NVIDIA V100 PCIe 32GB vs NVIDIA V100 SXM2 16GB

The V100 SXM2 16GB delivers 12% more FP16 throughput than the V100 PCIe 32GB (125 vs 112 TFLOPS dense). The V100 PCIe 32GB counters with 16 GB more memory (32 GB vs 16 GB).

V100 PCIe 32GB: Volta, 2018 V100 SXM2 16GB: Volta, 2017

FP16 dense lead

+12%

V100 SXM2 16GB: 125 vs 112 TFLOPS

Memory

32 vs 16 GB

16 GB more for the V100 PCIe 32GB

Bandwidth

900 GB/s vs 900 GB/s

Even

TDP

250 W vs 300 W

V100 PCIe 32GB draws 50 W less

See rental pricing from $0.79/hr Download comparison sheet Printable version →

NVIDIA V100 PCIe 32GB

Full specs →

What is the NVIDIA V100 PCIe 32GB?

The V100 PCIe 32GB is a GPU from NVIDIA on the Volta architecture, an accelerator for AI training, inference and high-performance computing. It is a PCIe add-in card, so it fits standard servers and workstations. V100 PCIe 32GB full specifications →

Can I rent the V100 PCIe 32GB in the cloud?

Flopper is not tracking any live rental listings for the V100 PCIe 32GB yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the V100 PCIe 32GB compare with other accelerators?

This page sets the V100 PCIe 32GB against the V100 SXM2 16GB. The V100 PCIe 32GB is the newer generation of the two. Both are NVIDIA parts; the V100 PCIe 32GB is a PCIe card while the V100 SXM2 16GB is an SXM module. The compare tool builds the same comparison for any two parts in the catalogue. All GPU comparisons →

Where do the V100 PCIe 32GB figures come from?

Every V100 PCIe 32GB figure on this page traces to NVIDIA's own datasheet or specification table. How Flopper verifies specifications →

NVIDIA V100 SXM2 16GB

Leads on FP16 Full specs →

What is the NVIDIA V100 SXM2 16GB?

The V100 SXM2 16GB is a GPU from NVIDIA on the Volta architecture, an accelerator for AI training, inference and high-performance computing. It is an SXM module, the mezzanine format used in dense multi-GPU servers. V100 SXM2 16GB full specifications →

Can I rent the V100 SXM2 16GB in the cloud?

Yes. Cloud providers tracked by Flopper rent the V100 SXM2 16GB by the hour, and the pricing page compares their current rates and the price history side by side. V100 SXM2 16GB rental prices →

How does the V100 SXM2 16GB compare with other accelerators?

This page sets the V100 SXM2 16GB against the V100 PCIe 32GB. The V100 SXM2 16GB is the older generation of the two. Both are NVIDIA parts; the V100 SXM2 16GB is an SXM module while the V100 PCIe 32GB is a PCIe card. The compare tool builds the same comparison for any two parts in the catalogue. All GPU comparisons →

Where do the V100 SXM2 16GB figures come from?

Every V100 SXM2 16GB figure on this page traces to NVIDIA's own datasheet or specification table. How Flopper verifies specifications →

Specifications

V100 PCIe 32GB

V100 SXM2 16GB

Architecture

VoltaVolta

Launch Year

20182017

Form Factor

PCIeSXM

Memory

32 GB16 GB

Memory Bandwidth

900 GB/s900 GB/s

TDP

250 W300 W

Process Node

12nm12nm

Spec

V100 PCIe 32GB

V100 SXM2 16GB

Architecture VoltaVolta

Launch Year 20182017

Form Factor PCIeSXM

Memory 32 GB16 GB

Memory Bandwidth 900 GB/s900 GB/s

TDP 250 W300 W

Process Node 12nm12nm

Performance (TFLOPS)

V100 PCIe 32GB

V100 SXM2 16GB

7 TFLOPS 7.8 TFLOPS

14 TFLOPS 16 TFLOPS

112 TFLOPS 125 TFLOPS

FP6

Precision

V100 PCIe 32GB

V100 SXM2 16GB

7 TFLOPS 7.8 TFLOPS

14 TFLOPS 16 TFLOPS

112 TFLOPS 125 TFLOPS

FP6

Figures are dense. Where a vendor publishes a figure it sits under the dense one and never decides a winner. Hover any precision for its definition, or read the GPU glossary.

FLOPS by Precision

What actually differs

Buy the PCIe 32GB, rent the SXM2 16GB. This is the one V100 pairing where the card that drops into a standard server slot also carries more memory: 32 GB against 16 GB, on the same HBM2 bandwidth, at 250 W instead of 300 W. The price is a lower boost clock, 112 TFLOPS FP16 against 125 TFLOPS FP16, and no NVLink.

Second-hand, that trade favours the PCIe card. The SXM2 module needs a baseboard only an SXM2 server provides, and NVLink earns its keep only when a model splits across GPUs, which twice the memory on one card often avoids.

Renting flips the answer. Flopper tracks a single PCIe 32GB listing at $2.34 per GPU-hour, while the SXM2 16GB rents from $0.17 on demand. If the job fits in 16 GB, rent the SXM2; if it needs 32 GB, check SXM2 32GB rates before paying the PCIe premium.

New to the terms? The glossary explains FP64, FP32, FP16, Sparsity and TFLOPS. For how the generations relate, see NVIDIA architectures explained.

Frequently asked questions

Is the V100 SXM2 faster than the V100 PCIe?

Slightly, on compute. Both use the same Volta die, but the SXM2 16GB boosts higher and runs 125 TFLOPS FP16 and 7.8 TFLOPS FP64 against 112 TFLOPS FP16 and 7 TFLOPS FP64 on the PCIe 32GB. Memory bandwidth is identical at 900 GB/s, so token generation runs at the same pace.

Which V100 should I buy second-hand, PCIe 32GB or SXM2 16GB?

The PCIe 32GB for most buyers. It fits a standard server slot at 250 W with passive cooling, and holds twice the memory. The SXM2 16GB module needs an SXM2 baseboard, so it only makes sense if you already own an SXM2 server or want NVLink between GPUs.

Does NVLink on the SXM2 matter more than the extra 16 GB?

Only when a model must split across GPUs. NVLink 2.0 gives each SXM2 module 300 GB/s against 32 GB/s over PCIe Gen3. A model that fits in 32 GB on one PCIe card never needs that link, and 16 GB fills quickly once weights, gradients and optimiser state share it.

How much does it cost to rent a V100 PCIe 32GB or a V100 SXM2 16GB?

Flopper's listings show one V100 PCIe 32GB listing, Shadeform at $2.34 per GPU-hour on demand. The V100 SXM2 16GB has 7 listings across 4 providers, from $0.09 spot and $0.17 on demand at DataCrunch to $3.06 for an AWS p3.16xlarge.

Where to Rent

NVIDIA V100 PCIe 32GB

No live rental listings tracked for the V100 PCIe 32GB yet.

Browse GPU cloud providers

NVIDIA V100 SXM2 16GB

Rent from

$0.79 / GPU-hr

Lowest of 6 live listings across 3 providers · median $2.34

See all V100 SXM2 16GB rental prices →

Price history and every provider on the pricing page. Some links are affiliate links.

Get Comparison Updates

New GPUs added weekly. Be the first to see how they compare.

Open in Full Compare Tool → V100 PCIe 32GB full specs V100 SXM2 16GB full specs Estimate cluster cost in the calculator →

NVIDIA V100 PCIe 32GB vs NVIDIA V100 SXM2 16GB

NVIDIA V100 PCIe 32GB

What is the NVIDIA V100 PCIe 32GB?

Can I rent the V100 PCIe 32GB in the cloud?

How does the V100 PCIe 32GB compare with other accelerators?

Where do the V100 PCIe 32GB figures come from?

NVIDIA V100 SXM2 16GB

What is the NVIDIA V100 SXM2 16GB?

Can I rent the V100 SXM2 16GB in the cloud?

How does the V100 SXM2 16GB compare with other accelerators?

Where do the V100 SXM2 16GB figures come from?

Specifications

Performance (TFLOPS)

FLOPS by Precision

What actually differs

Frequently asked questions

Is the V100 SXM2 faster than the V100 PCIe?

Which V100 should I buy second-hand, PCIe 32GB or SXM2 16GB?

Does NVLink on the SXM2 matter more than the extra 16 GB?

How much does it cost to rent a V100 PCIe 32GB or a V100 SXM2 16GB?

Where to Rent

NVIDIA V100 PCIe 32GB

NVIDIA V100 SXM2 16GB

Get Comparison Updates

Stay Updated on GPU Specs