#Gpus

Early AMD GPUs cost as little as $89 and $99 in 2006, with the most expensive being $449 ($750 adjusted for inflation), but now they cost as much as $1,100.

Plain DDP on a rented 4x RTX A6000 gave 2x. Measured all-reduce bandwidth, then bf16 gradients and accumulation took it to 3.6x. Commands inside.

The right GPU cloud depends on four things: the workload (steady inference, bursty inference, or...

Training LLMs at massive scale brings unique infrastructure challenges, especially as jobs span thousands of GPUs and run for extended periods.

Nvidia's Minerva maps GPUs, networks and jobs in AI data centres to spot failing parts early. IREN, Lambda, IBM and Tesla plan to use it.

AMD is expanding its FSR 4.1 upscaling technology beyond Radeon RX 9000-series GPUs to Xbox Series X/S consoles, Xbox Ally handhelds, laptops, and Radeon RX 6000-series graphics cards.…

Prominent leaker Kepler_L2 recently suggested that Nvidia's next-generation mainstream graphics card, tentatively called the RTX 6060, could be delayed until as late as 2029. Memory supply shortages...

And other fun facts about how Zuckercorp aims to keep a million GPUs running long enough to train a model

Uniform Manifold Approximation and Projection (UMAP) is a dimensionality reduction technique widely used for visualization and feature extraction.

Biomolecular structure prediction is now often run at proteome scale, where the goal is to move an entire worklist through the pipeline efficiently.

Supply chain problems are expanding across more SKUs, lanes, and constraints than ever before, while energy grids are balancing more distributed sources in real…

Stoa is the institutional marketplace to buy and sell GPUs and AI hardware, with verified counterparties and price discovery.

If xAI defaults on its debt, Apollo Global Management ends up in the GPU rental business.

The 3GB GDDR7 chips that the RTX 50 Super GPUs will use reportedly cost twice to thrice as much as the 2GB chips found on vanilla RTX 50-series graphics cards. This would likely push the retail price of these GPUs way beyond Nvidia's target MSRP.