NVIDIA A10G vs NVIDIA Tesla V100S PCIe 32 GB Comparison

NVIDIA
GEFORCE

NVIDIA A10G

CORE STATE GA102
VRAM 24 GB
CLOCK SPEED 1710 MHz
TDP 150 W
BUS WIDTH 384 bit
ARCHITECTURE Ampere
nm
PROCESS 8 nm
LAUNCH DATE 2021
VS
NVIDIA
GEFORCE

Tesla V100S PCIe 32 GB

CORE STATE GV100
VRAM 32 GB
CLOCK SPEED 1597 MHz
TDP 250 W
BUS WIDTH 4096 bit
ARCHITECTURE Volta
nm
PROCESS 12 nm
LAUNCH DATE 2019

PERFORMANCE BENCHMARKS

geekbench_opencl
158,063
194,415
geekbench_vulkan
145,863
N/A

Analysis: NVIDIA A10G vs NVIDIA Tesla V100S PCIe 32 GB

The NVIDIA Tesla V100S PCIe 32 GB and NVIDIA A10G are both end-of-life server accelerators, but they represent two very different design philosophies from NVIDIA. The V100S is a Volta-generation card with a massive HBM2 memory bus, while the A10G is an Ampere-generation card built on a newer process with GDDR6 memory. The benchmark data shows a clear single-test victory for the V100S, but the architectural differences tell a more nuanced story about which card suits which workload.

FAQ

Q: Which card has the higher Geekbench OpenCL score?

A: The NVIDIA Tesla V100S PCIe 32 GB scores 194415 in Geekbench OpenCL, while the NVIDIA A10G scores 158063. The V100S leads by 23% in this specific test.

Q: How does the A10G's average benchmark score compare to the V100S?

A: The A10G has an average benchmark score of 151963 across its two tests (Geekbench OpenCL and Vulkan), while the V100S has an average score of 194415 from its single OpenCL result. The V100S sits at the 98th percentile of all GPUs, while the A10G sits at the 97th percentile.

Q: What memory configurations do these cards use?

A: The V100S uses 32 GB of HBM2 on a 4096-bit bus with 1.13 TB/s bandwidth. The A10G uses 24 GB of GDDR6 on a 384-bit bus with 600.2 GB/s bandwidth. The V100S has roughly 88% more memory bandwidth.

Q: Which card has more shading units and tensor cores?

A: The A10G has 9216 shading units and 288 tensor cores, while the V100S has 5120 shading units and 640 tensor cores. The A10G has 80% more shading units, but the V100S has 122% more tensor cores.

Q: What are the power and physical differences?

A: The V100S is a dual-slot card with a 250 W TDP and 2x 8-pin power connectors, requiring a 600 W suggested PSU. The A10G is a single-slot card with a 150 W TDP and an 8-pin EPS connector, requiring a 450 W suggested PSU. The A10G measures 267 mm in length and 112 mm in height.

Q: Which card supports newer API features?

A: The A10G supports DirectX 12 Ultimate (12_2) and includes 72 RT cores for hardware ray tracing. The V100S supports DirectX 12 (12_1) and has no RT cores.

The Verdict

The data points to the NVIDIA Tesla V100S PCIe 32 GB for raw compute throughput in OpenCL workloads. Its 23% lead over the A10G in the single head-to-head benchmark, combined with its 98th percentile ranking versus the A10G's 97th, makes it the stronger choice for compute-heavy tasks that leverage its massive memory bandwidth and higher tensor core count.

The NVIDIA A10G is the pick when power efficiency and physical footprint matter more than peak OpenCL performance. Its 150 W TDP is 40% lower than the V100S's 250 W, and its single-slot design with a smaller 267 mm length makes it far easier to fit into dense server chassis. The A10G also brings modern Ampere features like RT cores and DirectX 12 Ultimate support, which the Volta-based V100S lacks entirely.

For buyers prioritizing raw compute in OpenCL, the V100S is the obvious choice. For those needing a lower-power, single-slot accelerator with newer API support and ray tracing capability, the A10G is the better fit. The V100S wins the benchmark comparison, but the A10G wins the practical deployment comparison.

Head-to-Head Benchmarks

The only direct benchmark comparison available is Geekbench OpenCL, and the results are decisive. The NVIDIA Tesla V100S PCIe 32 GB scores 194415, while the NVIDIA A10G scores 158063. This represents a 23% advantage for the V100S, which is a substantial margin in GPU computing.

This single-test result aligns with the nearest rival data for each card. The V100S is 3.9% ahead of the NVIDIA A100 SXM4 40 GB, which scores 187147, and 5.8% ahead of the A100 SXM4 80 GB at 183725. The A10G, by contrast, trails the AMD Radeon Pro W6800X by 5.4% (160671) and the A100 PCIe 40 GB by 6.5% (162504). The A10G does beat the AMD Instinct MI100 by 9.3%, but that card scores only 139035.

The V100S also edges out the RTX 6000D by a narrow margin, with the RTX 6000D scoring 195964 and the V100S trailing by just 0.8%. This places the V100S in elite company among the top-performing accelerators in the database. The A10G's best comparison is against the V100 PCIe 32 GB (non-S), where it leads by 1.1% with a score of 150305 for that older card.

Specification Differences

The two cards differ across nearly every major specification. The V100S uses a GV100 chip on a 12 nm TSMC process, while the A10G uses a GA102 chip on an 8 nm Samsung process. The A10G has more transistors (28,300 million versus 21,100 million) but a smaller die (628 mm² versus 815 mm²), giving it a much higher transistor density of 45.1M per mm² versus 25.9M per mm².

Clock speeds favor the A10G. The A10G has a base clock of 1320 MHz and a boost clock of 1710 MHz, compared to the V100S's 1245 MHz base and 1597 MHz boost. Memory clocks also differ significantly: the A10G runs at 1563 MHz (12.5 Gbps effective) versus the V100S's 1107 MHz (2.2 Gbps effective).

The V100S has a wider memory bus at 4096 bits versus 384 bits, and offers 32 GB versus 24 GB of memory. The A10G counters with more shading units (9216 versus 5120) and more TMUs (288 versus 320, actually the V100S has more TMUs), but the V100S has more ROPs (128 versus 96). The V100S has 640 tensor cores versus 288 on the A10G, but the A10G has 72 RT cores while the V100S has none.

Pixel and texture rates favor the V100S. The V100S delivers 204.4 GPixel/s and 511.0 GTexel/s, while the A10G delivers 164.2 GPixel/s and 492.5 GTexel/s. FP32 compute strongly favors the A10G at 31.52 TFLOPS versus 16.35 TFLOPS, while FP16 performance is nearly identical at 32.71 TFLOPS for the V100S and 31.52 TFLOPS for the A10G.

Architecture Differences

The architecture gap between these two cards is generational. The V100S is built on Volta, which debuted with a focus on tensor core compute for AI and HPC workloads. Its 640 tensor cores are nearly double the A10G's 288, and its HBM2 memory with a 4096-bit bus delivers 1.13 TB/s of bandwidth — a figure that remains impressive even by modern standards. Volta does not include ray tracing hardware, so the V100S is purely a compute accelerator.

The A10G is built on Ampere, which introduces several key improvements. The 8 nm Samsung process allows for 28,300 million transistors on a smaller die, yielding better power efficiency. Ampere brings 72 RT cores for ray tracing, which the V100S lacks entirely. The A10G also supports DirectX 12 Ultimate (12_2), while the V100S is limited to DirectX 12 (12_1).

Memory architecture is a major differentiator. The V100S uses HBM2 with a 4096-bit interface, which explains its massive bandwidth advantage. The A10G uses GDDR6 on a 384-bit bus, which is cheaper and easier to integrate but provides less than half the bandwidth. FP16 compute also differs in approach: the V100S achieves 32.71 TFLOPS via a 2:1 ratio (using tensor cores), while the A10G achieves 31.52 TFLOPS at a 1:1 ratio, meaning it can process FP16 at the same rate as FP32.

The bus interface also differs. The V100S uses PCIe 3.0 x16, while the A10G uses PCIe 4.0 x16. This gives the A10G double the host bandwidth for data transfer, which can matter in systems with multiple GPUs or high I/O workloads.

Where Each One Wins

The NVIDIA Tesla V100S PCIe 32 GB wins in scenarios that demand maximum memory bandwidth and tensor core throughput. Its 1.13 TB/s bandwidth is nearly double the A10G's 600.2 GB/s, making it superior for workloads that stream large datasets through memory, such as certain HPC simulations, large language model inference, and scientific computing tasks. The 640 tensor cores also give it an edge in dense matrix operations that leverage Volta's tensor core design. The 23% OpenCL benchmark lead confirms this advantage in real-world compute tests.

The NVIDIA A10G wins in power-constrained and space-constrained deployments. Its 150 W TDP is 100 W lower than the V100S, and its single-slot design with 267 mm length fits in chassis where the dual-slot V100S cannot. The A10G also wins on raw FP32 compute, delivering 31.52 TFLOPS versus 16.35 TFLOPS, which matters for workloads that don't use tensor cores and rely on traditional shader-based compute. The 72 RT cores make the A10G the only choice of the two for any ray tracing workload, and its DirectX 12 Ultimate support makes it more capable for graphics-adjacent tasks.

The A10G also wins on host data transfer with PCIe 4.0 x16 versus PCIe 3.0 x16, making it better for workloads that require frequent CPU-GPU communication. The V100S wins on overall benchmark percentile (98th versus 97th) and on memory capacity (32 GB versus 24 GB), which is critical for models that need to fit entirely in GPU memory.

In summary: the V100S is the compute monster with unmatched bandwidth and tensor throughput, while the A10G is the efficient, modern, ray-tracing-capable workhorse that sacrifices peak memory performance for flexibility and lower power draw. Choose based on whether your workload is memory-bound or compute-bound, and whether you need RT cores or can rely on pure tensor core performance.

DETAILED SPECIFICATIONS

SPECIFICATION
A10G
Tesla V100S PCIe 32 GB
Core Specs
Shading Units
9,216
5,120 -44.4%
Shaders
9,216
5,120 -44.4%
TMUs
288
320 +11.1%
ROPs
96
128 +33.3%
SM Count
72
80 +11.1%
Clocks
Base Clock
1320 MHz
1245 MHz
Boost Clock
1710 MHz
1597 MHz
Memory Clock
1563 MHz 12.5 Gbps effective
1107 MHz 2.2 Gbps effective
Memory
Memory Size
24 GB
32 GB
VRAM (MB)
24,576
32,768 +33.3%
Memory Type
GDDR6
HBM2
Memory Bus
384 bit
4096 bit
Bandwidth
600.2 GB/s
1.13 TB/s
Cache
L1 Cache
128 KB (per SM)
128 KB (per SM)
L2 Cache
6 MB
6 MB
Performance
Pixel Rate
164.2 GPixel/s
204.4 GPixel/s
Texture Rate
492.5 GTexel/s
511.0 GTexel/s
FP32 (TFLOPS)
31.52 TFLOPS
16.35 TFLOPS
FP64 (TFLOPS)
985.0 GFLOPS (1:32)
8.177 TFLOPS (1:2)
FP16 (TFLOPS)
31.52 TFLOPS (1:1)
32.71 TFLOPS (2:1)
AI/RT
RT Cores
72
—
Tensor Cores
288
640 +122.2%
Power
TDP
150 W
250 W
TDP (W)
150
250 +66.7%
Suggested PSU
450 W
600 W
Power Connectors
8-pin EPS
2x 8-pin
Architecture
Architecture
Ampere
Volta
GPU Name
GA102
GV100
Generation
Server Ampere (Axx)
Tesla Volta (Vxx)
Process Size
8 nm
12 nm
Transistors
28,300 million
21,100 million
Die Size
628 mm²
815 mm²
Foundry
Samsung
TSMC
Density
45.1M / mm²
25.9M / mm²
API Support
DirectX
12 Ultimate (12_2)
12 (12_1)
OpenGL
4.6
4.6
Vulkan
1.4
1.4
OpenCL
3.0
3.0
CUDA
8.6
7.0
Shader Model
6.8
6.8
Physical
Slot Width
Single-slot
Dual-slot
Length
267 mm 10.5 inches
—
Height
112 mm 4.4 inches
—
Outputs
No outputs
No outputs
Bus Interface
PCIe 4.0 x16
PCIe 3.0 x16
Other
Production
End-of-life
End-of-life
Predecessor
Tesla Turing
Tesla Pascal
Successor
Server Ada
Tesla Turing
View A10G Details View Tesla V100S PCIe 32 GB Details