• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

The Ryzen "ThreadRipper"... 16 cores of awesome

If it actually does turn out to be an MCM design, it'll be interesting to see if AMD has managed to work around the NUMA-related pitfalls that helped doom Quad FX a decade ago. And it'll also be interesting to see if Intel suddenly produces a 14C or 16C Core i9 to keep AMD's hands away from the performance crown.
 
It looks like AMD officially announced the ThreadRipper CPU today:

https://www.engadget.com/2017/05/17/amd-s-16-core-threadripper-cpu-is-built-for-ultra-high-end-pcs/

Well... that kinda steals the thunder of Intel's Core i9 processor announcement. We're supposed to get more technicial details at Computex next week, but I wonder when we'll see pricing information on this bad boy? AMD says "Summer", but that could be anywhere from June until September.

I like the name of this CPU. Trivial but it is far more badassery than Epyc. Now data centers will be ran epically on the epyc platform for epic performance...
 
If it actually does turn out to be an MCM design, it'll be interesting to see if AMD has managed to work around the NUMA-related pitfalls that helped doom Quad FX a decade ago. And it'll also be interesting to see if Intel suddenly produces a 14C or 16C Core i9 to keep AMD's hands away from the performance crown.

Ryzen5 and Ryzen7 already need to work around NUMA-related pitfalls.

Given the information presented to date - synching across the module is no different from synching across the CCX.
 
Will be interesting to see how this all plays out. From my point of view, I am really judging these larger core count chips by how well they can keep up in single threaded applications. It is a given they will do well in multi threaded situations but who wants a slow CPU for the times you are not using all of the cores? I hope AMD has put some thought into this aspect as Intel seems to have shown some promise in this area.
 
1800X dropped $40 since launch on Amazon, so maybe $500-550 for the entry level Threadripper CPU? I'm guessing they'll sell 10 -16 core versions.
 
Ryzen5 and Ryzen7 already need to work around NUMA-related pitfalls.

Given the information presented to date - synching across the module is no different from synching across the CCX.

Well, the issue with Quad FX was that you had two memory pools connected to different CPUs, and Windows XP had absolutely no idea how to keep track of what data was located in which memory pool. As a result, all memory accesses had to go over the HT bus to both processors, driving memory latency way up and killing performance. It'd be presumably the same scenario in any system which uses multiple Zen dies connected internally by HT, assuming all four of the chip's memory channels are populated.

It might not be an issue now, with AMD having more experience with NUMA designs and Windows 10 being a lot more adept at handling that kind of situation, but then the only other MCM chips AMD has produced to date are the Socket G34 Opterons, and memory latency was high on those anyway because they used external memory buffers, so it's difficult to get an exact idea.
 
Will be interesting to see how this all plays out. From my point of view, I am really judging these larger core count chips by how well they can keep up in single threaded applications. It is a given they will do well in multi threaded situations but who wants a slow CPU for the times you are not using all of the cores? I hope AMD has put some thought into this aspect as Intel seems to have shown some promise in this area.

I think for a lot of users that are targeted with this platform it won't be an issue. Latency will be a possible issue but not so much single-core performance. The reason is that if you're a content creator you get one of these primarily for the purpose of doing business. So if you're color grading film/tv content, or rendering graphic effects for example, then that's how you'll spend most of the time with this computer, not doing things that rely on fewer cores running faster (like gaming etc). So even if it's a "bottleneck" it'll be one that the users buying this platform mostly won't run into.

I do think latency might still be an issue however. In pro-audio for example low-latency performance is key, and if it hampers what you get out of the platform then clearly the platform isn't as valuable as it seems.
 
Threadripper isn't using the same socket as EPYC (Naples) is (SP3 vs. SP3+).
SP3 package uses four dies in MCM config, while SP3+ uses "just" two.

Makes no sense to use a huge package with eight channel memory & 128 PCIe link capability (hence 4094 LGA contacts), when the silicon in this case only has and uses half the amount.

Most likely SP3+ is somewhat similar to C32 vs. G34 socket situation seen in the past.

SP3+ could also be the same thing as SP4 (Snowy Owl), but equipped with LGA contacts instead of BGA balls.
 
I do think latency might still be an issue however. In pro-audio for example low-latency performance is key, and if it hampers what you get out of the platform then clearly the platform isn't as valuable as it seems.
ASIO latency will be just fine. Meanwhile, you can do tons more at the same time.
 
I agree but it is software dependent. I do video, photography and audio and find a mix between single and multi threaded situations. I wish everything was heavily threaded but it is not. So to get me to pay up for the higher core counts the CPU at least needs to be close to the lower core chips in single thread to justify the higher price. But I am a solo operator, not in a company where things might be more compartmentalized. My needs are for an all around performer. I like how Intel has implemented the per core overclocking to try to make up this gap. I hope they expand on this approach.
 
Actually $799 seems more realistic now.

I don't think $799 is ultra-premium enough. Maybe $1000+.


1800X dropped $40 since launch on Amazon, so maybe $500-550 for the entry level Threadripper CPU? I'm guessing they'll sell 10 -16 core versions.

$500-550?! Stop joking around. There's only 12 & 16-core ThreadRipper. Everything else you think you know are faked leakes.

LhTztdO.jpg
 
Have you guys seen the size of this thing?

http://www.guru3d.com/news-story/am...essors-16-cores-and-32-threads-confirmed.html

It's as big as the palm of your hand! I know they originally designed this with workstations in mind but damn, that thing is huge 😛

If AMD can deliver this at a reasonable price point I may be tempted to pick it up. But as I only do a moderate amount of video/photo editing I would have to see a quite significant bump in performance at a not break the bank price to move up from my R7 1700 build.
 
ASIO latency will be just fine. Meanwhile, you can do tons more at the same time.

I saw parts of that video before, and would have to watch it again. But the previous tests I saw using a standardized test showed that it didn't do as well with lower latencies using virtual instruments compared to Intel. I'm not saying the value proposition isn't there, because for people like me who can live with a higher latency it's a non-issue. I'm saying that for some it matters. But sure, it's a subset of a subset essentially.
 
Where did AMD state that they're only releasing a 12 and a 16 core version?

Stilt confirmed that the cores+cache combination on each CCX have to be identical on both die. So only really 12 and 16 make sense. You could do 8 as well in theory.

I don't think $799 is ultra-premium enough. Maybe $1000+.

I thought it was going to be $999, but Glo made the comment about disruptive pricing. Keep in mind that AMD is going to have to slash Ryzen prices anyway once Skylake-X arrives so it does sort of make sense.
 
Back
Top