News
DealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory’s NVIDIA & AMD Enterprise ExpansionDealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory’s NVIDIA & AMD Enterprise Expansion
EveryNeocloud

MFU

The share of a chip's peak arithmetic speed that a training job actually turns into useful model work.

Also: model FLOPs utilisation

Model FLOPs utilisation compares the arithmetic a model needs per step with what the hardware could theoretically do in the same time. The measure was set out in Google's PaLM paper of 2022. Large training runs commonly report MFU between about 30 and 50 per cent; the rest is lost waiting for memory, for network transfers between GPUs, and to restarts. Raising MFU is worth a great deal of money on a large cluster, which is why providers compete on network design, software tuning and reliability rather than GPU count alone.

See also

Find providers that mention MFU

The weekly digest

Every Monday: the provider of the day, new pages on EveryNeocloud, and the conferences and anniversaries of the week ahead.