NVIDIA Tesla P100 16GB HBM2 — Pascal GP100 Data Center GPU Accelerator
Technical Identity
The form factor of these units is NOT verified and must be confirmed from the unit label before ordering. NVIDIA shipped the Tesla P100 16GB (SKU TESLA-P100-16GB) in two different form factors: a PCIe full-height / full-length add-in card and an SXM2 mezzanine module. An SXM2 module requires an SXM2 baseboard and cannot be installed in a PCIe slot — there is no adapter between the two. Both share the Pascal GP100 GPU, 3584 CUDA cores and 16GB CoWoS HBM2 at 732 GB/s, but differ in rated compute, power and interconnect. Figures below are quoted separately per form factor from NVIDIA’s two Tesla P100 datasheets (Oct 2016). Condition: Refurbished.
Specifications
Brand: NVIDIA
Model / MPN: Tesla P100 16GB (SKU TESLA-P100-16GB)
NVIDIA ordering P/N: not stated — differs by form factor and OEM; verify from unit label
Product type: Data center GPU accelerator
GPU architecture / GPU: NVIDIA Pascal / GP100
CUDA cores: 3584 (identical on both form factors)
GPU memory: 16GB CoWoS HBM2 at 732 GB/s
ECC: Yes (SXM2 datasheet: native, no capacity or performance overhead)
Thermal solution: Passive (both form factors)
Compute APIs: CUDA, DirectCompute, OpenCL, OpenACC
PCIe variant (Tesla P100 for PCIe datasheet): 4.7 TFLOPS FP64, 9.3 TFLOPS FP32, 18.7 TFLOPS FP16; PCIe Gen3; 250 W max; PCIe Full Height/Length.
SXM2 variant (Tesla P100 SXM2 datasheet): 5.3 TFLOPS FP64, 10.6 TFLOPS FP32, 21.2 TFLOPS FP16; NVIDIA NVLink; 300 W max; SXM2 mezzanine module.
Form factor of these units: PCIe or SXM2 — NOT verified, confirm from unit label
Condition: Refurbished
Engineering Context
Pascal GP100 was NVIDIA’s first data center GPU to combine CoWoS packaging with HBM2, which is how the 16GB Tesla P100 reaches 732 GB/s, and it carries a high double-precision ratio. The PCIe card occupies a double-width full-height, full-length slot over PCIe Gen3; the SXM2 module bolts onto a vendor baseboard and uses NVLink at a higher 300 W envelope. Both are passively cooled.
Use Cases
FP64-weighted HPC codes: molecular dynamics, CFD, climate, quantum chemistry.
Legacy CUDA training and inference nodes on Pascal-era toolkits and driver branches.
Like-for-like replacement of a failed P100, if the form factor matches the one fitted.
What This Is NOT
NOT form-factor verified. An SXM2 module requires an SXM2 baseboard and will not fit a PCIe slot; a PCIe card will not fit an SXM2 socket.
NOT the 12GB configuration. This is 16GB HBM2 at 732 GB/s; the 12GB PCIe configuration runs 549 GB/s.
NOT a Volta, Turing, Ampere, Ada or Hopper part. The GPU is Pascal GP100 — the datasheets list no Tensor Cores, no RT Cores and no MIG.
NOT a display or workstation graphics card. The datasheets list no display outputs.
NOT actively cooled. Both datasheets state a passive thermal solution; the chassis must supply directed airflow.
NOT NVLink-equipped in the PCIe form factor. NVLink appears on the SXM2 datasheet only.
NOT supplied with baseboards, risers, power cables, brackets, driver media or licences.
Compatibility & Hard Constraints
Form factor is the hard constraint. Confirm PCIe versus SXM2 from the unit label and against your chassis before ordering.
Typical platforms, PCIe: servers with a double-width, full-height, full-length PCIe Gen3 x16 slot rated for a 250 W passively cooled accelerator, plus the required supplementary power connector.
Typical platforms, SXM2: vendor SXM2 baseboards designed for the 300 W NVLink Tesla P100 module. Baseboard, heatsink and retention hardware are not supplied here.
Cooling: passive on both variants — chassis-directed airflow is mandatory.
Software: a CUDA toolkit and driver branch that still support the Pascal compute capability of GP100.
Pre-deployment verification: form factor, OEM part number on the label, power connector type, chassis thermal rating, driver / CUDA support level.
Frequently Asked Questions (FAQ)
Q. Is this Tesla P100 16GB a PCIe card or an SXM2 module?
The form factor is NOT verified on these units. NVIDIA shipped the Tesla P100 16GB in both a PCIe full-height/full-length form factor and an SXM2 mezzanine form factor. Confirm the form factor from the unit label before ordering: an SXM2 module requires an SXM2 baseboard and cannot be installed in a PCIe slot. Contact T.E.S IT-SOLUTIONS and we will confirm what is in stock.
Q. Which Tesla P100 16GB specifications are identical on both form factors?
Both are Pascal GP100 accelerators with 3584 CUDA cores, 16GB CoWoS HBM2 at 732 GB/s, ECC, a passive thermal solution, and CUDA, DirectCompute, OpenCL and OpenACC compute APIs.
Q. How do the PCIe and SXM2 performance figures differ?
Per NVIDIA's datasheets: Tesla P100 for PCIe is rated 4.7 TFLOPS FP64, 9.3 TFLOPS FP32 and 18.7 TFLOPS FP16 at 250 W over PCIe Gen3. The SXM2 Tesla P100 is rated 5.3 TFLOPS FP64, 10.6 TFLOPS FP32 and 21.2 TFLOPS FP16 at 300 W with NVLink. NVIDIA notes these measurements use NVIDIA GPU Boost.
Q. Does the Tesla P100 16GB have NVLink?
NVLink is listed as the interconnect on the SXM2 Tesla P100 datasheet only; the PCIe datasheet lists PCIe Gen3. Because the form factor of these units is unverified, do not assume NVLink is present until the label is confirmed.
Q. Is the memory 12GB or 16GB, and what is the bandwidth?
This listing is the 16GB configuration at 732 GB/s. NVIDIA's PCIe datasheet quotes 16GB HBM2 at 732 GB/s or 12GB HBM2 at 549 GB/s; the SXM2 datasheet quotes 16 GB HBM2 at 732 GB/s.
Q. Does the Tesla P100 have Tensor Cores or display outputs?
No. The Tesla P100 is a Pascal GP100 compute accelerator. NVIDIA's datasheets list no Tensor Cores and no display outputs; the listed compute APIs are CUDA, DirectCompute, OpenCL and OpenACC.
Q. What condition is supplied by T.E.S IT-SOLUTIONS?
Refurbished. Each Tesla P100 16GB unit receives a cosmetic inspection and, where a matching test platform is available, a power-on and detection check. Because PCIe and SXM2 units trade at different market levels, pricing is set conservatively and final pricing is subject to confirmation of the exact model and form factor. EU stock, global B2B shipping.
Why T.E.S IT-SOLUTIONS
Every unit ships with functional verification where a test platform is available, capture of the unit label and OEM part number, cosmetic inspection, anti-static packaging, and global B2B shipping from EU stock.