V100 SXM2 32GB Specifications
First tensor-core datacenter GPU. No BF16 or TF32, so modern training recipes need code changes.
Buy it for
DGX-1 and OEM HGX V100 systems already in service. NVLink between modules gives it far better multi-GPU scaling than the PCIe version.
Look elsewhere if
You are buying a card to install. SXM2 needs a matching baseboard, so the module alone is useless.
Will it fit your system?
This is not a card you can install
SXM2 is a socketed module that mounts to an OEM baseboard, not a PCIe slot. You buy it inside a complete system — it cannot go into an existing desktop or a standard server under any circumstances.
- Form factor
- SXM2
- Cooling
- passive
- Power connectors
- None — SXM2 module, powered by the baseboard
- Board power
- 300 W
- Interface
- PCIe Gen 3
Supported formats & features
What the V100 SXM2 32GB does and does not support in hardware.
Blackwell-generation feature — B200 has it
Predates hardware FP8; use INT8 or BF16 instead — H100 has it
FP16 only, so training recipes may need loss scaling
Not available on this architecture
Quantised inference supported
Not supported on this generation
Precision must be managed manually
300 GB/s GPU-to-GPU
Compute-only die — no ray tracing hardware
Corrects single-bit errors — expected for long training runs
What fits in memory
Largest model that fits on one V100 SXM2 32GB, assuming ~20% of the 32 GB is reserved for KV cache and activations.
| Model |
|---|
| Llama 8B |
| Llama 70B |
| Mixtral 8x22B |
| Llama 405B |
Weights only. Long contexts, large batch sizes and optimiser state during training all need substantially more — full fine-tuning typically needs 3–4× the weight memory.
Workload character
Plenty of bandwidth per unit of compute. Memory-bound work — single-stream token decode above all — runs close to its ceiling, and compute is usually what runs out first.
Cost per unit of compute
Breakeven counts hardware cost against rental only. Power, cooling, hosting and depreciation push the real crossover substantially later — see colocation pricing for what hosting actually costs.
Cost to run
Electricity for one card at 100% duty, including cooling and PSU overhead (3,679 kWh/year).
At the US average rate, three years of continuous operation costs $1,325 — roughly 15% of the card's $9,000 launch price, before any cooling or hosting fees.
Availability & market
Cheap as loose modules precisely because so few buyers can use them. Buy the whole system or nothing.
Discontinued. Expect secondary-market stock only, with no manufacturer warranty and limited driver support ahead.
Memory
- Capacity
- 32 GB
- Type
- HBM2
- Bandwidth
- 900 GB/s
- Bus width
- 4,096-bit
- GB per TFLOPMemory relative to compute — higher favours large models
- 0.256
Compute
- FP64
- 7.8 TFLOPS
- FP32
- 15.7 TFLOPS
- TF32 dense
- —
- FP16 dense
- 125 TFLOPS
- FP16 sparseRequires structured sparsity; rarely reached in practice
- —
- FP8 dense
- —
- INT8 dense
- —
Silicon
- Codename
- GV100
- Process node
- TSMC 12nm FFN
- Transistors
- 21.1B
- Die size
- 815 mm²
- CUDA cores
- 5,120
- Tensor cores
- 640
- SM count
- 80
- Boost clock
- 1,530 MHz
Platform & power
- TDP
- 300 W
- TFLOPS per wattDense FP16 ÷ TDP — drives your power bill at scale
- 0.42
- Form factor
- SXM2
- Cooling
- passive
- Interconnect
- NVLink 2.0
- Interconnect speed
- 300 GB/s
- PCIe generation
- Gen 3
- MIG
- No
- Launch
- 2018-03
- MSRP at launch
- $9,000
V100 SXM2 32GB — common questions
Answered from the specifications above. Questions this card's data cannot answer with confidence are left out rather than guessed.
- How much VRAM does the V100 SXM2 32GB have?
- The V100 SXM2 32GB has 32 GB of HBM2 memory with 900 GB/s of bandwidth. In practice plan on roughly 25 GB for weights: KV cache, activations and fragmentation take the rest, and long contexts or large batches take considerably more.
- What size model can the V100 SXM2 32GB run?
- On a single V100 SXM2 32GB, about 12B parameters at FP16. Anything larger has to be split across several GPUs. These are weight-only figures — a long context window can consume tens of gigabytes on its own.
- Can you put the V100 SXM2 32GB in a desktop PC?
- No. The V100 SXM2 32GB is a SXM2 module, not a PCIe card. It bolts to an OEM baseboard and is only sold inside a complete system, so there is no way to install one in a desktop or a standard server.
- How much power does the V100 SXM2 32GB use?
- The V100 SXM2 32GB is rated at 300 W board power. It draws that from the baseboard rather than from PSU cables, so there is nothing to wire up yourself.
- How much does it cost to run the V100 SXM2 32GB?
- At the US commercial average of $0.12/kWh, running one V100 SXM2 32GB continuously costs about $442 a year in electricity. That figure includes cooling and power-supply overhead at a factor of 1.4 over board power, and over three years comes to roughly 15% of what the card cost at launch.
- Does the V100 SXM2 32GB support FP8?
- No. The V100 SXM2 32GB has no hardware FP8 path. FP8-quantised checkpoints will still load, but the values are upcast before they reach the tensor cores, so you get the memory saving without the speedup.
- Does the V100 SXM2 32GB support FP4?
- No. The V100 SXM2 32GB predates hardware FP4. You can run 4-bit quantised weights on it and the memory saving is real, but they are dequantised before compute, so there is no arithmetic speedup.
- Does the V100 SXM2 32GB support INT8?
- Yes. The V100 SXM2 32GB supports INT8 acceleration. INT8 remains the most widely supported quantised inference format across frameworks.
- Does the V100 SXM2 32GB support BF16?
- No. The V100 SXM2 32GB predates bfloat16 support and can only use FP16 for half-precision work. FP16 training on it usually needs loss scaling to stay numerically stable.
- Does the V100 SXM2 32GB have NVLink?
- Yes. The V100 SXM2 32GB supports NVLink (NVLink 2.0) at 300 GB/s, so GPU-to-GPU traffic bypasses PCIe. This matters for tensor-parallel serving, where every token crosses the link.
- Does the V100 SXM2 32GB have RT cores?
- No. The V100 SXM2 32GB is a compute-only part with no ray-tracing hardware. For rendering or virtual workstations, look at the L40S or an RTX workstation card instead.
- Does the V100 SXM2 32GB have ECC memory?
- Yes. The V100 SXM2 32GB has ECC, so single-bit memory errors are corrected rather than silently corrupting a long run.
- Is the V100 SXM2 32GB worth buying?
- Buy it for: DGX-1 and OEM HGX V100 systems already in service. NVLink between modules gives it far better multi-GPU scaling than the PCIe version. Look elsewhere if: You are buying a card to install. SXM2 needs a matching baseboard, so the module alone is useless.