What will be the maximum achievable flop utilization on the next generation of Nvidia server chips?
32
602
แน912แน1.7k
2025
1D
1W
1M
ALL
1.1%
<30%
1.5%
30-40%
7%
40-50%
10%
50-60%
27%
60-70%
24%
70-80%
9%
80-90%
20%
90-100%
Concretely, what is the best FLOPS (Floating Point Operations per Second) the next generation of Nvidia server cards will be able to achieve on fp16 matrix multiplications on matrices generated by the normal distribution, divided by the maximum theoretical FLOPS that Nvidia reports?
For example, for A100s, it's possible to achieve about 280+ TeraFLOPS out of a maximum of 312 TeraFLOPS, for a maximum flop utilization of ~90%.
On H100s, it seems to be closer around 700 TeraFLOPS, out of a maximum of 1000.
Will resolve when values seem clear after HNext cards are released, or a maximum of one year after Nvidia announces it.
To clarify, this will be for the B100, not the B200.
Get แน600 play money
Related questions
Will the Groq chip inspire Nvidia/AMD to produce radically new AI chips before 2026?
50% chance
Will NVIDIA have more than 75% of the Data center market share by revenue in 2024?
44% chance
Will NVIDIA maintain a >=75% of the Data center market share for at least 4 quarters by the end of 2025?
41% chance
At least one of the most powerful neural nets at end of 2026 will be trained using 10^27 FLOPs
77% chance
How many FLOPs will go into training the first ASL-3 model?
Will the purchase of 3k NVIDIA H100 chips through Saudi's KAUST lead to a functional form of generative AI by June 2024?
51% chance
Will a serious competitor to NVIDIA in the AI chip space emerge before EOY 2027?
71% chance
In how many startups will Nvidia invest in 2024?
Will there be a computing cluster with 10^20 FLOP/s before end of 2024?
67% chance
Will Nvidia still retain over 50% market share of PC gamer GPU usage by 2030
50% chance