Well as pointed out the 8 is for 4 Dies per CPU in a Dual CPU system. So it explains not having 2 Numa nodes on Ryzen and doesn't eliminate the Chance that TR has 2 nodes. But its all on how the board/CPU report themselves in windows. Numa is about maximising server task performance to the nth degree. Making apps realize that just starting another thread on another location would be sub optimal. It makes absolute sense to Node out even the CPU on EPYC because its those circumstances that AMD themselves majorly pushed Numa way back in the day. That said TR is for desktop program and workstation workloads. The last thing AMD wants is the their "ThreadRipper" to not be doing encode or rendering work across 16c because it's not Numa aware. Most desktop apps aren't going to be aware and whatever latency penalty happens when spanning all 16c, isn't going to be anywhere near as bad as only using 8.Well that's what I was thinking, that the OS doesn't see it on Ryzen, which would mean it's not simply a function of a node per CCX. 8 CCX in Epyc I thought explained 8 nodes. But then that'd mean two on Ryzen. And then whatever TR has in terms of total dies (is it 4 or 8?).
Either way, it'll be interesting to see the performance of TR when it appears.
its all on how the board/CPU report themselves in windows. Numa is about maximising server task performance to the nth degree. Making apps realize that just starting another thread on another location would be sub optimal. It makes absolute sense to Node out even the CPU on EPYC because its those circumstances that AMD themselves majorly pushed Numa way back in the day. That said TR is for desktop program and workstation workloads. The last thing AMD wants is the their "ThreadRipper" to not be doing encode or rendering work across 16c because it's not Numa aware. Most desktop apps aren't going to be aware and whatever latency penalty happens when spanning all 16c, isn't going to be anywhere near as bad as only using 8.
On Windows aplications will do the same optimization that games apply, using thread affinity masks to avoid unwanted bottlenecks or move unavoidable bottlenecks between threads that don't need to communicate as much.So given that some workstation tasks are either dependent on other data in other threads and/or sensitive to latency I have to wonder if this is going to be an issue regardless of what they do. Any thoughts?
Hard to say there are certainly jobs that are more sensitive to memory or cache latency. But I don't know if we really have a good feel for what desktop solutions would run afoul of this. But even in those tasks I can't see the latencies being hurt enough to be worth only having an R7. I mean if they can use 16 cores, it might not be 16x as fast as 1 core, but it's still going to be better than 12 (pretend that we can get a 12c die).Got it. Makes perfect sense. Thanks.
So given that some workstation tasks are either dependent on other data in other threads and/or sensitive to latency I have to wonder if this is going to be an issue regardless of what they do. Any thoughts?
Wasn't that the expected clock? its 2x1700XWow, base clock of 3.4 GHz? At what TDP?
Also still less than the $1.1k they could charge for it (basically 2x the 1800x) more more than 2x the 1700x. So . . . interesting.
AMD has priced their 16c part against Intel's 10c Skylake-X part. Let the games begin!
Wasn't that the expected clock? its 2x1700X
Anandtech:
Threadripper 1950X - 16/32 3.4/4.0 - $999 - Cinebench 15 score 3062
Threadripper 1920X - 12/24 3.5/4.0 - $799 - Cinebench 15 score 2431
There is still room below the top bins for cheaper 16c and 12c SKUs.
Remember Bits and Chips' starting the rumor craze of 16C TR at $850 that blew the internet up? Fake news (and bad memes) making machine or just a guy that speculates and people take his words as gospel?
AMD Ryzen™ Threadripper™ and Ryzen™ 3 Product Updates [Video by AMD]
The $999 one is for the 1950x the top-end 16c/32t so this does not invalidate the BnC's claim yet. BnC was only speculating $850 for the bottom-tier 16c/32t, which could very well be the vanilla 1950 (analogous to the vanilla 1700).
Do they really need to? Anything that doesn't meet the quality part could just be sold as a Ryzen. By pricing this so aggressively there probably isn't room for anything cheaper.
The scaling in Cinebench R15 from R7 1700X to TR 1950X is almost 100%.
Do they really need to? Anything that doesn't meet the quality part could just be sold as a Ryzen. By pricing this so aggressively there probably isn't room for anything cheaper.There is still room below the top bins for cheaper 16c and 12c SKUs.
The scaling in Cinebench R15 from R7 1700X to TR 1950X is almost 100%. Amazing. TR 1950X rips 7900x for serious rendering work. POVRay should end up 15% faster, Blender around 20% and Corona Photorealism around 25% faster on 1950x vs 7900x if scaling is similar to Cinebench.
http://www.anandtech.com/show/11170...review-a-deep-dive-on-1800x-1700x-and-1700/18
http://www.anandtech.com/show/11550...core-i9-7900x-i7-7820x-and-i7-7800x-tested/11