Accelerator
2,560 single-precision cores at 1607 MHz, 8,873 GFLOPS at the 1733 MHz boost clock, 8 GiB of GDDR5X at 320 GB/s, and 180 W.
GeForce GTX 1080 Founders Edition cards in a rack, the reference shroud and its blower fan in the foreground.
The first card here sold for installation in a personal computer. The H100 is an accelerator with no operating system or input. This card does the same job for people playing games, so it is also recorded as an accelerator. The host it plugs into is not this record.
Nine years separate it from the GeForce RTX 5090. Both cards are NVIDIA's, and both state a single-precision peak, a memory bandwidth and a board power. The catalog can therefore calculate ratios for the first two quantities across the decade.
NVIDIA states two clocks for this card, and they are two quantities: 1607 MHz is what it is specified to hold, while 1733 MHz is what it reaches when there is headroom. They belong to two comparability groups, so a boost clock cannot be compared with the Commodore 64's crystal-derived clock.
| Component | Role | Count | Clock | Capacity |
|---|---|---|---|---|
| NVIDIA GP104 (GeForce GTX 1080) The base clock. The card also states a 1733 MHz boost clock, which is a separate figure. | GPU | 1 | 1607 MHz Nominal clock 1733 MHz Peak clock | — |
| NVIDIA GeForce GTX 1080 frame buffer | Video memory | 1 | — | 8 GiB |
| Metric | Value | Evidence | Basis | Method | Source |
|---|---|---|---|---|---|
| Memory bandwidth whole-system | 320 GB/s | Confirmed | Derived | Theoretical peak Computed here from figures the vendor stated One pool of GDDR5X on the card, and nothing else on it. 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 |
|---|---|---|---|---|---|
| Launch price whole-system | unknown NVIDIA states no price for this card in either cited document. The one figure in them is an expectation about what add-in-card partners would charge — "pricing is expected to start at $599", hedged as varying by region — which is neither NVIDIA’s own asking price nor a price for the Founders Edition NVIDIA sold itself. | Not stated | Vendor-rated | List price Stated by the vendor |
| Metric | Value | Evidence | Basis | Method | Source |
|---|---|---|---|---|---|
| Clock frequency gpu | 1607 MHz | Confirmed | Vendor-rated | Nominal clock Stated by the vendor Reference card, as specified in NVIDIA's launch table. The base clock: the frequency the card is specified to hold, and the one of its two clocks that belongs in the same row as a fixed crystal frequency. The boost clock is recorded separately, because a ceiling reached when there is thermal headroom answers a different question. | |
| Clock frequency gpu | 1733 MHz | Confirmed | Vendor-rated | Peak clock Stated by the vendor Reference card, the frequency NVIDIA guarantees the card reaches when power and thermal headroom allow. The boost clock, and NVIDIA's own footnote is why it is a separate figure rather than a better version of the base clock: the vendor computes its rate figures from this frequency, not from the one the card is specified to hold. It sits in the same comparability group as the PlayStation 5's variable-frequency ceiling and in a different one from every fixed clock in the catalog. | |
| Peak FP32 rate gpu | 8873 GFLOP/s | Confirmed | Theoretical | Theoretical peak Stated by the vendor 2,560 single-precision cores at the 1733 MHz boost clock, per the whitepaper's footnote. A theoretical peak, not an achieved rate. NVIDIA's table labels this row only "GFLOPs". It is approved as a single-precision rate because the same vendor, in the launch article, describes the parts that produce it as "2560 single-precision CUDA Cores", and because the figure is exactly those cores multiplied by two operations and the boost clock the table's footnote says it uses. That is a precision the vendor stated, in the vendor's own documents, not a reading this catalog supplied, which is what keeps the console teraflops figures provisional. | |
| Geekbench Compute score, opencl gpu | 54 526 | 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 also lists this processor under “GeForce GTX 1080”, at 55 875; the row naming the product as its maker does was taken, and the disagreement is recorded in the conflict ledger. | |
| Thermal design power gpu | 180 W | Confirmed | Vendor-rated | Rated power Stated by the vendor "TDP: 180 W" for the reference card, alongside a 500 W recommended system power that describes a computer rather than this card. NVIDIA's own word for this figure is TDP, so it is recorded as a cooling-design target. The GeForce RTX 5090 nine years later states a "Total Graphics Power" instead, which is a different definition and therefore a different metric. The two sit in different rows on purpose, and neither would allow a ratio in any case. | |
| Transistor count gpu | 7 200 000 000 transistors | Confirmed | Vendor-rated | Die specification Stated by the vendor Whole die, as NVIDIA states it: the graphics, memory-controller and display blocks together. "7.2 billion" is two significant digits, not eleven. NVIDIA states the same count in the article and in the whitepaper, and neither breaks it down by block, so nothing here says how many of these transistors do arithmetic. | |
| Process node gpu | 16 nm | Confirmed | Vendor-rated | Die specification Stated by the vendor Node as named by the foundry and repeated by NVIDIA: "16nm FinFET". A node name rather than a measured dimension. Nothing on this die is sixteen nanometers across, and the metric forbids ratios for that reason: "four times finer than 4 nm" would be a statement about marketing rather than about silicon. |
| Metric | Value | Evidence | Basis | Method | Source |
|---|---|---|---|---|---|
| Memory bandwidth memory | 320 GB/s | Confirmed | Theoretical | Theoretical peak Stated by the vendor Peak transfer rate of the GDDR5X memory on a 256 bit interface. A bus maximum, stated identically by the launch article and the whitepaper. Same metric, same scope, same method as every other theoretical bandwidth here, which is what makes the multiplier against the GeForce RTX 5090 a legitimate one. | |
| Memory bus width memory | 256 bit | Confirmed | Vendor-rated | Design capacity Stated by the vendor Width of the interface between the GP104 and the GDDR5X fitted to the card. Stated in the article’s prose rather than in either specification table, which is why it took a second transcription of a source the catalog already held. | |
| Memory capacity whole-system | 8 GiB | Confirmed | Vendor-rated | Design capacity Stated by the vendor The card's own memory, on a 256 bit interface. No host memory is included. NVIDIA calls it a framebuffer and writes "8 GB", read as 8 GiB in the binary convention this catalog applies to every memory capacity. Recorded at whole-system scope because it is all the memory this card has. As with the H100, it answers a narrower question than a computer's, since work has to be copied here from a host first. |
320 000 000 000 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.
The release date is the date of the cited launch article. The configuration is labeled a reference card on editorial grounds: NVIDIA's table describes the card it built, and partner boards ship with clocks of their own that this record does not cover. The texel fill rate is not recorded because the catalog has no metric for it. The recommended system power is also excluded because it describes a computer this record does not contain.