Compute·Atlas

Supercomputers/Analysis/Silicon generation decay

Analysis 07

Refresh cadence and silicon generation decay

Of the accelerator generations in this dataset that have actually left the installed base, the median stayed 7 years. The longest-lived is NVIDIA Tesla K20X at 10 years, a part released in 2012 and still running well into the following decade. The median gap between a part being released and appearing in a large system is 1 year, which is the number most often assumed to be zero.

Phase 1 seed dataset, compiled by hand. These rows were built from public operator, laboratory and vendor sources. A mechanical second-reader pass has since fetched every cited source: 32 of 52 systems have a readable citation that names them, and 4 are genuinely weakly sourced. Every claim carries its source and a confidence tier. Treat anything below verified as a lead, not a citation.

Read the shape, not the spikes. This is the installed base of the 52 systems in this dataset: a curated seed chosen for supply-chain diversity, not a census of the market. At that sample size one machine entering or leaving service moves a share line by tens of points, so year-to-year jumps are composition effects rather than market events. The multi-year trends are the part worth quoting; the individual steps usually have one system's name on them, and the table view under the chart will tell you which.

  • NVIDIA
  • Intel
  • AMD
  • Other suppliers
Accelerator generations in the installed base Horizontal range bars showing, for each accelerator generation, the span of years it was present in the installed base of this dataset. IBM PowerXCell 8i IBM · released 2008 IBM PowerXCell 8i IBM in service 2008–2013 peak 1 PFlop/s in 2008 1 system NVIDIA Tesla M2050 NVIDIA · released 2010 NVIDIA Tesla M2050 NVIDIA in service 2010–2010 (still in service) peak 3 PFlop/s in 2010 1 system in service NVIDIA Tesla K20X NVIDIA · released 2012 NVIDIA Tesla K20X NVIDIA in service 2012–2021 peak 18 PFlop/s in 2012 2 systems Intel Xeon Phi 31S1P Intel · released 2013 Intel Xeon Phi 31S1P Intel in service 2013–2018 peak 61 PFlop/s in 2013 1 system Intel Xeon Phi 7250 (Knights Landing) Intel · released 2016 Intel Xeon Phi 7250 (Knights Landing) Intel in service 2016–2023 (still in service) peak 45 PFlop/s in 2017 3 systems in service NVIDIA Tesla P100 NVIDIA · released 2016 NVIDIA Tesla P100 NVIDIA in service 2017–2024 peak 29 PFlop/s in 2017 2 systems NUDT Matrix-2000 NUDT · released 2017 NUDT Matrix-2000 NUDT in service 2018–2018 (still in service) peak 61 PFlop/s in 2018 1 system in service NVIDIA Tesla V100 NVIDIA · released 2017 NVIDIA Tesla V100 NVIDIA in service 2018–2025 (still in service) peak 263 PFlop/s in 2018 3 systems in service NVIDIA A100 SXM4 40GB NVIDIA · released 2020 NVIDIA A100 SXM4 40GB NVIDIA in service 2020–2025 (still in service) peak 200 PFlop/s in 2021 5 systems in service AMD Instinct MI250X AMD · released 2021 AMD Instinct MI250X AMD in service 2022–2025 (still in service) peak 1733 PFlop/s in 2022 2 systems in service NVIDIA A100 SXM4 64GB NVIDIA · released 2020 NVIDIA A100 SXM4 64GB NVIDIA in service 2022–2025 (still in service) peak 241 PFlop/s in 2022 1 system in service Intel Data Center GPU Max 1550 Intel · released 2023 Intel Data Center GPU Max 1550 Intel in service 2023–2025 (still in service) peak 1012 PFlop/s in 2023 1 system in service NVIDIA H100 SXM5 NVIDIA · released 2022 NVIDIA H100 SXM5 NVIDIA in service 2023–2025 (still in service) peak 821 PFlop/s in 2023 5 systems in service AMD Instinct MI300A AMD · released 2023 AMD Instinct MI300A AMD in service 2024–2025 (still in service) peak 1742 PFlop/s in 2024 1 system in service NVIDIA GH200 Grace Hopper Superchip NVIDIA · released 2023 NVIDIA GH200 Grace Hopper Superchip NVIDIA in service 2024–2025 (still in service) peak 435 PFlop/s in 2024 4 systems in service 20082012201620202025
Each bar runs from the first year the part appears in a system here to the last year it was in service. A faded end marked “in service” means the generation has not been retired: the bar is open, not finished.
Table view: Accelerator generations in the installed base
GenerationSupplierReleasedIn serviceYearsRelease→deployPeak installedPeak yearSystems
IBM PowerXCell 8iIBM20082008–2013601.03 PFlop/s20081
NVIDIA Tesla M2050NVIDIA20102010–2010+102.57 PFlop/s20101
NVIDIA Tesla K20XNVIDIA20122012–202110017.59 PFlop/s20122
Intel Xeon Phi 31S1PIntel20132013–20186061.44 PFlop/s20131
Intel Xeon Phi 7250 (Knights Landing)Intel20162016–2023+8044.85 PFlop/s20173
NVIDIA Tesla P100NVIDIA20162017–20248129.36 PFlop/s20172
NUDT Matrix-2000NUDT20172018–2018+1161.44 PFlop/s20181
NVIDIA Tesla V100NVIDIA20172018–2025+81263.1 PFlop/s20183
NVIDIA A100 SXM4 40GBNVIDIA20202020–2025+60200.5 PFlop/s20215
AMD Instinct MI250XAMD20212022–2025+411.73 EFlop/s20222
NVIDIA A100 SXM4 64GBNVIDIA20202022–2025+42241.2 PFlop/s20221
Intel Data Center GPU Max 1550Intel20232023–2025+301.01 EFlop/s20231
NVIDIA H100 SXM5NVIDIA20222023–2025+31820.8 PFlop/s20235
AMD Instinct MI300AAMD20232024–2025+211.74 EFlop/s20241
NVIDIA GH200 Grace Hopper SuperchipNVIDIA20232024–2025+21434.9 PFlop/s20244
How this was calculated

A generation is present in a year if any system carrying it was in service that year. A system is in service from its first-operational date until it is retired.

Two modelling problems had to be fixed before this chart could be honest, and both are worth stating.

First, component edges originally had no time bounds, so a system that was rebuilt with different silicon made both generations look present for its whole life. Tianhe-2 is the case: Intel Xeon Phi from 2013, NUDT Matrix-2000 from 2018. Edges now carry optional era_start and era_end dates, and without them this page would have credited a discontinued Intel part with an extra seven years of installed life.

Second, a system recorded as operational with no retirement date means we found no source saying it was retired, which is an absence of evidence, not evidence of absence. Systems now carry a status_as_of date recording when the status was last supported by a source. Tianhe-1A forces this: read naively it would credit a 2010 Fermi part with sixteen years in service on the strength of a missing citation.

Spans that are still open are excluded from the median, because including a generation that has not finished would bias the figure downward: every currently-deployed part looks short-lived until it is retired. Only the 4 generations with a closed span contribute to the 7-year median.

“Peak installed” is the highest total Rmax of all systems carrying the part in any one year. It is a capability measure, not a unit count.

The read

The headline is how long generations last, and it is longer than the release cadence suggests. Vendors ship a new accelerator every eighteen months to two years; the installed base turns over on something closer to a seven-to-ten-year cycle. Kepler's K20X entered service in 2012 and was still running in 2021. The gap between those two clocks is where most of the confusion about “refresh cycles” lives: a procurement cycle and a product cycle are not the same cycle.

The second pattern is the release-to-deployment lag. Parts do not appear in leadership systems the year they launch; the median here is a year, and the exceptions run longer. That lag compounds with the announcement-to-first-light lag on the schedule page: a system announced around silicon that has not shipped is committing to two uncertainties at once.

The third is displacement, and the clearest case is not a technical one. Intel's Xeon Phi 31S1P has a closed span ending in 2018, and it did not end because something faster arrived: it ended because an export control blocked the upgrade path and NUDT substituted its own part. On this chart that is a generation dying at a policy boundary rather than a performance one, which is exactly the kind of thing a supply-chain view can show and a ranking cannot.

The bars at the right-hand edge are all open, which is the correct rendering and also the limitation. Every current generation looks short-lived because it has not finished yet. Nothing on the right third of this chart should be read as a lifespan.