A longitudinal register of GPU models observed on the public fast network. It records presence and worker count, not ownership, capacity, utilisation, price or performance.
Latest census sample: 22 Sep 2026. Page regenerated 2026-09-23 05:21 UTC from the dated archive.
280 days · 47 GPU models · 3,692 model-day observations · 20 models present now · first observed 17 Dec 2025
A dated register of the GPU models observed on the fast network. Thickness is observed worker count on a shared scale. A band that stops is a model no longer observed; LAST OBSERVED marks its final dated appearance in this window. Persistent absence is consistent with operators withdrawing that model, but the census does not establish why.
23 models plotted. 24 further GPU modeles peaked below 5 and are retained in the complete census and the table below.
| GPU model | First observed | Last observed | Now | Peak | In figure |
|---|---|---|---|---|---|
| RTX 5090 | 2025-12-17 | 2026-09-22 | 141 | 151 | yes |
| RTX 4090 | 2025-12-17 | 2026-09-22 | 69 | 69 | yes |
| RTX PRO 6000 Blackwell Server Edition | 2025-12-17 | 2026-09-22 | 19 | 25 | yes |
| RTX 5080 | 2025-12-17 | 2026-09-22 | 19 | 20 | yes |
| RTX PRO 5000 Blackwell | 2026-05-29 | 2026-09-22 | 6 | 9 | yes |
| RTX PRO 6000 Blackwell Workstation Edition | 2025-12-17 | 2026-09-22 | 6 | 8 | yes |
| GB10 | 2026-08-26 | 2026-09-22 | 5 | 5 | yes |
| RTX PRO 6000 Blackwell Max-Q Workstation Edition | 2026-03-28 | 2026-09-22 | 4 | 4 | yes |
| RTX 3090 | 2025-12-17 | 2026-09-22 | 2 | 42 | yes |
| RTX PRO 4000 Blackwell | 2026-09-06 | 2026-09-22 | 2 | 6 | yes |
| RTX 4090 D | 2026-01-06 | 2026-09-22 | 2 | 3 | yes |
| RTX 5070 Ti | 2025-12-17 | 2026-09-22 | 2 | 3 | yes |
| RTX 4080 | 2025-12-17 | 2026-09-22 | 2 | 2 | yes |
| RTX 5090 D | 2026-03-23 | 2026-09-22 | 2 | 2 | yes |
| RTX 3090 Ti | 2025-12-17 | 2026-09-22 | 1 | 6 | yes |
| RTX PRO 2000 Blackwell | 2026-09-10 | 2026-09-22 | 1 | 5 | yes |
| RTX 5060 Ti | 2025-12-31 | 2026-09-22 | 1 | 5 | yes |
| RTX PRO 4500 Blackwell | 2026-04-15 | 2026-09-22 | 1 | 3 | yes |
| CMP 170HX | 2026-08-28 | 2026-09-22 | 1 | 2 | yes |
| RTX 4090 Laptop GPU | 2026-08-29 | 2026-09-22 | 1 | 2 | yes |
| H100 80GB HBM3 | 2025-12-17 | 2026-09-15 | 0 | 21 | yes |
| B200 | 2025-12-17 | 2026-03-08 | 0 | 12 | yes |
| L40S | 2025-12-17 | 2026-09-15 | 0 | 11 | yes |
| L4 (22GB) | 2026-03-02 | 2026-03-11 | 0 | 4 | no |
| H100 NVL | 2025-12-17 | 2026-05-19 | 0 | 4 | no |
| RTX 6000 Ada Generation | 2025-12-17 | 2026-09-19 | 0 | 4 | no |
| Unknown GPU | 2026-03-12 | 2026-07-05 | 0 | 4 | no |
| RTX 4080 SUPER | 2025-12-17 | 2026-09-15 | 0 | 3 | no |
| A100 80GB PCIe | 2025-12-25 | 2026-09-13 | 0 | 3 | no |
| RTX A6000 | 2025-12-17 | 2026-09-19 | 0 | 2 | no |
| B300 SXM6 AC | 2025-12-17 | 2025-12-17 | 0 | 2 | no |
| RTX 6000D | 2026-09-16 | 2026-09-16 | 0 | 2 | no |
| RTX 3080 | 2026-05-16 | 2026-09-10 | 0 | 1 | no |
| RTX 4060 Ti | 2026-09-21 | 2026-09-21 | 0 | 1 | no |
| L4 | 2026-02-24 | 2026-03-10 | 0 | 1 | no |
| H200 | 2026-01-07 | 2026-03-30 | 0 | 1 | no |
| RTX A4000 | 2026-03-06 | 2026-07-22 | 0 | 1 | no |
| RTX 4090 (24GB) | 2026-03-05 | 2026-03-11 | 0 | 1 | no |
| RTX A5000 | 2025-12-17 | 2025-12-20 | 0 | 1 | no |
| RTX 4070 Ti SUPER | 2026-04-13 | 2026-07-30 | 0 | 1 | no |
| H100 PCIe | 2026-01-02 | 2026-09-16 | 0 | 1 | no |
| Graphics Device | 2026-07-26 | 2026-07-27 | 0 | 1 | no |
| RTX PRO 5000 72GB Blackwell | 2026-09-06 | 2026-09-06 | 0 | 1 | no |
| A800 80GB PCIe | 2025-12-25 | 2026-01-23 | 0 | 1 | no |
| A100-SXM4-40GB | 2026-06-03 | 2026-09-16 | 0 | 1 | no |
| L40 | 2025-12-25 | 2025-12-25 | 0 | 1 | no |
| RTX A4500 | 2026-04-26 | 2026-04-26 | 0 | 1 | no |
Observed workers are concentrated: the leading GPU model holds 141 of 287 observed workers (49%). 3 established GPU models no longer reach the present.
The census is a separate open dataset. It is not a performance ranking and never weights the goodput ratios.
Scope. One public, unauthenticated fast-network census, sampled once per dated day from 2025-12-17 through 2026-09-22. Counts show observed workers, not capacity, utilisation, price, ownership or the reason a model appeared or disappeared.
Method. Lifelines include every GPU model present now or peaking at 5 workers or more. Band thickness uses one shared scale. The complete table and downloads retain every observed model, including those below the figure threshold.
Separate measurement. Workload-specific production goodput lives at /market. The census is context only and is not combined with it.
Citation. Reuse is licensed under CC BY 4.0.
REPLICATE THIS RECORD ON A SECOND FLEET
Publish daily workload, GPU, attempts, completions and occupied GPU-hours under your own fleet identifier. Fleets are never averaged; earnings and worker identity are not accepted. Protocol validation is the gate - submitting does not mean acceptance.