Quote:
We present to you the interview of Mark Papermaster, CTO, AMD who talks about Ryzen, Radeon Instinct, VEGA and a lot other interesting stuff.
I believe Mr. Papermaster has done an excellent job. Congratulations on your teams success.
It is better than what nVIDIA did with NVLink. nVIDIA cannot offer a NVLink between the GPUs and CPUs if the CPUs are x86 based. AMDs solution works with x86 as well as other CPU designs (being an open source solution and seeing as AMD have an x86 license).Originally Posted by Defoler
Their CPU to GPU server connection and P2P connection sounds a lot like what nvidia did with ibm on the nvlink connection, except AMD also own the CPU so they can do it without needing someone else.
Also 14/16nm seems to be on their plan to stay for several good years as well, so no 10nm anytime soon from AMD.
The interview doesn't really have any extra information, but at least they are strongly believe in their next architectures.
At least they have been working hard to stay competitive. I hope they actually deliver, but I'll keep my enthusiasm on a low burn until the CPUs and GPUs are actually out, not just a paper-mastered (his name is killing me)
Quote:Originally Posted by Defoler
Their CPU to GPU server connection and P2P connection sounds a lot like what nvidia did with ibm on the nvlink connection, except AMD also own the CPU so they can do it without needing someone else.
Also 14/16nm seems to be on their plan to stay for several good years as well, so no 10nm anytime soon from AMD.
The interview doesn't really have any extra information, but at least they are strongly believe in their next architectures.
At least they have been working hard to stay competitive. I hope they actually deliver, but I'll keep my enthusiasm on a low burn until the CPUs and GPUs are actually out, not just a paper-mastered (his name is killing me)
Remember AMD buying SeaMicro awhile back?Originally Posted by Mahigan
It is better than what nVIDIA did with NVLink. nVIDIA cannot offer a NVLink between the GPUs and CPUs if the CPUs are x86 based. AMDs solution works with x86 as well as other CPU designs (being an open source solution and seeing as AMD have an x86 license).
Bandwidth wise they are similar (except that AMD has far more bandwidth with their solution when it comes to the P to P communication between GPUs (512 GBps vs 160GBps for NVLink). NVLink can deliver around 40GBps of bandwidth between other devices where as Infinity Fabric can deliver anywhere between 30 to 512GBps for the same task (customizeable).
Looks like NVLink is far slower than Infinity Fabric as things stand right now (based on what I have read so far).
I honestly, based on current performance figures, see AMDs HPC solution as being more than extremely competitive. RyZen looks like it might make quick work of IBMs CPUs and the Vega platform looks like it will be superior to the GP100 in terms of compute capabilities.
AMD have ported CUDA to OpenCL as well therefore I don't see CUDA as being a roadblock anymore. Looks like competition is back in this market
Hang on a second, I thought that that was a terrible business decision and another nail in the coffin for the crumbling AMD that really needed to be bought out by someone else (Samsung LOL) in order to succeed?
Keep in mind, we still don't have a single product from AMD yet. No Zen, No Vega, No Infinity whatever, so before we all officially declare that AMD is back to it's former glory, lets wait until we actually get something other than an interview and some slides.