AMD Instinct MI100 vs NVIDIA Tesla V100S PCIe 32 GB Comparison

AMD
RADEON

AMD Instinct MI100

CORE STATE Arcturus
VRAM 32 GB
CLOCK SPEED 1502 MHz
TDP 300 W
BUS WIDTH 4096 bit
ARCHITECTURE CDNA 1.0
nm
PROCESS 7 nm
LAUNCH DATE 2020
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
139,035
194,415

Analysis: AMD Instinct MI100 vs NVIDIA Tesla V100S PCIe 32 GB

The NVIDIA Tesla V100S PCIe 32 GB and the AMD Instinct MI100 are both end-of-life enterprise accelerators aimed at high-performance computing, but the benchmark data shows a decisive performance gap. In the only available Geekbench OpenCL test, the V100S scores 194,415, while the MI100 scores 139,035. This gives the NVIDIA card a 39.8% lead over its AMD competitor. The V100S also sits at the 98th percentile of all GPUs, whereas the MI100 ranks at the 96th percentile, confirming that the NVIDIA part occupies a higher tier of overall compute capability.

Head-to-Head Benchmarks

The single benchmark result in the data paints a stark picture. The NVIDIA Tesla V100S PCIe 32 GB delivers an OpenCL score of 194,415, which is significantly higher than the 139,035 achieved by the AMD Instinct MI100. This represents a 39.8% advantage for the NVIDIA accelerator, a substantial margin that indicates a clear performance hierarchy between these two products. The V100S’s lead is not a marginal victory; it is a dominant one, placing it nearly 40% ahead of the MI100 in raw compute throughput as measured by this workload.

Looking at the nearest rivals for each card provides additional context. The V100S is 3.9% ahead of the NVIDIA A100 SXM4 40 GB and 5.8% ahead of the A100 SXM4 80 GB, placing it firmly within the top tier of NVIDIA’s own data center lineup. Its closest competitor is the RTX 6000D, which is only 0.8% slower. On the AMD side, the MI100's closest rival is the NVIDIA Tesla V100 PCIe 16 GB, which trails by just 0.7%. The MI100 also edges out the NVIDIA Tesla V100 SXM2 32 GB by 0.9% and the AMD Radeon PRO V620 by 1.9%. This shows that while the MI100 is competitive with the older, non-S V100 variants, it is in a completely different performance class than the V100S.

The delta between the two cards is consistent with their respective competitive positions. The V100S outperforms its nearest rivals, while the MI100 only narrowly beats its closest competitors. When compared directly, the V100S’s 39.8% lead over the MI100 is a decisive indicator of which accelerator delivers more compute performance in this benchmark. The data shows no benchmark where the MI100 comes out ahead, as the record lists 1 win for the NVIDIA card and 0 for the AMD card.

Architecture Differences

The architectural divide between these two accelerators is fundamental, beginning with the manufacturing process and core design. The V100S is built on NVIDIA's Volta architecture and uses the GV100 chip, fabricated on a 12 nm process at TSMC. In contrast, the MI100 is based on AMD's CDNA 1.0 architecture, featuring the Arcturus chip, and is manufactured on a more advanced 7 nm process, also at TSMC. This process difference is reflected in transistor density: the MI100 packs 25,600 million transistors into a 750 mm² die, yielding a density of 34.1M transistors per mm², while the V100S houses 21,100 million transistors on a larger 815 mm² die, for a density of 25.9M per mm².

The compute configurations diverge sharply. The MI100 has 7,680 shading units, 480 texture mapping units (TMUs), and 64 render output units (ROPs). The V100S, conversely, has 5,120 shading units, 320 TMUs, and 128 ROPs. While the MI100 has 50% more shading units and 50% more TMUs, the V100S has double the ROP count. This results in different throughput characteristics: the MI100's pixel rate is 96.13 GPixel/s, while the V100S achieves 204.4 GPixel/s. However, in texture rate, the MI100 leads with 721.0 GTexel/s versus the V100S's 511.0 GTexel/s.

A critical difference lies in the absence of dedicated tensor cores on the AMD part. The V100S includes 640 tensor cores, a feature entirely missing from the MI100. This is a key architectural distinction, as tensor cores are designed to accelerate AI and deep learning workloads. The MI100 also lacks any ray tracing cores, but neither card has those. In terms of floating-point performance, the MI100 delivers 23.07 TFLOPS for FP32 and 46.14 TFLOPS for FP16 (2:1), whereas the V100S offers 16.35 TFLOPS for FP32 and 32.71 TFLOPS for FP16 (2:1). The AMD card is clearly superior in raw FP32 and FP16 throughput.

The memory subsystems are similar in capacity but differ in bandwidth. Both cards feature 32 GB of HBM2 memory with a 4096-bit bus. However, the MI100’s memory runs at 1200 MHz (2.4 Gbps effective), providing a bandwidth of 1.23 TB/s, while the V100S’s memory runs at 1107 MHz (2.2 Gbps effective), yielding 1.13 TB/s. The MI100 also supports PCIe 4.0 x16, while the V100S is limited to PCIe 3.0 x16. Both cards are dual-slot designs with no display outputs, requiring 2x 8-pin power connectors. The MI100 has a higher thermal design power (TDP) of 300 W compared to the V100S’s 250 W, and it recommends a 700 W power supply versus the 600 W suggested for the NVIDIA card.

FAQ

Q: Which GPU is faster in the Geekbench OpenCL benchmark?

A: The NVIDIA Tesla V100S PCIe 32 GB is faster, scoring 194,415 compared to the AMD Instinct MI100's 139,035, a 39.8% difference.

Q: Does the AMD Instinct MI100 have more shading units than the NVIDIA card?

A: Yes, the MI100 has 7,680 shading units, which is 50% more than the V100S’s 5,120 shading units.

Q: What is the memory bandwidth difference between the two cards?

A: The AMD MI100 has a higher memory bandwidth of 1.23 TB/s, while the NVIDIA V100S has 1.13 TB/s. Both use 32 GB of HBM2 memory on a 4096-bit bus.

Q: Are there any tensor cores on the AMD Instinct MI100?

A: No, the MI100 does not have any tensor cores. The NVIDIA V100S includes 640 tensor cores, which are absent from the AMD design.

Q: How does the transistor density compare between the two chips?

A: The AMD MI100's Arcturus chip has a transistor density of 34.1M per mm², which is higher than the NVIDIA V100S’s GV100 chip at 25.9M per mm².

Q: Which card has a higher FP32 performance rating?

A: The AMD Instinct MI100 has a higher FP32 rating at 23.07 TFLOPS, compared to the NVIDIA V100S’s 16.35 TFLOPS.

Specification Differences

  • Process Node: The NVIDIA V100S uses a 12 nm process, while the AMD MI100 uses a 7 nm process.
  • Transistors: The V100S has 21,100 million transistors, while the MI100 has 25,600 million.
  • Die Size: The V100S has an 815 mm² die, while the MI100 has a slightly smaller 750 mm² die.
  • Transistor Density: The MI100 is denser at 34.1M / mm², compared to the V100S’s 25.9M / mm².
  • Base Clock: The V100S runs at 1245 MHz, while the MI100 runs at 1000 MHz.
  • Boost Clock: The V100S boosts to 1597 MHz, while the MI100 boosts to 1502 MHz.
  • Memory Clock: The V100S memory runs at 1107 MHz (2.2 Gbps effective), while the MI100 runs at 1200 MHz (2.4 Gbps effective).
  • Memory Bandwidth: The V100S offers 1.13 TB/s, while the MI100 offers 1.23 TB/s.
  • Shading Units: The V100S has 5,120, while the MI100 has 7,680.
  • TMUs: The V100S has 320, while the MI100 has 480.
  • ROPs: The V100S has 128, while the MI100 has 64.
  • Tensor Cores: The V100S has 640, while the MI100 has none.
  • Pixel Rate: The V100S achieves 204.4 GPixel/s, while the MI100 achieves 96.13 GPixel/s.
  • Texture Rate: The V100S achieves 511.0 GTexel/s, while the MI100 achieves 721.0 GTexel/s.
  • FP32 Performance: The V100S is rated at 16.35 TFLOPS, while the MI100 is rated at 23.07 TFLOPS.
  • FP16 Performance: The V100S is rated at 32.71 TFLOPS (2:1), while the MI100 is rated at 46.14 TFLOPS (2:1).
  • TDP: The V100S has a 250 W TDP, while the MI100 has a 300 W TDP.
  • Suggested PSU: The V100S recommends a 600 W PSU, while the MI100 recommends a 700 W PSU.
  • Bus Interface: The V100S uses PCIe 3.0 x16, while the MI100 uses PCIe 4.0 x16.
  • API Support: The V100S supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4, while the MI100 lists N/A for all three APIs.
  • Dimensions: The MI100 has a listed length of 267 mm and height of 111 mm, while the V100S has no listed dimensions.

Where Each One Wins

The NVIDIA Tesla V100S PCIe 32 GB is the clear winner in overall compute performance as measured by the Geekbench OpenCL benchmark. Its 39.8% lead over the MI100 makes it the superior choice for workloads that rely on raw, single-precision compute throughput in general-purpose GPU computing. The V100S also benefits from its 640 tensor cores, a feature absent on the AMD card, making it the only one of the two suited for accelerated AI and deep learning inference or training tasks that leverage tensor cores. Furthermore, its higher pixel rate of 204.4 GPixel/s suggests an advantage in workloads that are heavily dependent on rasterization-like operations or pixel processing, despite both cards lacking display outputs. The V100S also does this with a lower TDP of 250 W compared to the MI100’s 300 W, indicating better energy efficiency for the performance delivered.

The AMD Instinct MI100, despite losing the overall benchmark, has specific hardware advantages that make it the better option for certain compute-heavy tasks. Its architecture provides significantly higher theoretical peak performance in FP32 and FP16, with 23.07 TFLOPS and 46.14 TFLOPS respectively, compared to the V100S’s 16.35 and 32.71 TFLOPS. This suggests that in highly optimized code that can saturate the GPU’s arithmetic units, the MI100 could potentially outperform the V100S, even though the OpenCL benchmark shows otherwise. The MI100 also has a higher texture rate of 721.0 GTexel/s, which could benefit texture-heavy scientific visualization workloads. Additionally, its faster memory bandwidth of 1.23 TB/s and support for PCIe 4.0 x16 provide a more modern I/O interface, which could reduce data transfer bottlenecks in systems that support the newer standard. For pure compute density and memory bandwidth, the MI100 holds the theoretical edge, but the V100S wins where it counts in the available benchmark.

DETAILED SPECIFICATIONS

SPECIFICATION
Instinct MI100
Tesla V100S PCIe 32 GB
Core Specs
Shading Units
7,680
5,120 -33.3%
Shaders
7,680
5,120 -33.3%
TMUs
480
320 -33.3%
ROPs
64
128 +100.0%
Compute Units
120
—
SM Count
—
80
Clocks
Base Clock
1000 MHz
1245 MHz
Boost Clock
1502 MHz
1597 MHz
Memory Clock
1200 MHz 2.4 Gbps effective
1107 MHz 2.2 Gbps effective
Memory
Memory Size
32 GB
32 GB
VRAM (MB)
32,768
32,768 0.0%
Memory Type
HBM2
HBM2
Memory Bus
4096 bit
4096 bit
Bandwidth
1.23 TB/s
1.13 TB/s
Cache
L1 Cache
16 KB (per CU)
128 KB (per SM)
L2 Cache
8 MB
6 MB
Performance
Pixel Rate
96.13 GPixel/s
204.4 GPixel/s
Texture Rate
721.0 GTexel/s
511.0 GTexel/s
FP32 (TFLOPS)
23.07 TFLOPS
16.35 TFLOPS
FP64 (TFLOPS)
11.54 TFLOPS (1:2)
8.177 TFLOPS (1:2)
FP16 (TFLOPS)
46.14 TFLOPS (2:1)
32.71 TFLOPS (2:1)
AI/RT
Tensor Cores
—
640
Power
TDP
300 W
250 W
TDP (W)
300
250 -16.7%
Suggested PSU
700 W
600 W
Power Connectors
2x 8-pin
2x 8-pin
Architecture
Architecture
CDNA 1.0
Volta
GPU Name
Arcturus
GV100
Generation
Instinct (MIx)
Tesla Volta (Vxx)
Process Size
7 nm
12 nm
Transistors
25,600 million
21,100 million
Die Size
750 mm²
815 mm²
Foundry
TSMC
TSMC
Density
34.1M / mm²
25.9M / mm²
API Support
DirectX
—
12 (12_1)
OpenGL
—
4.6
Vulkan
—
1.4
OpenCL
2.1
3.0
CUDA
—
7.0
Shader Model
—
6.8
Physical
Slot Width
Dual-slot
Dual-slot
Length
267 mm 10.5 inches
—
Height
111 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
Radeon Instinct
Tesla Pascal
Successor
—
Tesla Turing
View Instinct MI100 Details View Tesla V100S PCIe 32 GB Details