According to Ryan Smith, nVidia announced at ISC that they would be shipping P100 in PCI-E form in Q4.
http://www.anandtech.com/show/10433/nvidia-announces-pci-express-tesla-p100
Interestingly there will be two versions, a full and cut. Unlike traditional GPUs though, the cut version has all the CC's of the full version, but has one of the HBM stacks disabled. I could definitely see this being used by AMD/nVidia in the consumer space as well as they start to ship larger numbers of HBM GPUs.
I suppose it's not surprising that NVLINK is also cut, but it would have been interesting if they'd kept it and routed it to a new top edge connector similar to SLI.
I have to say, this is earlier than I expected this product to be sold in this form.
http://www.anandtech.com/show/10433/nvidia-announces-pci-express-tesla-p100
Interestingly there will be two versions, a full and cut. Unlike traditional GPUs though, the cut version has all the CC's of the full version, but has one of the HBM stacks disabled. I could definitely see this being used by AMD/nVidia in the consumer space as well as they start to ship larger numbers of HBM GPUs.
I suppose it's not surprising that NVLINK is also cut, but it would have been interesting if they'd kept it and routed it to a new top edge connector similar to SLI.
I have to say, this is earlier than I expected this product to be sold in this form.
Facebook
Twitter