 use FP16 accumulate. FP8 with FP32 accumulate is half as high.

Clocks

2 of 5 differ

Base clockiLowest rated clock under load that the card maker states for this card in MHz. NVIDIA's base clock in GHz is a chip value, and a figure such as 2.33 GHz gives no exact MHz value, so it is not converted. A row labelled only "Clock Speed" next to a boost clock row, as on PNY, is the base clock. On an NVIDIA card, a row labelled Graphics Clock next to a Boost Clock row, as on Palit, is the base clock too. A retail or distributor feed that repeats NVIDIA's reference base clock does not count. A card maker's own page that states the reference base clock for a card at reference clocks counts.

Not available

2,017MHz1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet, GPU Engine Specs: Base Clock 2017 MHz. INNO3D states it for this part number. It is the same as NVIDIA's reference base clock, so the card runs its base at reference.

Game clockiAMD Game Clock or Intel Graphics Clock of this card. Both are a typical clock in games, below the peak boost. On an Intel card, a card maker figure labelled Graphics Clock or Engine Clock is this clock, also with "up to". A figure the card maker labels Boost Clock stays a boost clock.

Not available

Not available

Boost clockiRated boost clock of this card out of the box, e.g. ASUS "Default mode" or MSI "Boost". Not a measured clock, the OC mode clock or another card's rating. AMD's boost clock is a maximum burst clock. NVIDIA's is its rated boost clock. Compare within one chip maker only. When the card ships in its OC BIOS position, as the PowerColor Hellhound does, that BIOS clock is stored here and the OC mode field stays empty. The other BIOS clock goes in the source note.

2,655MHz1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Core clock: 2655 MHz (Reference card: 2407MHz). The row does not name a BIOS mode. Gigabyte's dual BIOS has Performance and Silent. A Memory Express copy of the Gigabyte page says Performance is the factory default, but that page was not readable here. No source states the Silent clock or a separate OC mode clock, so the OC mode field stays empty.8Reviewnoobfeed.com/hardware/gigabyte-aorus-geforce-rtx-5090-master-ice-reviewRead 15 October 2026Out of the box the card boosts to 2,655MHz, against 2,407MHz on the Founders Edition. The review also logs an average of 2,820MHz in a one-hour F1 24 loop with the optional reverse fan fitted. That is not stored, because the fan is not the default setup and the review states no resolution.

2,452MHz1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Boost Clock 2452 MHz. NVIDIA's reference boost is 2407 MHz, so this is the factory overclock.

Boost clock in OC modeiBoost clock the card maker states for its OC mode, set in its software or with a BIOS switch, e.g. ASUS "OC mode" or MSI "Extreme Performance".

Not available

Not available

Reference boost clockiThe chip maker's boost clock for this GPU. It shows how far the card is factory overclocked.

2,407MHz1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Core clock row: Reference card 2407MHz.9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026Boost clock 2.41 GHz. The exact 2407 MHz is in the whitepaper and in Gigabyte's reference row.10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: GPU boost clock 2407 MHz.

2,407MHz9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: GPU Boost Clock 2407 MHz.

Memory

0 of 5 differ

VRAMiGraphics memory on the card. A card with another memory size is another product.

32GB1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 202632 GB GDDR7.9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 202632 GB GDDR7.10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 32 GB GDDR7.

32GB1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Standard Memory Config 32GB.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026

VRAM type

GDDR71Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 202610Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: GDDR7.

GDDR71Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: the Memory Interface row says GDDR7 and the Memory Interface Width row says 512-bit. The labels are swapped on the page.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 20269Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: 32 GB GDDR7.

Memory bus width

512bits1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Memory bus: 512 bit.9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026Bus width 512-bit.10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 512-bit.

512bits1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Memory Interface Width 512-bit.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026

Memory speediEffective data rate per pin, e.g. 28 for GDDR7 at 28 Gbps. A memory clock published as 28000 MHz effective is 28 Gbps.

28Gbps1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Memory speed: 28 Gbps.10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: memory clock (data rate) 28 Gbps.

28Gbps1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Memory Clock 28Gbps.9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: Memory Clock (Data Rate) 28 Gbps.

Memory bandwidthiPeak bandwidth between the GPU and its VRAM, as published. Effective bandwidth figures that include the cache do not count.

1,792GB/s10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 1792 GB/sec.

1,792GB/s1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Memory Bandwidth 1792 GB/sec.9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: Memory Bandwidth 1792 GB/sec.

Connectivity

0 of 12 differ

PCIe generationiHighest PCIe generation the card supports. In an older slot it runs at that slot's generation.

51Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Bus standard: PCI-E 5.0.9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026PCIe Gen 5.10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: PCI Express Gen 5.

51Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Bus Support PCI Express Gen 5.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026

PCIe lanesiElectrical lanes the card uses. Many 8-lane cards have a full-length x16 connector. In a PCIe 4.0 slot an 8-lane card has half the link bandwidth of a 16-lane card.

162Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Interface type: PCI Express x16 5.0.3Retailer pageparadigit.ie/gigabyte-aorus-geforce-rtx-5090-master-ice-32g/21757552/productRead 15 October 2026Interface type: PCI Express x16 5.0.

163Retailer pagecoolblue.nl/en/product/963046/inno3d-geforce-rtx-5090-x3-oc-32gb.htmlRead 16 October 2026Specifications: Bus type PCI Express 5.0 x16. INNO3D states only PCI Express Gen 5.6Spec databasegeizhals.de/inno3d-geforce-rtx-5090-x3-oc-n50903-32d7x-17593928-a3444085.htmlRead 16 October 2026Anbindung: PCIe 5.0 x16.

Display outputsiPhysical video connectors on the card, one row per type and version. USB-C is a USB Type-C port that carries video.

AGIGABYTE AORUS GeForce RTX 5090 MASTER ICE 32G

Type	Version	Count	Max bandwidth (Gbps)
HDMI	2.1b	1	Not available
DisplayPort	2.1b UHBR20	3	80

Source1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Display outputs: DisplayPort 2.1b *3 / HDMI 2.1b *1. Gigabyte's product title and the Gigabyte datasheet write DisplayPort 2.1a. The Gigabyte spec tab copy and NVIDIA write 2.1b, so 2.1b is stored, as on the catalog's other Gigabyte RTX 50 cards. The UHBR20 class and the 80 Gbps link rate come from NVIDIA's RTX 5090 page.2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026HDMI 1x version 2.1b and DisplayPort 3x. The sheet writes DisplayPort 2.1a, and the product title says 3 x DP 2.1a. Other Gigabyte copies and NVIDIA write 2.1b.9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026Footnote: DisplayPort 2.1b with UHBR20, which is 80 Gbps. HDMI 2.1b. Reference layout: 3x DisplayPort and 1x HDMI.

BINNO3D GeForce RTX 5090 X3 OC

Type	Version	Count	Max bandwidth (Gbps)
HDMI	2.1b	1	Not available
DisplayPort	2.1b UHBR20	3	80

Source1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Standard Display Connectors 3x DisplayPort 2.1b, 1x HDMI 2.1b. The UHBR20 class and the 80 Gbps link rate come from NVIDIA's RTX 5090 page footnote.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026Footnote 2: DisplayPort 2.1b with UHBR20, up to 4K 480 Hz or 8K 165 Hz with DSC. The 80 Gbps rate is UHBR20.

Display outputs in totaliPhysical video connectors on the card.

4=CalculatedFrom Display outputs

4=CalculatedFrom Display outputs

HDMI ports

1=CalculatedFrom Display outputs

1=CalculatedFrom Display outputs

DisplayPort ports

3=CalculatedFrom Display outputs

3=CalculatedFrom Display outputs

Highest HDMI version

2.1b=CalculatedFrom Display outputs

2.1b=CalculatedFrom Display outputs

Highest DisplayPort version

2.1b UHBR20=CalculatedFrom Display outputs

2.1b UHBR20=CalculatedFrom Display outputs

Fastest DisplayPort linkiHighest stated DisplayPort link rate on the card.

80Gbps=CalculatedFrom Display outputs

80Gbps=CalculatedFrom Display outputs

Max displaysiDisplays the card drives at the same time.

41Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Maximum displays: 4.9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026Multi-monitor: up to 4 displays.

41Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Multi Monitor up to 4.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026Multi Monitor: up to 4 displays.

Max output resolutioniHighest stated output resolution and its refresh rate at that resolution. Do not combine a resolution and a refresh rate from different modes. DSC and port conditions belong in the source note. When the card maker states only the resolution, take the refresh rate at that resolution from the chip maker. When two ports differ, store the highest mode and name the port in the source note.

Resolution7680 × 4320 pxRefresh rate165 Hz9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026Max resolution: 4K at 480 Hz or 8K at 165 Hz with DSC. Gigabyte states 7680 x 4320 and no refresh rate, so the rate at that resolution comes from NVIDIA. DSC is needed for this mode.

Resolution7680 × 4320 pxRefresh rate165 Hz1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Maximum Digital Resolution 4K at 480Hz or 8K at 165Hz with DSC. The 8K mode is stored. 7680 x 4320 is the 8K resolution.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026Max Digital Resolution: 4K at 480 Hz or 8K at 165 Hz with DSC. MSI's sheet gives 7680 x 4320. The 165 Hz is NVIDIA's.

HDCP version

2.39Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026HDCP 2.3.

2.31Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: HDCP 2.3.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026HDCP: 2.3.

Video engine

0 of 4 differ

Video encodersiHardware encode engines, e.g. 3 for "3x NVENC". Intel multi-format codec engines both encode and decode and count here and in video decoders. AMD publishes no engine counts. Leave it empty until an AMD document gives them.

39Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026NVENC: 3x ninth generation.10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 3 x NVENC (9th gen).11Maker product pagedeveloper.nvidia.com/video-encode-decode-support-matrixRead 15 October 20263 NVENC units.

31Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: NVIDIA Encoder 3x Ninth Generation.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026NVENC: 3x Ninth Generation.11Maker product pagedeveloper.nvidia.com/video-encode-decode-support-matrixRead 16 October 20263 NVENC units.

Video decodersiHardware decode engines, e.g. 2 for "2x NVDEC". AMD publishes no engine counts. Leave it empty until an AMD document gives them.

29Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026NVDEC: 2x sixth generation.10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: 2 x NVDEC (6th gen).11Maker product pagedeveloper.nvidia.com/video-encode-decode-support-matrixRead 15 October 20262 NVDEC units.

21Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: NVIDIA Decoder 2x Sixth Generation.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026NVDEC: 2x Sixth Generation.11Maker product pagedeveloper.nvidia.com/video-encode-decode-support-matrixRead 16 October 20262 NVDEC units.

Hardware video encodeiCodecs the maker states the GPU encodes in hardware.

H.264, H.265 (HEVC), AV19Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026AV1 encode: yes. H.264 and HEVC encode are in the video encode and decode support matrix.11Maker product pagedeveloper.nvidia.com/video-encode-decode-support-matrixRead 15 October 2026GeForce RTX 5090 (GB202): 3 NVENC units with H.264, HEVC and AV1 encode.

H.264, H.265 (HEVC), AV111Maker product pagedeveloper.nvidia.com/video-encode-decode-support-matrixRead 16 October 2026NVENC table, GeForce RTX 5090, 9th Gen: H.264, H.265 and AV1 encode are YES.

Hardware video decodeiCodecs the maker states the GPU decodes in hardware.

H.264, H.265 (HEVC), VP9, AV19Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026AV1 decode: yes. The other codecs are in the video encode and decode support matrix.11Maker product pagedeveloper.nvidia.com/video-encode-decode-support-matrixRead 15 October 2026GeForce RTX 5090 (GB202): 2 NVDEC units with H.264, HEVC, VP9 and AV1 decode. The page also lists MPEG-1, MPEG-2, MPEG-4, VC-1 and VP8, which have no option in the field.

H.264, H.265 (HEVC), VP9, AV111Maker product pagedeveloper.nvidia.com/video-encode-decode-support-matrixRead 16 October 2026NVDEC table, GeForce RTX 5090, 6th Gen: VP9, H.264, H.265 and AV1 decode are YES. H.266 is not listed.

Graphics features

0 of 5 differ

DirectXiAs published, e.g. "12 Ultimate".

12 Ultimate9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026APIs: DirectX 12 Ultimate.

12 Ultimate1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Microsoft DirectX 12 Ultimate.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026Microsoft DirectX 12 Ultimate.

VulkaniAs published. It can rise with drivers.

1.49Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026APIs: Vulkan 1.4.

1.41Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Vulkan API 1.4. OpenGL 4.6.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026Vulkan 1.4, OpenGL 4.6.

OpenGLiAs published.

4.69Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026APIs: OpenGL 4.6.

4.61Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: OpenGL 4.6.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026Vulkan 1.4, OpenGL 4.6.

Upscaling technologiesiThe maker's upscaling names with version, e.g. "DLSS 4", "FSR 4", "XeSS 2". Game support varies. A bundle name such as FSR Redstone is not an upscaler name, so its upscaler is written with its version, such as FSR 4. When the card maker and the chip maker write different versions, the chip maker's spec table wins. When the chip maker names no version, as NVIDIA writes DLSS for RTX 30 cards, the name is stored without one.

DLSS 412Press releasenvidianews.nvidia.com/news/nvidia-blackwell-geforce-rtx-50-series-opens-new-world-of-ai-computer-graphicsRead 15 October 2026DLSS 4 is supported on GeForce RTX 50 Series GPUs.

DLSS 41Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: NVIDIA DLSS 4.8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026The DLSS row lists DLSS 4 with Super Resolution, DLAA and Ray Reconstruction.

Frame generationiThe maker's frame generation names, e.g. "DLSS Multi Frame Generation", "FSR Frame Generation", "AMD Fluid Motion Frames 2". Driver-level features count. Game support varies. Chip-level lists are copied the same way into every card with that GPU. Versions are written with the major number only, so AMD Fluid Motion Frames 2.1 is AMD Fluid Motion Frames 2.

DLSS Frame Generation, DLSS Multi Frame Generation12Press releasenvidianews.nvidia.com/news/nvidia-blackwell-geforce-rtx-50-series-opens-new-world-of-ai-computer-graphicsRead 15 October 2026DLSS 4 debuts Multi Frame Generation, which generates up to three frames per rendered frame. DLSS Frame Generation carries over from the RTX 40 series and is listed as in the catalog's other RTX 50 cards.

DLSS Frame Generation, DLSS Multi Frame Generation8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026The DLSS row lists Frame Generation and Multi Frame Generation.

Power

1 of 4 differ

Board poweriDefault board power or power limit of this card. Use the card maker's figure first: NVIDIA Total Graphics Power, AMD Total or Typical Board Power, Intel TBP. A row on the card maker's own page counts under any label, such as Power Consumption or TDP. When the card maker states none, a reviewer's reading of the default power limit counts, and the source note names the reader and the tool. Retail feed fields such as "Power consumption (typical)" do not count. AMD's Typical Board Power and the power limit can differ. When both are known, store the default-mode power limit. The reference figure is not proof of a partner card's power. Other BIOS modes go in the source note. A reviewer who only relays a maker figure has not read the power limit.

600W14Reviewpcgameshardware.de/Geforce-RTX-5090-Grafikkarte-281029/Specials/Test-Release-kaufen-Takt-Temperaturen-Lautheit-1464748/2Read 15 October 2026PCGH text: the card ships with a 600 W power limit in both the Performance and Silent BIOS. The table row Max. Prozent Powerlimit reads 100 percent for both. PCGH read the limit with HWiNFO and MSI Afterburner (method on page 1: which power limit the maker set). Gigabyte states no board power in the sources read. The same article gives 575 W for the Founders Edition and the Zotac Solid.

Not available

Reference board poweriThe chip maker's board power for this GPU. It is a chip spec and repeats in every card with this GPU.

575W9Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 15 October 2026Total Graphics Power 575 W.10Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 15 October 2026Table 3: TGP 575 W.

575W8Maker product pagenvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090Read 16 October 2026Total Graphics Power 575 W.9Maker spec sheet (PDF)images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdfRead 16 October 2026Table 3: TGP 575 W.

PSU wattage (card maker)iThe PSU wattage the card maker lists for a full system with this card, whether it calls it Recommended, Minimum or Required. NVIDIA calls it Required System Power, AMD Minimum PSU Recommendation. When a maker lists both a recommendation and a lower minimum, store the recommendation and note the minimum. Makers differ for the same GPU, so compare cards of one maker only.

1,000W1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Required power supply: 1000W.2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Minimum system power supply: 1000 W.

1,000W1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Minimum System Power Requirement 1000 W. INNO3D calls it a minimum and gives no separate recommendation.

Power connectoriSupplementary power inputs on the card. 16-pin with no revision means the maker does not say 12V-2x6 or 12VHPWR. Use 12V-2x6 or 12VHPWR only when the card maker's page or manual names it. Do not infer it from the GPU generation or from a reviewer's or retailer's wording.

16-pin (revision not stated)1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Power connector: 16 pin*1. The row does not name 12V-2x6 or 12VHPWR, so the 16-pin option is stored.2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Supplementary power connectors: 1x 16-pin. The sheet does not name the revision.

16-pin (revision not stated)1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: Supplementary Power Connectors 4x PCIe 8-pin cables (adapter in box) OR 1x 600 W PCIe Gen 5 cable. INNO3D does not name the connector revision. Coolblue prints a 16-pin PCIe 5.0 connector.

Cooling

2 of 5 differ

Cooler typeiOpen air covers axial fans that blow onto the heatsink, including flow-through designs. All-in-one liquid is a card with an attached radiator. A water block needs a custom loop. Passive means no fan, and the fan count is 0.

Open air2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Cooling type: Active, WINDFORCE 3X with three fans.8Reviewnoobfeed.com/hardware/gigabyte-aorus-geforce-rtx-5090-master-ice-reviewRead 15 October 2026Triple-fan WindForce cooler on enlarged heatsinks and a vapor chamber.

Open air1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Features: enormous cooler with triple fan blades, a huge vapour chamber with full coverage for the GPU and memory, a die-casting metal base and a multifunctional backplate for hot air extraction.

FansiFans mounted on the card. Radiator fans of an all-in-one cooler are not counted.

32Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Number of fans: 3. Cooling technology: GIGABYTE WINDFORCE 3X.3Retailer pageparadigit.ie/gigabyte-aorus-geforce-rtx-5090-master-ice-32g/21757552/productRead 15 October 2026Number of fans: 3.8Reviewnoobfeed.com/hardware/gigabyte-aorus-geforce-rtx-5090-master-ice-reviewRead 15 October 2026Triple-fan array.

31Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Features: triple fan blades, 10 cm scythe fan blades. The central fan spins the opposite way (anticlockwise rotation).6Spec databasegeizhals.de/inno3d-geforce-rtx-5090-x3-oc-n50903-32d7x-17593928-a3444085.htmlRead 16 October 2026Kuehlung: 3x Axial-Luefter (100mm).

Radiator sizeiLiquid-cooled cards only. The nominal size class from the fan positions, e.g. 240 or 360.

Not available

Not available

Fan stop at low loadiThe fans stop below a temperature or load threshold, e.g. ASUS 0dB or MSI Zero Frozr. A maker's semi-passive or 0dB text counts. A threshold goes in the source note.

Yes14Reviewpcgameshardware.de/Geforce-RTX-5090-Grafikkarte-281029/Specials/Test-Release-kaufen-Takt-Temperaturen-Lautheit-1464748/2Read 15 October 2026Table row Lautstaerke Luefter Idle: 0 (semi-passiv) for the Performance and the Silent BIOS. The fans stop at idle.

Not available

Dual BIOSiA switch on the card selects between two factory firmware profiles, e.g. Performance and Quiet.

Yes2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Feature bullet: Dual BIOS (Performance / Silent).8Reviewnoobfeed.com/hardware/gigabyte-aorus-geforce-rtx-5090-master-ice-reviewRead 15 October 2026The dual-BIOS switch toggles between OC mode and Silent mode. Gigabyte's datasheet calls the modes Performance and Silent.14Reviewpcgameshardware.de/Geforce-RTX-5090-Grafikkarte-281029/Specials/Test-Release-kaufen-Takt-Temperaturen-Lautheit-1464748/2Read 15 October 2026The card has a Performance and a Silent BIOS, both tested.

Not available

Dimensions and weight

4 of 8 differ

LengthiCard length along the PCIe connector, from the bracket to the far end. Makers write it as L. Makers write L x W x H. W is the card height and H is the thickness. NVIDIA calls the height Width. Retail feeds relabel the sides as Height, Width or Depth, so decide by size: length is the longest side and thickness the shortest. When the maker lists sizes with and without the bracket, store the size with the bracket.

360mm1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Dimensions: L=360 W=150 H=75 mm. Gigabyte's L x W x H is length, height, thickness. Whether the bracket is included is not stated.

333mm1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Graphics Card Dimensions: Length 333, Height 137, Width 60, unit not printed and read as mm. The page does not say whether the bracket is included.6Spec databasegeizhals.de/inno3d-geforce-rtx-5090-x3-oc-n50903-32d7x-17593928-a3444085.htmlRead 16 October 2026Abmessungen: 333x137x60mm (LxBxH).

HeightiFrom the PCIe connector edge to the top of the card. The source note says when the bracket or the power plug is included. Makers write L x W x H. W is the card height and H is the thickness. NVIDIA calls the height Width. Retail feeds relabel the sides as Height, Width or Depth, so decide by size: length is the longest side and thickness the shortest. When the maker lists sizes with and without the bracket, store the size with the bracket.

150mm1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Second value of L=360 W=150 H=75 mm.

137mm1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Height 137.6Spec databasegeizhals.de/inno3d-geforce-rtx-5090-x3-oc-n50903-32d7x-17593928-a3444085.htmlRead 16 October 2026Same row, middle value.

ThicknessiFrom the bracket face to the back of the cooler. Makers often write it as H. Makers write L x W x H. W is the card height and H is the thickness. NVIDIA calls the height Width. Retail feeds relabel the sides as Height, Width or Depth, so decide by size: length is the longest side and thickness the shortest. When the maker lists sizes with and without the bracket, store the size with the bracket.

75mm1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Third value of L=360 W=150 H=75 mm.

60mm1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026INNO3D labels the 60 mm side Width. It is the thickness, since it is the shortest side.6Spec databasegeizhals.de/inno3d-geforce-rtx-5090-x3-oc-n50903-32d7x-17593928-a3444085.htmlRead 16 October 2026Same row, last value.

Slot widthiThickness in expansion-slot units as the card maker states it, e.g. 2.5 or 3.125. Not the number of slots the card covers in a case. Only a slot figure the card maker states. Do not compute it from the thickness. Retail feeds give different values for the same card. A slot figure in a row labelled Height, as on Palit, counts.

3.7slots2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Number of slots: 3.7. Paradigit and LDLC print the same figure. The sheet does not give the thickness measurement for this figure.3Retailer pageparadigit.ie/gigabyte-aorus-geforce-rtx-5090-master-ice-32g/21757552/productRead 15 October 2026Number of slots: 3.7.

Not available

Low profileiThe maker states the card fits low-profile slots, usually with a half-height bracket in the box. False when a data sheet says full-height bracket (FH/FL or FH/HL). Left out otherwise.

No2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Form factor: Full-Height/Full-Length (FH/FL), so not low profile.3Retailer pageparadigit.ie/gigabyte-aorus-geforce-rtx-5090-master-ice-32g/21757552/productRead 15 October 2026Form factor: Full-Height/Full-Length (FH/FL).

No6Spec databasegeizhals.de/inno3d-geforce-rtx-5090-x3-oc-n50903-32d7x-17593928-a3444085.htmlRead 16 October 2026Bauform: PCIe-Karte (full height). Slotblende: full height. A full-height bracket is a spec database's wording here, not an INNO3D datasheet line.

NVIDIA SFF-ReadyiThe card is on NVIDIA's SFF-Ready Enthusiast GeForce list. Only GeForce cards can qualify. For a GPU class the list covers, the list decides, and a card missing from it is false. Below the 70 class, which the list does not cover, the card maker's own SFF-Ready claim counts. A retail title does not.

No13Certification listnvidia.com/en-us/geforce/news/small-form-factor-sff-readyRead 15 October 2026The SFF-Ready Enthusiast list names three RTX 5090 cards (Founders Edition and two ASUS ProArt). The AORUS Master ICE is not on it, and the list limits are 304 mm length, 151 mm height and 2.5 slots. The card is 360 mm long and 3.7 slots.

No7Certification listnvidia.com/en-us/geforce/news/small-form-factor-sff-readyRead 16 October 2026The list covers the 5090 class. It names the Founders Edition and two ASUS ProArt RTX 5090 cards, and no INNO3D card. False is stored.

WeightiThe card alone, without support bracket, adapter or packaging. A liquid-cooled card includes its attached radiator and tubes. A weight rounded to whole kilograms, as some retail and distributor feeds print it, is not stored. A colour edition can state its own weight, so it is per variant. A feed weight that does not say whether it is the card or the package is not stored.

Not available

Not available

Package weightiOne retail box with its contents. A weight rounded to whole kilograms in a retail or distributor feed is not stored. A feed weight that does not say whether it is the card or the package is not stored. A feed's gross weight is the package weight, and the same rounding rule applies.

Not available

Not available

Lighting and software

2 of 3 differ

RGB lightingiDecorative lighting with selectable colours. A white logo or a single fixed colour does not count.

Yes1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Description: RGB Halo lighting, customizable and synchronized in GIGABYTE Control Center.2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Feature bullet: RGB Halo.8Reviewnoobfeed.com/hardware/gigabyte-aorus-geforce-rtx-5090-master-ice-reviewRead 15 October 2026RGB logo on the backplate, RGB Aorus emblem on the side and an RGB strip along the card.

Not available

Lighting synciSync standards the maker names, as a short name without a version, e.g. "ASUS Aura Sync", "MSI Mystic Light", "Gigabyte RGB Fusion". Leave it empty when the maker names only its own app. That goes in Software.

Not available

Not available

SoftwareiThe card maker's tuning software, not the chip maker's driver.

GIGABYTE Control Center1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Description: lighting is customized and synchronized through GIGABYTE Control Center.

Not available

Measured performance

1 of 6 differ

GPU clock while gaming (measured)iAverage GPU clock a reviewer logged in a game at the card's default settings, after warm-up. Game, resolution and BIOS mode are in the source note.

2,620MHz14Reviewpcgameshardware.de/Geforce-RTX-5090-Grafikkarte-281029/Specials/Test-Release-kaufen-Takt-Temperaturen-Lautheit-1464748/2Read 15 October 2026Table row Taktrate GPU: 2620 MHz in the Performance BIOS (2560 MHz in the Silent BIOS). Metro Exodus Enhanced Edition, church scene in the Volga section, at the power limit. Open test system with a passive CPU cooler and a semi-passive PSU, room about 21 C. The optional back fan was not fitted. CPU and driver are not given. Noobfeed logged an average of 2820 MHz in a one-hour F1 24 loop with the optional fan fitted, a different load.

Not available

3DMark Time Spy graphicsiGraphics score, not the overall score. Measured by a reviewer, not claimed by the maker. A score the review does not label as graphics is left out.

Not available

Not available

3DMark Steel NomadiMeasured by a reviewer, not claimed by the maker.

Not available

Not available

3DMark Port RoyaliMeasured by a reviewer, not claimed by the maker.

Not available

Not available

3DMark Speed WayiMeasured by a reviewer, not claimed by the maker.

Not available

Not available

Game frame ratesiReviewer frame rates for this card. The test CPU, memory and driver, frame generation, and whether the card ran at factory or reference clocks are in the source note.

Not available

Not available

Measured noise and temperature

2 of 5 differ

Fan noise per loadiSound pressure a reviewer measured from the card. Profile is the BIOS mode or fan profile as the reviewer names it. Fixed fan-speed tests are not a load. Compare only at the same distance. A row needs a distance. Take it from the review or from the site's test-method page. Leave the row out when neither states it. A synthetic or render load, such as a 3DMark loop or FurMark, is max load. Sone values are not converted. When a review gives two distances for one test, the row is left out.

Not available

Not available

Fan noise while gamingiLoudest gaming row, so the loudest BIOS mode tested. Compare only at the same distance.

Not available

Not available

GPU temperature while gamingiGPU core sensor after warm-up in a game, at default fan settings. Not the hot spot. A synthetic stress test does not count. The average or plateau reading after the card has warmed up, never the peak. A run of a few minutes does not count. The run length goes in the source note.

72.5°C14Reviewpcgameshardware.de/Geforce-RTX-5090-Grafikkarte-281029/Specials/Test-Release-kaufen-Takt-Temperaturen-Lautheit-1464748/2Read 15 October 2026Table row Temperatur GPU: 72.5 C in the Performance BIOS (80 C in the Silent BIOS), same Metro Exodus run at the power limit with the card in its normal setup. Read with HWiNFO and MSI Afterburner. This article gives no run length. PCGH's later method page (Asus ROG Astral RTX 5090 test, 28.09.2025) states at least 20 minutes of settling for the same scene. No hot spot value is given.

Not available

Hot spot temperature while gamingiHottest sensor on the GPU die under the same game load. The average or plateau reading after the card has warmed up, never the peak. A run of a few minutes does not count. The run length goes in the source note.

Not available

Not available

Memory temperature while gamingiVRAM sensor under the same game load. The average or plateau reading after the card has warmed up, never the peak. A run of a few minutes does not count. The run length goes in the source note.

74°C14Reviewpcgameshardware.de/Geforce-RTX-5090-Grafikkarte-281029/Specials/Test-Release-kaufen-Takt-Temperaturen-Lautheit-1464748/2Read 15 October 2026Table row Temperatur VRAM (junction): 74 C in the Performance BIOS (80 C in the Silent BIOS), same run.

Not available

Measured power

None of 2 known

Card power at idle (measured)iPower of the card alone at the desktop with one display. A hardware capture of slot and cable power comes first, a software reading of total board power second. The method is in the source note. Wall power does not count.

Not available

Not available

GPU power while gaming (measured)iAverage power of the card alone in a game at default settings. Same method rule as idle power. A mix of games, synthetic loads or stress tests does not count. Total board power, such as PCGH's Boardpower, not the GPU chip or Graphics Power reading alone.

Not available

Not available

In the box

2 of 3 differ

In the box

Quick Guide, Warranty Registration, Adapter, AORUS Stickers, Metal Sticker, VGA Holder5Retailer pagenetcodex.phRead 15 October 2026Accessories: Quick Guide, Warranty Registration, Adapter, AORUS Stickers, Metal Sticker, VGA Holder. The retailer copies Gigabyte's spec tab. The adapter type is not named. The store also lists the DisplayPort as 2.1a, a conflict kept out of the stored value.

Not available

Power adapter cable includediAn adapter from PCIe 8-pin cables to the card's 16-pin input is in the box.

Yes2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Packaging content: Cables included, Power. The sheet does not name the cable type. Netcodex lists Adapter in the box.5Retailer pagenetcodex.phRead 15 October 2026Accessories list names an Adapter.

Yes1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026Datasheet: 4x PCIe 8-pin cables (adapter in box).

Support bracket includediAn anti-sag holder or support bracket is in the box.

Yes2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Feature bullet: reinforced structure and versatile VGA holder. The VGA holder is the support bracket.5Retailer pagenetcodex.phRead 15 October 2026Accessories list names a VGA Holder.8Reviewnoobfeed.com/hardware/gigabyte-aorus-geforce-rtx-5090-master-ice-reviewRead 15 October 2026A reinforced metal support bracket is in the box.

Not available

Links

Not compared

Product page

www.gigabyte.com/Graphics-Card/GV-N5090AORUSM-ICE-32GD6Maker product pagegigabyte.com/Graphics-Card/GV-N5090AORUSM-ICE-32GDRead 15 October 2026Gigabyte product page link, taken from search results and the task's evidence link. gigabyte.com is on the blocked list, so the page was not opened. The specs above come from retailer and datasheet copies of Gigabyte's spec tab.

www.inno3d.com/product/inno3d-geforce-rtx-5090-x3-oc1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026The spec tab is on the product page.

Spec sheet

Not available

www.inno3d.com/product/inno3d-geforce-rtx-5090-x3-oc1Maker product pageinno3d.com/product/inno3d-geforce-rtx-5090-x3-ocRead 16 October 2026The datasheet is a spec table on the product page. No downloadable file is linked.

Manual

Not available

Not available

EPREL registration numberiId in the EU energy label database. Each size or model of a series has its own id, so it is per variant.

Not available

Not available

Warranty

0 of 1 differ

Manufacturer warranty (source market)iThe maker's warranty period for the exact SKU or configuration and market named in the source note. It applies only to that stated market, not every regional variant. Statutory consumer rights are excluded. The base term is stored. An extension that needs registration goes in the source note.

3years1Retailer pagehoanghapc.vn/vga-gigabyte-aorus-geforce-rtx-5090-master-ice-32gRead 15 October 2026Warranty: 36 months. The market is Vietnam. The Gigabyte datasheet lists 4 years with online registration, so only the 3 year base term is stored.2Spec databaseshop.aodatacloud.es/medias/datasheet_gv-n5090aorusm_ice-32gd.pdfRead 15 October 2026Warranty period 4 years. The feature bullet says Online registration required, so the base term is shorter. Not stored from this sheet.

3years2Maker spec sheet (PDF)inno3d.com/sites/default/files/2025-11/The%20INNO3D%20FAQ_0.pdfRead 16 October 2026The INNO3D FAQ (German support, 2 pages) says all graphics cards bought from June 2019 on have a three-year manufacturer's warranty. iCHILL cards always get three years.

Both are also in the API. Back to all graphics cards.