Accelerator
34 TFLOP/s of FP64 and 67 of FP32, 80 GiB of memory at 3.35 TB/s, and a 700 W ceiling. Its floating-point figures state their precision.
NVIDIA's specification table names the precision of every rate it publishes: FP64 at 34 TFLOP/s, FP64 through the tensor cores at 67, FP32 at 67, and FP16 through the tensor cores at 1,979 with structural sparsity. Other floating-point figures in this dataset, from Sega, Sony and Apple, are bare numbers whose precision the reader must infer. Those figures are provisional; these are approved.
A single "teraflops" figure would obscure the difference. The chip performs double-precision arithmetic at 34 or 67 TFLOP/s depending on which units run it, and half-precision arithmetic at 1,979 when the workload has the required sparsity pattern. The catalog keeps these capabilities in separate comparability groups, so no ratio crosses between them.
This is an accelerator rather than a computer. It has no operating system or input, and a host gives it work. Its memory figure therefore means something different from a console's.
| Component | Role | Count | Clock | Capacity |
|---|---|---|---|---|
| NVIDIA H100 (SXM) NVIDIA states no clock frequency for it, as with every other modern record in this catalog. | GPU | 1 | unknown Nominal clock | — |
| NVIDIA H100 (SXM) GPU memory | Video memory | 1 | — | 80 GiB |
| Metric | Value | Evidence | Basis | Method | Source |
|---|---|---|---|---|---|
| Memory bandwidth whole-system | 3.35 TB/s | Confirmed | Derived | Theoretical peak Computed here from figures the vendor stated One pool of high-bandwidth memory on the package, and nothing else on the card. Adds 1 memory-bandwidth figure obtained by theoretical-peak. Nothing in the catalog is left out of it. A total is only as trustworthy as the decision about what belongs in it, so both the terms and the exclusions are named records. |
| Metric | Value | Evidence | Basis | Method | Source |
|---|---|---|---|---|---|
| Clock frequency gpu | unknown NVIDIA’s specification table for this accelerator gives floating-point rates at four precisions, a memory bandwidth and a power figure, and no clock at all. | Not stated | Vendor-rated | Nominal clock Stated by the vendor Unknown rather than not-applicable, and worth noticing beside the two consumer cards in this catalog: the GeForce GTX 1080 is specified with a base clock and a boost clock, while the data-center part from the same maker is specified with neither. What NVIDIA sells here is throughput, and the frequency that produces it is left inside the figures. |
| Metric | Value | Evidence | Basis | Method | Source |
|---|---|---|---|---|---|
| Peak FP32 rate gpu | 67 TFLOP/s | Confirmed | Vendor-rated | Theoretical peak Stated by the vendor Single-precision rate on the general floating-point units. Equal to this chip's double-precision tensor-core rate, which is a coincidence of two different units and not a statement that precision costs nothing here: on the general units, single precision is twice as fast as double. | |
| Peak FP64 rate gpu | 34 TFLOP/s | Confirmed | Vendor-rated | Theoretical peak Stated by the vendor Double-precision rate on the general floating-point units, as distinct from the tensor cores. Approved rather than provisional because the source names the precision, the first floating-point figure in this catalog of which that is true. It remains a peak: no workload sustains it, and the same chip's tensor cores double it for the same precision, which is recorded separately because a different unit produced it. | |
| Peak FP64 rate gpu | 67 TFLOP/s | Confirmed | Vendor-rated | Tensor core peak Stated by the vendor Double-precision rate through the fourth-generation tensor cores, which perform matrix operations rather than general arithmetic. Twice the general-purpose double-precision rate, and available only to matrix operations shaped the way the tensor cores require. It is in a comparability group of its own for that reason: quoting it beside another machine's general floating-point peak would compare a specialized unit with a general one. | |
| Peak FP16 rate gpu | 1979 TFLOP/s | Confirmed | Vendor-rated | Tensor core peak with sparsity Stated by the vendor Half-precision rate through the tensor cores, with structural sparsity — the source's asterisk footnote reads "With sparsity". Conditional on the data, not just the hardware: the figure assumes a sparsity pattern the workload must actually have, and halves without it. The method records that condition, so this rate cannot be compared with any dense figure, which is why it is recorded separately rather than rounded into a headline number. | |
| Geekbench Compute score, opencl gpu | 341 973 | Confirmed | Measured | Benchmark chart average Published independently geekbench-compute 7 , opencl A dated capture of the benchmark publisher’s rolling average for this graphics processor, read from the 25 July 2026 OpenCL chart. Version 7 calibrates a compute result against a baseline of 100,000, so it may be compared only with another Geekbench 7 OpenCL figure from this curation round, never with a Metal figure, an earlier version or a processor score. The chart’s device string names the memory rather than the form factor; it is read as the SXM module because the PCIe and NVL cards hold rows of their own. | |
| Thermal design power gpu | 700 W | Confirmed | Vendor-rated | Rated power Stated by the vendor "Max Thermal Design Power (TDP): Up to 700W (configurable)" — an upper bound for the module, and one operators may set lower. A cooling-design target for one component, not a machine's consumption, and explicitly configurable downward, so "700 W" is a ceiling that a given deployment may never approach. The metric forbids ratios, which also keeps it away from the whole-system ratings recorded for the PlayStation 2 and the Mac mini: those describe complete computers, this describes a card. |
| Metric | Value | Evidence | Basis | Method | Source |
|---|---|---|---|---|---|
| Memory bandwidth memory | 3.35 TB/s | Confirmed | Theoretical | Theoretical peak Stated by the vendor Peak transfer rate between the accelerator's memory and its processing units. A bus maximum like every other bandwidth figure here, and directly comparable with them: same metric, same scope, same method. That makes it the one place in this record where a multiplier across the whole span of the catalog is legitimate. | |
| Memory capacity whole-system | 80 GiB | Confirmed | Vendor-rated | Design capacity Stated by the vendor Memory on the SXM module. The accelerator has no other, and no host memory is included. NVIDIA writes "80GB", read as 80 GiB in the binary convention this catalog applies to every memory capacity. Recorded at whole-system scope because it is all the memory this device has, but it answers a narrower question than a computer's: work has to be copied here from a host before the accelerator can touch it. |
3 350 000 000 000 B/s
a + b + … sum v1 Adds 1 memory-bandwidth figure obtained by theoretical-peak. Nothing in the catalog is left out of it. A total is only as trustworthy as the decision about what belongs in it, so both the terms and the exclusions are named records.
70×
a ÷ b ratio v1 Compares memory-bandwidth at memory scope, obtained by theoretical-peak. Valid only within this comparability group, and reported to 2 significant digits — the precision of the least precise input. At least one input is a theoretical peak, which is an upper bound rather than an achieved result.
The release date is editorial context: the cited page carries no publication date and states none for the product. The same table's H100 NVL column is not recorded because it describes a different card. Transcribing a second column would double the record without adding a comparison the catalog can make. The BFLOAT16, FP8 and INT8 rates are omitted because the catalog has no metric for those formats. NVLink's 900 GB/s is excluded because the catalog has no metric for interconnect bandwidth.