NVIDIA A10G vs NVIDIA GeForce RTX 5090 Comparison

The NVIDIA A10G and the NVIDIA GeForce RTX 5090 are two very different interpretations of what a high-end GPU should be, and the recorded data makes that split unusually clear. One is a single-slot server accelerator built for density and efficient compute deployment; the other is an active, current-generation consumer flagship with gaming and workstation ambitions. Benchmark results indicate the RTX 5090 dominates every shared compute test, yet the A10G retains a distinct profile in the database thanks to its class-leading percentile placement and its specialist server design. What follows is a walk through the measurements, category by category.

Head-to-Head Benchmarks

Only two tests exist in the database for both cards, and the RTX 5090 wins both by wide margins.

In Geekbench OpenCL, the RTX 5090 posts 334,370 against the A10G's 158,063. That is a gap of 52.7 percent, meaning the RTX 5090 delivers roughly twice the OpenCL throughput of the server card. In Geekbench Vulkan the margin widens further: 376,728 versus 145,863, a difference of 61.3 percent. The Vulkan result is particularly notable because Vulkan is an API both cards formally support at version 1.4, so this is not a case of one product lacking driver-level capability. The Blackwell-based card simply moves through the same workload far faster.

The surrounding data explains why. The RTX 5090's FP32 throughput is 104.8 TFLOPS against the A10G's 31.52 TFLOPS, its FP16 throughput is likewise 104.8 TFLOPS versus 31.52 TFLOPS, and its memory bandwidth of 1.79 TB/s is nearly triple the A10G's 600.2 GB/s. Texture fill rate shows the same shape: 1,636.8 GTexel/s for the RTX 5090 versus 492.5 GTexel/s for the A10G. With 21,760 shading units, 680 tensor cores, and 170 RT cores facing 9,216 shading units, 288 tensor cores, and 72 RT cores, the raw compute hierarchy favors the RTX 5090 at every level that the shared benchmarks exercise.

There is, however, an important caveat about how the database ranks each card against the full GPU population. The A10G sits at the 97th percentile against all GPUs, while the RTX 5090 sits at the 92nd percentile. This apparent inversion comes from each card's distinct rival set and benchmark pool, not from head-to-head strength. The A10G's average benchmark score of 151,963 places it 1.1 percent ahead of the Tesla V100 PCIe 32 GB, 5.4 percent behind the Radeon Pro W6800X, 6.5 percent behind the A100 PCIe 40 GB, and 9.3 percent ahead of the Instinct MI100. The RTX 5090's average score of 79,842 puts it 0.3 percent ahead of the Tesla P100 PCIe 16 GB, 0.6 percent ahead of the Tesla P100 PCIe 12 GB, 1.1 percent ahead of the Radeon RX 6850M XT, and 1.4 percent behind the Radeon Pro Vega 64X. The percentile figures therefore describe different competitive neighborhoods, and the head-to-head results remain the cleanest comparison available.

Where Each One Wins

The RTX 5090 wins every shared benchmark, and it also holds an exclusive benchmark footprint that the A10G does not appear in at all. The database records a 3DMark Steel Nomad DX12 score of 18,355, Passmark G3D at 39,650, Passmark GPU Compute at 26,756, Passmark G2D at 1,413, and DirectX 9, 10, 11, and 12 Passmark results of 395, 226, 341, and 185 respectively. This is a graphics card with a full display subsystem: one HDMI 2.1b output and three DisplayPort 2.1b outputs. For rendering, gaming, and desktop work, it is the only one of the two that can drive a monitor directly.

The A10G wins on deployment characteristics rather than raw scores. It is a single-slot card with a 150 W TDP, an 8-pin EPS power connector, a 450 W suggested PSU, and no display outputs at all. It was designed to be packed densely into server chassis where passive or shared cooling and per-slot power budgets matter more than peak throughput. Its 24 GB of GDDR6 on a 384-bit bus remains a substantial memory allocation for inference and virtualization workloads that fit within its compute envelope. Its rivals in the database, the V100, the A100, the W6800X, and the MI100, confirm its station as a datacenter part rather than a desktop contender.

The dimensional data reinforces the split. The A10G measures 267 mm long and 112 mm tall; the RTX 5090 measures 304 mm long, 137 mm tall, and 40 mm wide in a dual-slot form, with a 575 W TDP, a 16-pin power connector, and a 950 W suggested PSU. One card is built to disappear into a rack, the other to anchor a high-end workstation or gaming build.

FAQ

Q: Which card is faster in the shared benchmarks?

A: The RTX 5090, decisively. It leads Geekbench OpenCL 334,370 to 158,063, a 52.7 percent advantage, and Geekbench Vulkan 376,728 to 145,863, a 61.3 percent advantage.

Q: Does the A10G have any benchmark wins over the RTX 5090?

A: No. The database records zero wins for the A10G and two for the RTX 5090 in head-to-head tests.

Q: How do the two compare against the broader GPU database?

A: The A10G ranks in the 97th percentile versus all GPUs, and the RTX 5090 ranks in the 92nd percentile. These figures reflect different benchmark pools and rival sets, so they should not be read as a head-to-head result.

Q: Which card has more memory?

A: The RTX 5090, with 32 GB of GDDR7 on a 512-bit bus delivering 1.79 TB/s. The A10G has 24 GB of GDDR6 on a 384-bit bus at 600.2 GB/s.

Q: Are both cards still in production?

A: No. The A10G is listed as end-of-life, released in April 2021. The RTX 5090 is active, released in January 2025.

Q: Can the A10G drive a display?

A: No. It has no display outputs. The RTX 5090 has one HDMI 2.1b and three DisplayPort 2.1b outputs.

Specification Differences

The specification table diverges almost everywhere. The A10G uses the GA102 chip on the Ampere architecture, fabricated at Samsung on an 8 nm process, with 28,300 million transistors on a 628 mm² die at 45.1M per mm². The RTX 5090 uses the GB202 chip on Blackwell 2.0, fabricated at TSMC on a 5 nm process, with 92,200 million transistors on a 750 mm² die at 122.9M per mm². Transistor density roughly triples, and total transistor count more than triples.

Clocks differ sharply: the A10G runs a 1320 MHz base and 1710 MHz boost, with memory at 1563 MHz (12.5 Gbps effective), while the RTX 5090 runs 2017 MHz base and 2407 MHz boost with memory at 1750 MHz (28 Gbps effective). The functional unit counts follow the same pattern: 680 texture mapping units and 176 render output units on the RTX 5090 against 288 and 96 on the A10G. Pixel rate is 423.6 GPixel/s versus 164.2 GPixel/s.

The platform details diverge as well. The A10G connects over PCIe 4.0 x16, the RTX 5090 over PCIe 5.0 x16. Power delivery differs: 150 W through an 8-pin EPS connector against 575 W through a 16-pin connector, with suggested PSUs of 450 W and 950 W respectively. Both support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, so the API surface is identical. The RTX 5090 had a launch MSRP of 1,999 USD; no launch MSRP is recorded for the A10G.

Architecture Differences

The A10G belongs to the Server Ampere generation, the successor to Tesla Turing, and it was itself succeeded by Server Ada. The RTX 5090 belongs to the GeForce 50-series, the successor to the GeForce 40, with the GeForce 60 generation following it. These lineage positions tell the story: the A10G is a middle entry in a server line where the card is a replaceable compute module, while the RTX 5090 is a current flagship in a consumer line where the card is the centerpiece of the system.

The node transition from Samsung 8 nm to TSMC 5 nm underpins most of the performance gap, enabling both higher clocks and a far larger functional allocation. The memory subsystem change, from GDDR6 at 12.5 Gbps effective to GDDR7 at 28 Gbps effective on a wider bus, multiplies bandwidth to 1.79 TB/s. Both architectures expose 1:1 FP16 throughput relative to FP32, at 31.52 TFLOPS for the A10G and 104.8 TFLOPS for the RTX 5090.

The Verdict

For any workload the database measures directly, the data is unambiguous: the RTX 5090 is the faster card, ahead by 52.7 percent in OpenCL and 61.3 percent in Vulkan, with roughly three times the FP32 throughput and nearly three times the memory bandwidth, plus 32 GB of memory to the A10G's 24 GB. It is the right choice for gaming, rendering, GPU compute, and any workload that benefits from its exclusive Passmark and 3DMark results.

The A10G's case rests on deployment characteristics the RTX 5090 cannot match: single-slot width, a 150 W TDP, an 8-pin EPS connector, no display hardware, and a 97th-percentile standing against the full GPU database. In dense server configurations where per-slot power and physical volume are the binding constraints, those attributes matter more than raw benchmark scores. The A10G is also an end-of-life product, while the RTX 5090 remains active. Choose the RTX 5090 when performance per card is the goal; choose the A10G only when its server-grade form factor and power envelope fit a deployment the consumer flagship cannot.