The sixth Hardware podcast sees Antony, Clive, Harry and Rich chat about all the latest hardware goings on. Discussed in the episode is Nvidia’s new GPU architecture, Fermi. Yes, yes, Nvidia’s got a Fermi – stop sniggering at the back!
[break]
However, Tim’s put us in line about the technical side of things, so this is actually how the MIMD stuff works. Just like previous Nvidia GPUs, the whole chip can either operate in graphics mode or compute mode. In graphics mode, it can do all the rasterisation stuff – pixel, vertex shading and so on – but if you want to run compute-type tasks through the GPU, you need to wait till all that’s done before slinging PhysX instructions (oh, OK, kernels) through the GPU.
However, Fermi can accept two sets of work simultaneously (as long they’re both intended for graphics or compute) and can run 16 kernels of that work concurrently. So if you’ve got two Compute tasks, one with six kernels (think ‘threads’ as a simplification) and the other with ten, then Fermi can handle all of that work at once. This makes Fermi very flexible, in contrast with its name.
I’m sure even that’s an over-simplification, but I’m tired and this pod isn’t going to cast itself… Tim can correct things in the Comments section.
Also discussed is the brilliant Samsung SpinPoint F3 and how we create our Hardware Buyer’s Guide every month.
Listen in for your chance to win Zeno Clash, and don’t forget all our other competitions:
Win a Sapphire Radeon HD 5870 Vapor-X
Win £600-worth of hardware in the reader survey
Discuss your thoughts in the forums.
[break]
However, Tim’s put us in line about the technical side of things, so this is actually how the MIMD stuff works. Just like previous Nvidia GPUs, the whole chip can either operate in graphics mode or compute mode. In graphics mode, it can do all the rasterisation stuff – pixel, vertex shading and so on – but if you want to run compute-type tasks through the GPU, you need to wait till all that’s done before slinging PhysX instructions (oh, OK, kernels) through the GPU.
However, Fermi can accept two sets of work simultaneously (as long they’re both intended for graphics or compute) and can run 16 kernels of that work concurrently. So if you’ve got two Compute tasks, one with six kernels (think ‘threads’ as a simplification) and the other with ten, then Fermi can handle all of that work at once. This makes Fermi very flexible, in contrast with its name.
I’m sure even that’s an over-simplification, but I’m tired and this pod isn’t going to cast itself… Tim can correct things in the Comments section.
Also discussed is the brilliant Samsung SpinPoint F3 and how we create our Hardware Buyer’s Guide every month.
Listen in for your chance to win Zeno Clash, and don’t forget all our other competitions:
Win a Sapphire Radeon HD 5870 Vapor-X
Win £600-worth of hardware in the reader survey
Discuss your thoughts in the forums.
MSI MPG Velox 100R Chassis Review
October 14 2021 | 15:04
Want to comment? Please log in.