Factual AI compute, hardware and data

The Tesla K10 offers 3x single precision performance and 1.8x memory bandwidth (the slide does not state the baseline for comparison).

Mostly true True in substance, wrong in a detail.

Factual: Statements about the present or the past, checked against the record: did they hold up?

The second judge would not resolve this claim; the verdict shown is the first judge's.

The evidence is the person's or their organisation's own records.

Said for an organisation they lead or work for; it counts on both records.

What would count as true or false

Judged as of May 2012 against the Tesla K10 specs NVIDIA published, compared with the obvious baseline, its predecessor Tesla M2090/M2075 (Fermi). TRUE: the published K10 peak single-precision FLOPS is about 3x the predecessor's (roughly 2.7x–3.3x) and its memory bandwidth about 1.8x (roughly 1.65x–1.95x). MOSTLY TRUE: one ratio is right and the other is a little outside that band (for example 2.5x or 2.0x), or the ratios hold only against a slightly different Fermi baseline. MIXED: one ratio holds and the other is clearly wrong. MOSTLY FALSE: both ratios are off but come from real spec comparisons, for example aggregate dual-GPU figures compared against a single GPU in a way that misleads. FALSE: the published specs contradict both ratios against any reasonable Fermi Tesla baseline.

Set before the research began.

Evidence (8)

E1 [PDF] nvidia® tesla® kepler - gpu computing accelerators NVIDIA · May 2012 Own record

“With an aggregate performance of 4.58 teraflop peak single precision and 320 gigabytes per second memory bandwidth for both GPUs put together,”
E2 [PDF] nvidia® tesla® kepler - gpu computing accelerators NVIDIA · May 2012 Own record

“delivers up to 2x the performance for single precision applications compared to the previous generation Fermi-based Tesla M2090 in the same power envelope.”
E3 TESLA™ M-CLASS GPU CoMPUTinG ModULES NVIDIA · August 2011 Own record

“Single precision peak performance is over one Teraflop per GPU.”
E4 TESLA™ M-CLASS GPU CoMPUTinG ModULES NVIDIA · August 2011 Own record

“Memory bandwidth (ECC off) 177 GBytes/sec”
E5 Datasheet: NVIDIA Tesla Kepler Computing Accelerators NVIDIA · May 2012 Own record

“delivers up to 2x the performance for single precision applications compared to the previous generation Fermi-based Tesla M2090 in the same power envelope.”
E6 Datasheet: NVIDIA Tesla Kepler Computing Accelerators NVIDIA · May 2012 Own record

“With an aggregate performance of 4.58 teraflop peak single precision and 320 gigabytes per second memory bandwidth for both GPUs put together”
E7 John Ashley, NVIDIA Speakerdeck (oxfordtkp) · 9 July 2012 Own record

“Fermi Kepler GK104 # of GPUs 1 2 Per GPU Board Single Precision Flops 1.3 TF 2.29 TF 4.58 TF Double Precision Flops 0.66 TF 0.095 TF 0.190 TF # CUDA Cores 512 1536 3072 Memory size 6 GB 4GB 8 GB Memory BW (ECC off) 177.6 GB/s 160GB/s 320 GB/s”
E8 NVIDIA Pioneers New Standard for High Performance Computing With Tesla GPUs Built on Kepler Architecture NVIDIA · 15 May 2012 Own record

“a single Tesla K10 accelerator board features two GK104 Kepler GPUs that deliver an aggregate performance of 4.58 teraflops of peak single-precision floating point and 320 GB per second memory bandwidth.”

Reasoning

NVIDIA's May 2012 datasheet and press release give the Tesla K10 board an aggregate of 4.58 TFLOPS peak single precision and 320 GB/s memory bandwidth (E1, E8). The M2090 datasheet gives 177 GB/s, and an NVIDIA comparison slide lists Fermi at 1.3 TF single precision and 177.6 GB/s (E4, E7). That puts memory bandwidth at about 1.8x, inside the band. Single precision comes out at about 3.5x, slightly above the 2.7x–3.3x range. Both ratios compare the K10's two-GPU board against one Fermi GPU, but they hold board-for-board, so the claim resolves as mostly true.

Cite this claim

Jensen Huang, “The Tesla K10 offers 3x single precision performance and 1.8x memory bandwidth (the slide does not state the baseline for comparison).” (said May 2012). Verdict: mostly true. Reliability API, claim c_e0159a6ff4fa, method 0.1.0+03d23c0e, dataset release v1.1.2. https://epistacklabs.com/claims/c_e0159a6ff4fa

This claim is in dataset release v1.1.2, with the same verdict.

Reply to this claim

More factual claims by Jensen Huang

Mostly true Factual 4 September 2026

“we invented a technology called a GPU, and basically a GPU does parallel processing tasks”

NVIDIA invented the GPU, which does parallel processing tasks such as fluid dynamics, particle physics, quantum chemistry, graphics and AI.
Resolved false Factual 1 June 2026

“Today, NVIDIA’s token cost is the lowest in the world, not by 10%, by X factors, orders of magnitude, all because we did extreme co-design”

NVIDIA's token cost is the lowest in the world, lower than others by large factors and orders of magnitude, due to extreme co-design.
Mostly true Factual 1 June 2026

“It’s the first one to be PCI Express Gen 6.”

Vera is the first CPU to support PCI Express Gen 6.
Mixed Factual 17 March 2026

“We have demonstrated that NVFP4 can achieve inference without loss of precision, while delivering significant performance and energy efficiency improvements, and it is equally suitable for training.”

NVIDIA has demonstrated that NVFP4 can achieve inference without loss of precision, with significant performance and energy-efficiency gains, and is equally suitable for training.
Mixed Factual 17 March 2026

“Over the past two years, the computational demands for inference have increased approximately 10,000 times, while usage has increased approximately 100 times.”

Over the past two years, computational demand for inference rose about 10,000 times while usage rose about 100 times.

All of them

Verdicts are made automatically; how verdicts are made.