AMD Instinct MI100 vs NVIDIA Tesla V100 PCIe 16 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 V100 PCIe 16 GB

CORE STATE GV100
VRAM 16 GB
CLOCK SPEED 1380 MHz
TDP 300 W
BUS WIDTH 4096 bit
ARCHITECTURE Volta
nm
PROCESS 12 nm
LAUNCH DATE 2017

PERFORMANCE BENCHMARKS

geekbench_opencl
139,035
163,063
geekbench_vulkan
N/A
113,062

Analysis: AMD Instinct MI100 vs NVIDIA Tesla V100 PCIe 16 GB

The AMD Instinct MI100 and NVIDIA Tesla V100 PCIe 16 GB are both end-of-life, dual-slot accelerators aimed at compute workloads, yet they present a distinct split in their strengths. The data shows a clear separation: the NVIDIA card wins the single available benchmark test decisively, while the AMD card counters with a substantially larger memory pool and higher raw throughput metrics. The choice between them hinges on whether a user prioritizes the measured compute performance in the Geekbench suite or the sheer capacity and bandwidth of on-board memory.

Where Each One Wins

The benchmark data, based on the Geekbench OpenCL test, gives a clear victory to the NVIDIA Tesla V100 PCIe 16 GB. It scores 163,063 points, which is 14.7% higher than the AMD Instinct MI100's score of 139,035. This is the only head-to-head benchmark available, and it shows a significant performance gap in favor of the NVIDIA card.

However, the AMD Instinct MI100 wins decisively in the field of memory and compute resources. It offers double the memory capacity with 32 GB of HBM2 compared to the Tesla V100's 16 GB. Its memory bandwidth is also substantially higher at 1.23 TB/s versus 897.0 GB/s. In terms of raw compute, the MI100's specifications are superior on paper, with a higher FP32 throughput of 23.07 TFLOPS and FP16 throughput of 46.14 TFLOPS, compared to the V100's 14.13 TFLOPS and 28.26 TFLOPS, respectively. The MI100 also has more shading units (7680 vs 5120) and texture mapping units (480 vs 320), though the V100 has double the ROPs (128 vs 64).

Architecture Differences

The two accelerators are built on fundamentally different architectures and manufacturing processes. The AMD Instinct MI100 uses the Arcturus chip, based on the CDNA 1.0 architecture, and is fabricated on a 7 nm process at TSMC. This allows for a transistor count of 25,600 million on a smaller die size of 750 mm², resulting in a higher transistor density of 34.1M per mm².

In contrast, the NVIDIA Tesla V100 PCIe 16 GB uses the GV100 chip, based on the Volta architecture, and is fabricated on a 12 nm process, also at TSMC. This older process results in 21,100 million transistors on a larger die size of 815 mm², yielding a lower transistor density of 25.9M per mm². The V100 also includes 640 tensor cores, a feature that is absent from the MI100's specifications.

The memory subsystems differ in capacity and bandwidth, as noted, but both use HBM2 with a 4096-bit bus width. The MI100's memory is clocked at 1200 MHz (2.4 Gbps effective), while the V100's memory is clocked at 876 MHz (1752 Mbps effective). The bus interfaces also differ, with the MI100 using PCIe 4.0 x16 and the V100 using PCIe 3.0 x16.

Head-to-Head Benchmarks

The sole head-to-head benchmark is the Geekbench OpenCL test, where the NVIDIA Tesla V100 PCIe 16 GB is the clear winner. The NVIDIA card scores 163,063, while the AMD Instinct MI100 scores 139,035. This represents a 14.7% delta in favor of the NVIDIA card.

This is a significant margin in a compute-oriented benchmark. The data indicates that, despite the MI100's higher theoretical specifications in memory bandwidth and FP32/FP16 throughput, the V100 achieves a higher actual score in this specific test. The V100's advantage might be attributed to its architecture, software optimization, or the specific workload characteristics of the Geekbench OpenCL test, but the result is unambiguous: the NVIDIA card outperforms the AMD card by a wide margin in this measurement.

The data provides no other head-to-head benchmarks, so this single test is the only quantitative comparison available. The result shows a clear win for the NVIDIA Tesla V100 PCIe 16 GB in this specific workload.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The AMD Instinct MI100 has a higher average benchmark score of 139,035, compared to the NVIDIA Tesla V100 PCIe 16 GB's average score of 138,063. The MI100's score is 0.7% higher than the V100's.

Q: How much memory does each accelerator have?

A: The AMD Instinct MI100 has 32 GB of HBM2 memory, while the NVIDIA Tesla V100 PCIe 16 GB has 16 GB of HBM2 memory. This means the MI100 has twice the memory capacity.

Q: What are the FP32 performance figures for each?

A: The AMD Instinct MI100 has an FP32 throughput of 23.07 TFLOPS, which is significantly higher than the NVIDIA Tesla V100 PCIe 16 GB's FP32 throughput of 14.13 TFLOPS.

Q: Does the NVIDIA Tesla V100 have tensor cores?

A: Yes, the NVIDIA Tesla V100 PCIe 16 GB includes 640 tensor cores. The AMD Instinct MI100's specifications do not list tensor cores.

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

A: The AMD Instinct MI100 has a memory bandwidth of 1.23 TB/s, which is higher than the NVIDIA Tesla V100 PCIe 16 GB's bandwidth of 897.0 GB/s.

Q: In the Geekbench OpenCL test, which GPU wins and by how much?

A: The NVIDIA Tesla V100 PCIe 16 GB wins the Geekbench OpenCL test with a score of 163,063, which is 14.7% higher than the AMD Instinct MI100's score of 139,035.

The Verdict

Based strictly on the data, the choice between these two accelerators depends on the workload. For a workload represented by the Geekbench OpenCL benchmark, the NVIDIA Tesla V100 PCIe 16 GB is the clear winner, delivering a 14.7% higher score. This makes it the preferable option for tasks that align with that specific benchmark's characteristics.

The AMD Instinct MI100, however, offers a compelling alternative for workloads that demand large memory capacity and high bandwidth. Its 32 GB of HBM2 memory, double that of the V100, and its 1.23 TB/s bandwidth are significant advantages for data sets that exceed 16 GB or for memory-bound operations. Its higher FP32 and FP16 throughput figures also suggest it could be faster in certain compute tasks, despite its loss in the Geekbench test.

Users who rely on the specific performance measured by Geekbench OpenCL should select the NVIDIA Tesla V100 PCIe 16 GB. Those who need to process larger datasets or require the higher memory bandwidth and theoretical compute throughput should consider the AMD Instinct MI100. The data supports either choice, depending on which metric is more relevant to the user's application.

Specification Differences

| Specification | AMD Instinct MI100 | NVIDIA Tesla V100 PCIe 16 GB |

|:--- |:--- |:--- |

| Chip | Arcturus | GV100 |

| Architecture | CDNA 1.0 | Volta |

| Process Node | 7 nm | 12 nm |

| Transistors | 25,600 million | 21,100 million |

| Die Size | 750 mm² | 815 mm² |

| Transistor Density | 34.1M / mm² | 25.9M / mm² |

| Base Clock | 1000 MHz | 1245 MHz |

| Boost Clock | 1502 MHz | 1380 MHz |

| Memory Size | 32 GB | 16 GB |

| Memory Type | HBM2 | HBM2 |

| Memory Clock | 1200 MHz (2.4 Gbps effective) | 876 MHz (1752 Mbps effective) |

| Memory Bandwidth | 1.23 TB/s | 897.0 GB/s |

| Shading Units | 7680 | 5120 |

| TMUs | 480 | 320 |

| ROPs | 64 | 128 |

| Tensor Cores | N/A | 640 |

| Pixel Rate | 96.13 GPixel/s | 176.6 GPixel/s |

| Texture Rate | 721.0 GTexel/s | 441.6 GTexel/s |

| FP32 Performance | 23.07 TFLOPS | 14.13 TFLOPS |

| FP16 Performance | 46.14 TFLOPS (2:1) | 28.26 TFLOPS (2:1) |

| Bus Interface | PCIe 4.0 x16 | PCIe 3.0 x16 |

| APIs (DirectX) | N/A | 12 (12_1) |

| APIs (OpenGL) | N/A | 4.6 |

| APIs (Vulkan) | N/A | 1.4 |

| Release Date | 2020-11-15 | 2017-06-20 |

| Predecessor | Radeon Instinct | Tesla Pascal |

| Successor | N/A | Tesla Turing |

DETAILED SPECIFICATIONS

SPECIFICATION
Instinct MI100
Tesla V100 PCIe 16 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
1380 MHz
Memory Clock
1200 MHz 2.4 Gbps effective
876 MHz 1752 Mbps effective
Memory
Memory Size
32 GB
16 GB
VRAM (MB)
32,768
16,384 -50.0%
Memory Type
HBM2
HBM2
Memory Bus
4096 bit
4096 bit
Bandwidth
1.23 TB/s
897.0 GB/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
176.6 GPixel/s
Texture Rate
721.0 GTexel/s
441.6 GTexel/s
FP32 (TFLOPS)
23.07 TFLOPS
14.13 TFLOPS
FP64 (TFLOPS)
11.54 TFLOPS (1:2)
7.066 TFLOPS (1:2)
FP16 (TFLOPS)
46.14 TFLOPS (2:1)
28.26 TFLOPS (2:1)
AI/RT
Tensor Cores
—
640
Power
TDP
300 W
300 W
TDP (W)
300
300 0.0%
Suggested PSU
700 W
700 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 V100 PCIe 16 GB Details