AMD Instinct MI100 vs NVIDIA Tesla P100 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 P100 PCIe 16 GB

CORE STATE GP100
VRAM 16 GB
CLOCK SPEED 1329 MHz
TDP 250 W
BUS WIDTH 4096 bit
ARCHITECTURE Pascal
nm
PROCESS 16 nm
LAUNCH DATE 2016

PERFORMANCE BENCHMARKS

geekbench_opencl
139,035
79,605

Analysis: AMD Instinct MI100 vs NVIDIA Tesla P100 PCIe 16 GB

AMD Instinct MI100 vs NVIDIA Tesla P100 PCIe 16 GB

The recorded database shows a single head-to-head benchmark between these two accelerators, and the result is decisive. The AMD Instinct MI100 scores 139035 in Geekbench OpenCL, while the NVIDIA Tesla P100 PCIe 16 GB scores 79605. That is a 74.7% higher score for the MI100, placing it in the 96th percentile of all GPUs, compared to the P100’s 92nd percentile. The gap is substantial, but the P100’s nearest rivals include the GeForce RTX 5090 at 79842 and the Radeon Pro Vega 64X at 80959, which means the P100 sits in a competitive neighborhood even if it trails the MI100 by a wide margin.

Head-to-Head Benchmarks

The only benchmark in the database is Geekbench OpenCL, and the MI100 wins it outright. The MI100’s 139035 score is 74.7% higher than the P100’s 79605. To put that in context, the MI100’s nearest rival is the NVIDIA Tesla V100 PCIe 16 GB at 138063, which is only 0.7% behind. The Tesla V100 SXM2 32 GB is at 137731 (0.9% behind), and the AMD Radeon PRO V620 is at 136472 (1.9% behind). So the MI100 leads its immediate competitor group by a small margin, but the P100 is far behind, in a different tier. The P100’s closest rivals are the Tesla P100 PCIe 12 GB at 79396 (0.3% behind), the GeForce RTX 5090 at 79842 (0.3% ahead, since the delta is negative), and the AMD Radeon RX 6850M XT at 78940 (0.8% behind). The P100 is also 1.7% behind the Radeon Pro Vega 64X at 80959. This means the P100 is effectively mid-pack among its own peers, while the MI100 is at the top of a higher-performance cluster.

The delta of 74.7% is not a marginal edge; it is a generational jump. The MI100 achieves this with a 7 nm process, while the P100 uses 16 nm. The MI100 has 25,600 million transistors on a 750 mm² die, compared to the P100’s 15,300 million on 610 mm². Transistor density is 34.1M per mm² for the MI100 versus 25.1M for the P100. The MI100’s FP32 throughput is 23.07 TFLOPS, while the P100 is at 9.526 TFLOPS. The MI100’s FP16 is 46.14 TFLOPS (2:1) versus the P100’s 19.05 TFLOPS (2:1). The MI100 also has 7680 shading units, 480 TMUs, and 64 ROPs, while the P100 has 3584 shading units, 224 TMUs, and 96 ROPs. The MI100’s texture rate is 721.0 GTexel/s versus 297.7 GTexel/s, but the P100’s pixel rate is higher at 127.6 GPixel/s versus 96.13 GPixel/s. The MI100’s memory bandwidth is 1.23 TB/s, far above the P100’s 732.2 GB/s, and it carries 32 GB of HBM2 versus 16 GB. These specifications explain the benchmark result, but the benchmark itself is the only direct measurement.

The Verdict

From the data, the AMD Instinct MI100 is the clear performance winner in the recorded OpenCL test. It delivers a 74.7% higher score than the Tesla P100 PCIe 16 GB and sits in the 96th percentile of all GPUs, while the P100 is in the 92nd percentile. If the decision is based solely on raw compute throughput as measured by Geekbench OpenCL, the MI100 is the stronger choice. The MI100 also offers double the memory capacity (32 GB versus 16 GB) and more than double the memory bandwidth (1.23 TB/s versus 732.2 GB/s). Its FP32 and FP16 rates are approximately 2.4 times higher. The P100, however, has a higher pixel rate (127.6 GPixel/s versus 96.13 GPixel/s) and a higher base clock (1190 MHz versus 1000 MHz), but the boost clock favors the MI100 (1502 MHz versus 1329 MHz). The P100 also has a lower TDP (250 W versus 300 W) and requires a single 8-pin connector versus the MI100’s two 8-pin connectors. The P100’s suggested PSU is 600 W versus 700 W for the MI100. For users prioritizing power efficiency, the P100 is less demanding, but for raw compute performance, the MI100 dominates.

The P100’s only advantage in the recorded data is its pixel fill rate and lower power draw. It also has a higher ROP count (96 versus 64) and a higher base clock. The P100 supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3, while the MI100 lists N/A for all APIs. This suggests the P100 has broader software compatibility for graphics-oriented workloads, despite both cards having no display outputs. The P100’s launch MSRP was 5,699 USD, but no such figure exists for the MI100, so no price comparison is possible from the data. The verdict from the benchmark is unambiguous: the MI100 is the faster card in OpenCL, but the P100 offers lower power consumption and more advanced graphics API support.

Where Each One Wins

The MI100 wins in every compute-heavy metric recorded in the database. It has higher FP32 (23.07 TFLOPS versus 9.526 TFLOPS), higher FP16 (46.14 versus 19.05), higher texture rate (721.0 GTexel/s versus 297.7 GTexel/s), higher memory bandwidth (1.23 TB/s versus 732.2 GB/s), and more memory (32 GB versus 16 GB). It also has more shading units (7680 versus 3584) and more TMUs (480 versus 224). The MI100’s OpenCL score is 139035, which is 74.7% higher than the P100’s 79605. The MI100 also sits in the 96th percentile versus the P100’s 92nd, and its nearest rivals are all above 135000, while the P100’s nearest rivals are all below 81000. This indicates the MI100 is designed for high-throughput parallel workloads, such as dense matrix operations or scientific simulation, where FP32 and FP16 rates matter.

The P100 wins in a few specific areas. Its pixel rate is 127.6 GPixel/s, which is higher than the MI100’s 96.13 GPixel/s. Its ROP count is 96 versus 64, and its base clock is 1190 MHz versus 1000 MHz. The P100 also has a lower TDP (250 W versus 300 W), a single 8-pin connector instead of two, and a 600 W suggested PSU versus 700 W. The P100 supports DirectX 12, OpenGL 4.6, and Vulkan 1.3, while the MI100 has no API support listed. This makes the P100 a better fit for rasterization-heavy tasks or graphics API compatibility, even if it has no display outputs. The P100’s memory clock is 715 MHz (1430 Mbps effective), while the MI100’s is 1200 MHz (2.4 Gbps effective), which contributes to the bandwidth difference. The P100 is also built on a 16 nm process, which is older than the MI100’s 7 nm, but it achieves a higher pixel rate nonetheless.

For use cases, the MI100 is the choice for FP32 or FP16 compute, large memory footprints, and high-bandwidth data movement. The P100 is the choice for scenarios where pixel fill rate matters, where power consumption is a constraint, or where software requires graphics API support. The P100’s PCIe 3.0 x16 interface is older than the MI100’s PCIe 4.0 x16, so the MI100 also has a faster host connection. The MI100’s transistor count is 25,600 million versus 15,300 million, and its die is larger at 750 mm² versus 610 mm², but the P100’s lower density (25.1M versus 34.1M per mm²) suggests a less packed design. The data shows the MI100 is the superior compute accelerator, while the P100 retains niche advantages in pixel throughput and power efficiency.

FAQ

Q: Which card has the higher OpenCL benchmark score?

A: The AMD Instinct MI100 scores 139035, which is 74.7% higher than the NVIDIA Tesla P100 PCIe 16 GB’s 79605.

Q: What are the memory capacities and bandwidths of each card?

A: The MI100 has 32 GB of HBM2 with 1.23 TB/s bandwidth. The P100 has 16 GB of HBM2 with 732.2 GB/s bandwidth.

Q: Which card has a higher FP32 performance?

A: The MI100 has 23.07 TFLOPS FP32, while the P100 has 9.526 TFLOPS FP32. The MI100 is about 2.4 times higher.

Q: Does the P100 have any advantage over the MI100?

A: Yes, the P100 has a higher pixel rate (127.6 GPixel/s versus 96.13 GPixel/s), more ROPs (96 versus 64), a lower TDP (250 W versus 300 W), and supports DirectX 12, OpenGL 4.6, and Vulkan 1.3, while the MI100 lists no API support.

Q: How does each card compare to its nearest rivals?

A: The MI100 leads the Tesla V100 PCIe 16 GB by 0.7% and the Tesla V100 SXM2 32 GB by 0.9%. The P100 is 0.3% ahead of the Tesla P100 PCIe 12 GB, 0.3% behind the GeForce RTX 5090, and 1.7% behind the Radeon Pro Vega 64X.

Q: What are the power connector and PSU requirements?

A: The MI100 uses two 8-pin connectors and a 700 W suggested PSU. The P100 uses one 8-pin connector and a 600 W suggested PSU.

Architecture Differences

The MI100 is built on AMD’s CDNA 1.0 architecture, using the Arcturus chip, while the P100 uses NVIDIA’s Pascal architecture with the GP100 chip. The MI100 is fabricated on a 7 nm process at TSMC, while the P100 uses a 16 nm process at TSMC. The MI100 has 25,600 million transistors on a 750 mm² die, giving a density of 34.1M transistors per mm². The P100 has 15,300 million transistors on a 610 mm² die, with a density of 25.1M per mm². The MI100’s memory clock is 1200 MHz (2.4 Gbps effective), while the P100’s is 715 MHz (1430 Mbps effective). Both use HBM2 memory with a 4096-bit bus, but the MI100’s bandwidth is 1.23 TB/s versus 732.2 GB/s for the P100. The MI100 has 7680 shading units, 480 TMUs, and 64 ROPs. The P100 has 3584 shading units, 224 TMUs, and 96 ROPs. Neither card has RT cores or tensor cores listed. The MI100’s FP16 rate is 46.14 TFLOPS (2:1), while the P100’s is 19.05 TFLOPS (2:1). The MI100’s FP32 is 23.07 TFLOPS, and the P100’s is 9.526 TFLOPS.

The MI100’s architecture is compute-focused with no graphics API support (DirectX, OpenGL, and Vulkan are all N/A). The P100 supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3, despite also having no display outputs. The MI100 uses PCIe 4.0 x16, while the P100 uses PCIe 3.0 x16. The MI100’s release date is 2020-11-15, and its predecessor is Radeon Instinct. The P100’s release date is 2016-06-19, with predecessor Tesla Maxwell and successor Tesla Volta. Both are end-of-life. The MI100’s pixel rate is 96.13 GPixel/s, and its texture rate is 721.0 GTexel/s. The P100’s pixel rate is 127.6 GPixel/s, and its texture rate is 297.7 GTexel/s. The MI100’s base clock is 1000 MHz with a boost of 1502 MHz. The P100’s base clock is 1190 MHz with a boost of 1329 MHz.

Specification Differences

The two cards differ in nearly every core specification. The MI100 has 32 GB of HBM2 memory, while the P100 has 16 GB. Memory bandwidth is 1.23 TB/s for the MI100 and 732.2 GB/s for the P100. The MI100 has 7680 shading units, 480 TMUs, and 64 ROPs; the P100 has 3584 shading units, 224 TMUs, and 96 ROPs. FP32 performance is 23.07 TFLOPS for the MI100 versus 9.526 TFLOPS for the P100. FP16 is 46.14 TFLOPS versus 19.05 TFLOPS. The MI100’s pixel rate is 96.13 GPixel/s, lower than the P100’s 127.6 GPixel/s. The MI100’s texture rate is 721.0 GTexel/s, higher than the P100’s 297.7 GTexel/s. Clock speeds differ: base 1000 MHz versus 1190 MHz, boost 1502 MHz versus 1329 MHz. Memory clock is 1200 MHz versus 715 MHz. TDP is 300 W for the MI100 and 250 W for the P100. Power connectors are 2x 8-pin for the MI100 and 1x 8-pin for the P100. Suggested PSU is 700 W versus 600 W. Bus interface is PCIe 4.0 x16 for the MI100 and PCIe 3.0 x16 for the P100.

The MI100 has no listed API support, while the P100 supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3. The MI100’s dimensions are 267 mm in length and 111 mm in height; the P100’s length is 267 mm, with no height listed. Both are dual-slot cards with no display outputs. The MI100’s process node is 7 nm, and the P100’s is 16 nm. Transistor count is 25,600 million versus 15,300 million, and die size is 750 mm² versus 610 mm². Transistor density is 34.1M per mm² versus 25.1M per mm². The MI100’s generation is Instinct (MIx), while the P100’s is Tesla Pascal (Pxx). The MI100’s predecessor is Radeon Instinct, and the P100’s predecessor is Tesla Maxwell, with successor Tesla Volta. The P100 has a launch MSRP of 5,699 USD; the MI100 has no launch MSRP in the data. The MI100’s OpenCL score is 139035, and the P100’s is 79605, with the MI100 winning the head-to-head by 74.7%.

DETAILED SPECIFICATIONS

SPECIFICATION
Instinct MI100
Tesla P100 PCIe 16 GB
Core Specs
Shading Units
7,680
3,584 -53.3%
Shaders
7,680
3,584 -53.3%
TMUs
480
224 -53.3%
ROPs
64
96 +50.0%
Compute Units
120
—
SM Count
—
56
Clocks
Base Clock
1000 MHz
1190 MHz
Boost Clock
1502 MHz
1329 MHz
Memory Clock
1200 MHz 2.4 Gbps effective
715 MHz 1430 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
732.2 GB/s
Cache
L1 Cache
16 KB (per CU)
24 KB (per SM)
L2 Cache
8 MB
4 MB
Performance
Pixel Rate
96.13 GPixel/s
127.6 GPixel/s
Texture Rate
721.0 GTexel/s
297.7 GTexel/s
FP32 (TFLOPS)
23.07 TFLOPS
9.526 TFLOPS
FP64 (TFLOPS)
11.54 TFLOPS (1:2)
4.763 TFLOPS (1:2)
FP16 (TFLOPS)
46.14 TFLOPS (2:1)
19.05 TFLOPS (2:1)
Power
TDP
300 W
250 W
TDP (W)
300
250 -16.7%
Suggested PSU
700 W
600 W
Power Connectors
2x 8-pin
1x 8-pin
Architecture
Architecture
CDNA 1.0
Pascal
GPU Name
Arcturus
GP100
Generation
Instinct (MIx)
Tesla Pascal (Pxx)
Process Size
7 nm
16 nm
Transistors
25,600 million
15,300 million
Die Size
750 mm²
610 mm²
Foundry
TSMC
TSMC
Density
34.1M / mm²
25.1M / mm²
API Support
DirectX
—
12 (12_1)
OpenGL
—
4.6
Vulkan
—
1.3
OpenCL
2.1
3.0
CUDA
—
6.0
Shader Model
—
6.0
Physical
Slot Width
Dual-slot
Dual-slot
Length
267 mm 10.5 inches
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
Launch Price
—
5,699 USD
Production
End-of-life
End-of-life
Predecessor
Radeon Instinct
Tesla Maxwell
Successor
—
Tesla Volta
View Instinct MI100 Details View Tesla P100 PCIe 16 GB Details