With Hyper-Q, Kepler supports 32 concurrent work queues versus one on Fermi.
Resolved true The record shows it happened or was so.
Factual: Statements about the present or the past, checked against the record: did they hold up?
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
True: NVIDIA's Kepler (GK110) documentation, such as the Kepler GK110 whitepaper or GTC 2012 materials, states that Hyper-Q provides 32 simultaneous hardware work queues, compared with a single work queue on Fermi. Mostly true: 32 queues is confirmed, but the Fermi comparison is slightly imprecise, or Hyper-Q applies only to GK110 or compute capability 3.5 rather than all Kepler parts. Mostly false: the queue count or the Fermi baseline is materially wrong while Hyper-Q did increase concurrency. False: Kepler's Hyper-Q did not provide 32 concurrent work queues.
Set before the research began.
Evidence (4)
-
E1 Whitepaper (NVIDIA's Next Generation CUDA Compute Architecture: Kepler GK110/210, V1.1) NVIDIA Own record
“Hyper-Q increases the total number of connections (work queues) between the host and the CUDA Work Distributor (CWD) logic in the GPU by allowing 32 simultaneous, hardware-managed connections (compared to the single connection available with Fermi).”
-
E2 www.nvidia.com · content · damDatasheet: NVIDIA Kepler Next-Generation Cuda Compute ... NVIDIA · May 2012 Own record
“by allowing 32 simultaneous, hardware managed connections, compared to the single connection available with Fermi.”
-
E3 Microsoft PowerPoint - Inside Kepler.pptx [Read-Only] NVIDIA (GTC 2012 session S0642, 'Inside Kepler' by Lars Nyland and Stephen Jones) Own record
“Kepler allows 32-way concurrency”
-
E4 Microsoft PowerPoint - Inside Kepler.pptx [Read-Only] NVIDIA (GTC 2012 session S0642, 'Inside Kepler' by Lars Nyland and Stephen Jones) Own record
“But CUDA streams multiplex into a single queue”
Reasoning
NVIDIA's Kepler GK110/210 architecture whitepaper (E1) says Hyper-Q allows "32 simultaneous, hardware-managed connections (compared to the single connection available with Fermi)." NVIDIA's May 2012 Kepler compute architecture datasheet (E2) uses the same wording. The GTC 2012 'Inside Kepler' slides say Kepler allows 32-way concurrency [E3] and that CUDA streams multiplex into a single queue [E4]. That matches the True criterion, which asks for NVIDIA's own Kepler GK110 documentation, though all of these sources are NVIDIA's own.
Cite this claim
Jensen Huang, “With Hyper-Q, Kepler supports 32 concurrent work queues versus one on Fermi.” (said May 2012). Verdict: resolved true. Reliability API, claim c_d1757cbb8222, method 0.1.0+03d23c0e, dataset release v1.1.2. https://epistacklabs.com/claims/c_d1757cbb8222
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
“we invented a technology called a GPU, and basically a GPU does parallel processing tasks”
-
Resolved false
“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”
-
Mostly true
“It’s the first one to be PCI Express Gen 6.”
-
Mixed
“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.”
-
Mixed
“Over the past two years, the computational demands for inference have increased approximately 10,000 times, while usage has increased approximately 100 times.”
Verdicts are made automatically; how verdicts are made.