NVIDIA A100 PCIe 40 GB vs NVIDIA A100 SXM4 40 GB Comparison

NVIDIA
GEFORCE

NVIDIA A100 PCIe 40 GB

CORE STATE GA100
VRAM 40 GB
CLOCK SPEED 1410 MHz
TDP 250 W
BUS WIDTH 5120 bit
ARCHITECTURE Ampere
nm
PROCESS 7 nm
LAUNCH DATE 2020
VS
NVIDIA
GEFORCE

A100 SXM4 40 GB

CORE STATE GA100
VRAM 40 GB
CLOCK SPEED 1410 MHz
TDP 400 W
BUS WIDTH 5120 bit
ARCHITECTURE Ampere
nm
PROCESS 7 nm
LAUNCH DATE 2020

PERFORMANCE BENCHMARKS

geekbench_opencl
178,627
201,096
geekbench_vulkan
146,380
173,198

Analysis: NVIDIA A100 PCIe 40 GB vs NVIDIA A100 SXM4 40 GB

The NVIDIA A100 SXM4 40 GB and the NVIDIA A100 PCIe 40 GB share the same GA100 chip, Ampere architecture, and memory configuration, yet they are distinct products with different physical requirements and performance profiles. The SXM4 module edges ahead in compute benchmarks, while the PCIe card offers a more flexible, lower-power form factor for standard servers. The data below breaks down the architectural similarities, benchmark deltas, and specification differences that define these two siblings.

FAQ

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

A: The NVIDIA A100 SXM4 40 GB scores 201,096, which is 12.6% higher than the NVIDIA A100 PCIe 40 GB’s score of 178,627.

Q: Is there a difference in the Geekbench Vulkan results?

A: Yes. The SXM4 scores 173,198, beating the PCIe version’s 146,380 by 18.3%. This is the largest performance gap between the two in the head-to-head results.

Q: Do both cards use the same amount of memory and bandwidth?

A: Yes. Both have 40 GB of HBM2e memory on a 5120-bit bus, delivering 1.56 TB/s of bandwidth.

Q: What is the physical form factor difference?

A: The SXM4 is a module (SXM Module) with no power connectors, while the PCIe version is a dual-slot card measuring 267 mm (10.5 inches) in length and 111 mm (4.4 inches) in height, requiring an 8-pin EPS power connector.

Q: How do the thermal design power (TDP) ratings compare?

A: The SXM4 has a 400 W TDP, while the PCIe card has a 250 W TDP. The recommended power supply also differs: 800 W for the SXM4 versus 600 W for the PCIe card.

Q: Which card has a higher percentile ranking among all GPUs?

A: The SXM4 sits in the 98th percentile, while the PCIe version is in the 97th percentile based on their average benchmark scores.

Architecture Differences

Both cards are built on the same GA100 chip using a 7 nm process at TSMC. They contain identical transistor counts of 54,200 million on a die size of 826 mm², resulting in the same transistor density of 65.6M / mm². The architecture is Ampere, belonging to the Server Ampere (Axx) generation, with a predecessor of Tesla Turing and a successor of Server Ada.

The core configuration is identical: 6,912 shading units, 432 texture mapping units, 160 raster output pipelines, and 432 tensor cores. Neither card has dedicated ray tracing cores. The pixel rate is 225.6 GPixel/s and the texture rate is 609.1 GTexel/s for both. FP32 performance is listed at 19.49 TFLOPS, and FP16 performance is 77.97 TFLOPS (4:1) for both.

The significant architectural difference lies in the base clock. The SXM4 runs at 1095 MHz base, while the PCIe card runs at 765 MHz base. Both share the same 1410 MHz boost clock and the same memory clock of 1215 MHz (2.4 Gbps effective). This base clock delta is the primary driver of the performance differences seen in benchmarks, even though the boost clocks and raw compute specifications match.

Head-to-Head Benchmarks

The head-to-head comparison shows a clean sweep for the SXM4 in both tested workloads. In Geekbench OpenCL, the SXM4 scores 201,096 against the PCIe card’s 178,627. The delta is 12.6%, meaning the SXM4 delivers over a tenth more performance in this compute-oriented test. This result aligns with the higher base clock, which allows the SXM4 to sustain higher throughput in workloads that do not consistently hit the boost ceiling.

In Geekbench Vulkan, the gap widens to 18.3%. The SXM4 scores 173,198, while the PCIe card trails at 146,380. This is a substantial margin, suggesting that the SXM4’s power delivery and thermal headroom (400 W TDP) allow it to maintain higher clocks under graphics API workloads. The PCIe card’s 250 W TDP likely forces more aggressive clock throttling in sustained loads.

The average benchmark score reinforces this trend. The SXM4 has an average score of 187,147, while the PCIe card averages 162,504. This 24,643-point gap (approximately 15.2%) reflects the SXM4’s consistent advantage across the two tests. The SXM4’s nearest rival, the NVIDIA RTX 5000 Ada Generation, scores 184,664 for a delta of 1.3%, meaning the SXM4 barely edges out that card. In contrast, the PCIe card’s nearest rival, the NVIDIA RTX 4500 Ada Generation, scores 166,094 for a delta of -2.2%, meaning the PCIe card actually trails that rival.

Specification Differences

The two cards diverge on several key specification fields, even though their core compute architectures are identical.

  • Base Clock: The SXM4 runs at 1095 MHz, while the PCIe card runs at 765 MHz. This is the most consequential difference for performance.
  • TDP: The SXM4 is rated at 400 W, versus 250 W for the PCIe card. This reflects the SXM4’s more aggressive power envelope.
  • Slot Width: The SXM4 is an SXM Module, while the PCIe card is dual-slot.
  • Power Connectors: The SXM4 has no power connectors, as it draws power from the SXM socket. The PCIe card requires an 8-pin EPS connector.
  • Suggested PSU: The SXM4 requires an 800 W power supply, while the PCIe card recommends a 600 W unit.
  • Dimensions: The PCIe card has listed dimensions of 267 mm (10.5 inches) in length and 111 mm (4.4 inches) in height. The SXM4 has no listed length, height, or width dimensions.
  • Average Benchmark Score: The SXM4 averages 187,147, while the PCIe card averages 162,504.
  • Percentile: The SXM4 is in the 98th percentile, the PCIe card in the 97th.
  • Release Date: The SXM4 was released on 2020-05-13, while the PCIe card followed on 2020-06-21.

All other fields are identical: memory size (40 GB), memory type (HBM2e), bus width (5120 bit), bandwidth (1.56 TB/s), shading units (6,912), TMUs (432), ROPs (160), tensor cores (432), pixel rate (225.6 GPixel/s), texture rate (609.1 GTexel/s), FP32 (19.49 TFLOPS), FP16 (77.97 TFLOPS), boost clock (1410 MHz), memory clock (1215 MHz), bus interface (PCIe 4.0 x16), and display outputs (none).

Where Each One Wins

The NVIDIA A100 SXM4 40 GB wins in every measured benchmark category. In Geekbench OpenCL, it holds a 12.6% lead, and in Geekbench Vulkan, it leads by 18.3%. Its higher base clock and 400 W TDP make it the stronger choice for sustained compute workloads, particularly those that stress the GPU for extended periods. The SXM4 also holds a higher average benchmark score (187,147 versus 162,504) and a higher percentile ranking (98th versus 97th). For users seeking maximum throughput in a dense server environment where the SXM module form factor is supported, the SXM4 is the clear performance pick.

The NVIDIA A100 PCIe 40 GB wins on integration flexibility and power efficiency. Its 250 W TDP and 600 W recommended PSU make it easier to deploy in standard servers with less robust power delivery. The dual-slot PCIe form factor with an 8-pin EPS connector allows installation in conventional chassis, whereas the SXM4 requires a specialized SXM socket and motherboard. The PCIe card’s lower base clock of 765 MHz is a trade-off, but its boost clock of 1410 MHz matches the SXM4, meaning burst workloads can still reach the same peak performance. The PCIe card’s nearest rival delta of -2.2% against the NVIDIA RTX 4500 Ada Generation shows it competes closely with newer cards in its segment, though it does not top them.

For pure benchmark scores, the SXM4 is the better performer. For environments that cannot accommodate an SXM module or that prioritize lower power consumption and simpler cooling, the PCIe card remains a capable option with identical memory capacity and bandwidth. The choice hinges on whether raw compute (SXM4) or deployment practicality (PCIe) matters more for the specific server infrastructure.

DETAILED SPECIFICATIONS

SPECIFICATION
A100 PCIe 40 GB
A100 SXM4 40 GB
Core Specs
Shading Units
6,912
6,912 0.0%
Shaders
6,912
6,912 0.0%
TMUs
432
432 0.0%
ROPs
160
160 0.0%
SM Count
108
108 0.0%
Clocks
Base Clock
765 MHz
1095 MHz
Boost Clock
1410 MHz
1410 MHz
Memory Clock
1215 MHz 2.4 Gbps effective
1215 MHz 2.4 Gbps effective
Memory
Memory Size
40 GB
40 GB
VRAM (MB)
40,960
40,960 0.0%
Memory Type
HBM2e
HBM2e
Memory Bus
5120 bit
5120 bit
Bandwidth
1.56 TB/s
1.56 TB/s
Cache
L1 Cache
192 KB (per SM)
192 KB (per SM)
L2 Cache
40 MB
40 MB
Performance
Pixel Rate
225.6 GPixel/s
225.6 GPixel/s
Texture Rate
609.1 GTexel/s
609.1 GTexel/s
FP32 (TFLOPS)
19.49 TFLOPS
19.49 TFLOPS
FP64 (TFLOPS)
9.746 TFLOPS (1:2)
9.746 TFLOPS (1:2)
FP16 (TFLOPS)
77.97 TFLOPS (4:1)
77.97 TFLOPS (4:1)
AI/RT
Tensor Cores
432
432 0.0%
BF16
311.84 TFLOPS (16:1)
311.84 TFLOPS (16:1)
TF32
155.92 TFLOPs (8:1)
155.92 TFLOPs (8:1)
Power
TDP
250 W
400 W
TDP (W)
250
400 +60.0%
Suggested PSU
600 W
800 W
Power Connectors
8-pin EPS
None
Architecture
Architecture
Ampere
Ampere
GPU Name
GA100
GA100
Generation
Server Ampere (Axx)
Server Ampere (Axx)
Process Size
7 nm
7 nm
Transistors
54,200 million
54,200 million
Die Size
826 mm²
826 mm²
Foundry
TSMC
TSMC
Density
65.6M / mm²
65.6M / mm²
API Support
OpenCL
3.0
3.0
CUDA
8.0
8.0
Physical
Slot Width
Dual-slot
SXM Module
Length
267 mm 10.5 inches
—
Height
111 mm 4.4 inches
—
Outputs
No outputs
No outputs
Bus Interface
PCIe 4.0 x16
PCIe 4.0 x16
Other
Production
End-of-life
End-of-life
Predecessor
Tesla Turing
Tesla Turing
Successor
Server Ada
Server Ada
View A100 PCIe 40 GB Details View A100 SXM4 40 GB Details