NVIDIA Tesla P40 24GB Specs, Benchmarks & Pricing
The NVIDIA Tesla P40 is a passively cooled, dual-slot PCIe inference accelerator announced in September 2016. It is built on the Pascal GP102 GPU with 3,840 CUDA cores and 24 GB of GDDR5 on a 384-bit bus. It has no Tensor Cores, and its FP16 runs at 1/64 the FP32 rate, so it suits FP32 and INT8 (DP4A) work. Used cards are a common low-cost way to get 24 GB of VRAM for running LLMs at home.
- Release Date: September 13, 2016
- MSRP: $7,358 USD†
- GPU Architecture: pascal
- Hardware-Accelerated GEMM Operations:FP64 FP32 TF32 BF16 FP16 FP8 FP4 INT8 INT4
- CUDA Compute Capability : 6.1
† Estimated MSRP of $7,358 USD. This is an OEM configurator price, not an NVIDIA MSRP, and it dates from about 9 months after the 2016-09-13 launch. The Next Platform ("Tesla GPU Accelerator Bang For The Buck, Kepler To Volta", 2018-03-05) says NVIDIA published no list or suggested street prices for Tesla cards, and that Dell's PowerEdge R730 configurator in June 2017 priced the Tesla P40 at $7,358 (P4 $2,402, P100 PCIe 16 GB $7,278, 12 GB $5,780). An earlier figure of $5,699 from aggregators had no traceable source and was dropped.
Considerations
- Lower FP32 compute performance compared to other GPUs
- Lower FP16 compute performance compared to other GPUs
Specifications for NVIDIA Tesla P40
| Specification | Performance Ranking |
|---|---|
| FP32 TFLOPs | |
| FP16 TFLOPs | |
| Tensor Core Count | Data not available |
| Memory Capacity (GB) | |
| Memory Bandwidth (GB/s) | |
| Int8 TOPs | |
| FP8 TFLOPs | Data not available |
| FP4 TFLOPs | Data not available |
Real-time NVIDIA Tesla P40 GPU Prices
Compare Price/Performance to other GPUs
Compare NVIDIA Tesla P40 to Another GPU
Price History
NVIDIA Tesla P40 Price History
Product Identifiers
Manufacturer Part Numbers (1)
- GPU CHIP
- GP102-895
Available from 1 Partners (1 products)
- HP NVIDIA Tesla P40 24GB GDDR5 Graphics Card
- Q0V80A(part number)
References
- https://images.nvidia.com/content/pdf/tesla/184427-Tesla-P40-Datasheet-NV-Final-Letter-Web.pdf
- https://videocardz.net/nvidia-tesla-p40
- https://www.nextplatform.com/ai/2018/03/05/tesla-gpu-accelerator-bang-for-the-buck-kepler-to-volta/1645835
- https://www.ebay.com/p/27032254618
Notes
- fp32TFLOPS of 12 is the single-precision figure in the NVIDIA Tesla P40 datasheet (CUDA cores, no TF32 path on Pascal). memoryCapacityGB 24, memoryBandwidthGBs 346, maxTDPWatts 250 and the 2016 date are from the same datasheet.
- fp16TFLOPS of 0.183 is FP16 vector throughput at 1/64 of FP32 (about 11.76 TFLOPS at boost clock / 64, as third-party databases give it). The datasheet does not list FP16; the 1/64 rate is a GP102 property and is not in the NVIDIA documents cited here, so treat it as unverified against NVIDIA. Unlike the P100, the P40 has no 2x FP16 path, so validate-specs.ts has a GP102-specific ratio range.
- int8TOPS of 47 is the datasheet INT8 figure. It is the DP4A ALU instruction (a CUDA-core path), the same convention as the V100, so INT8 stays out of supportedHardwareOperations. fp8TFLOPS and fp4TFLOPS are null because Pascal predates both.
- supportedHardwareOperations is an empty array: Pascal has no matrix or tensor-core path. DP4A is an ALU instruction, not a tensor path (see gpu-researcher.md).
- Estimated MSRP of $7,358 USD. This is an OEM configurator price, not an NVIDIA MSRP, and it dates from about 9 months after the 2016-09-13 launch. The Next Platform ("Tesla GPU Accelerator Bang For The Buck, Kepler To Volta", 2018-03-05) says NVIDIA published no list or suggested street prices for Tesla cards, and that Dell's PowerEdge R730 configurator in June 2017 priced the Tesla P40 at $7,358 (P4 $2,402, P100 PCIe 16 GB $7,278, 12 GB $5,780). An earlier figure of $5,699 from aggregators had no traceable source and was dropped.
- supportedCUDAComputeCapability 6.1 is the GP102 compute capability (the P100's GP100 is 6.0).
- Third-party part number Q0V80A is from the eBay product page (the listing names HP as the maker). It is the only source for that part number.