Go Back   OC3D Forums > [OC3D] General Forums > OC3D News
Reply
 
Thread Tools Display Modes
 
  #1  
Old 24-05-19, 11:42 AM
WYP's Avatar
WYP WYP is offline
News Guru
 
Join Date: Dec 2010
Location: Northern Ireland
Posts: 17,725
Ryzen 3rd Generation 16-core rumoured to be a Threadripper-killer with insane Cineben

That many points on a single desktop-grade CPU?



Read more about AMD's rumoured Ryzen 3rd Generation 16-core Cinebench performance.

__________________
_______________________________
Twitter - @WYP_PC
Reply With Quote
  #2  
Old 24-05-19, 12:04 PM
AlienALX's Avatar
AlienALX AlienALX is offline
OC3D Elite
 
Join Date: Mar 2015
Location: West Sussex
Posts: 14,126
LOL look at the CPUs it's on the same list as. That's nuts.

That's like the who's who of CPUs.
__________________

Reply With Quote
  #3  
Old 24-05-19, 01:03 PM
Avet's Avatar
Avet Avet is offline
OC3D Elite
 
Join Date: Dec 2016
Posts: 1,411
Quote:
Originally Posted by AlienALX View Post
LOL look at the CPUs it's on the same list as. That's nuts.

That's like the who's who of CPUs.
Yes. But... That is only in rendering types of workload. For everything else Intel is better.

This is interesting because it shifts traditional concept of massive single workstation PC that you use for everything. Wendell (Level1), and Eposvox did a lot of testing, and recommend one PC for editing based on 9900K, or 9700K because it is snappy, has fluent workflow etc... And another based on Threadripper that you use only as a rendering station. This way you get the best from both worlds, and you don't need the fastest CPU on the planet because you loose editing time wile it renders. This gives you much more room for scaling your systems.

This is where dual system cases like Corsair 1000D come in. One Intel Z390 build for editing, general use, and a Threadripper monster just for rendering. Any content creator's wet dream system.

I think this is the way reviewers should approach this "new generation" of CPUs.
Reply With Quote
  #4  
Old 24-05-19, 01:05 PM
tgrech tgrech is online now
OC3D Elite
 
Join Date: Jun 2013
Location: UK
Posts: 1,855
Quote:
Originally Posted by Avet View Post
Yes. But... That is only in rendering types of workload. For everything else Intel is better.

This is interesting because it shifts traditional concept of massive single workstation PC that you use for everything. Wendell (Level1), and Eposvox did a lot of testing, and recommend one PC for editing based on 9900K, or 9700K because it is snappy, has fluent workflow etc... And another based on Threadripper that you use only as a rendering station. This way you get the best from both worlds, and you don't need the fastest CPU on the planet because you loose editing time wile it renders. This gives you much more room for scaling your systems.

This is where dual system cases like Corsair 1000D come in. One Intel Z390 build for editing, general use, and a Threadripper monster just for rendering. Any content creator's wet dream system.

I think this is the way reviewers should approach this "new generation" of CPUs.
Dual-mobo systems would likely become kinda redundant if these numbers are true(Since this would mean Threadripper has performance that matches Intel even in low core workloads). Essentially these are showing AMD matching Intel in IPC(Roughly same per core performance as the 9900K) in a benchmark that generally favours Intel. This would probably mean AMD surpassing Intel in a wider benchmark suite if true, unless the nature of Zen and its strong points has changed significantly.
Reply With Quote
  #5  
Old 24-05-19, 01:34 PM
MiNo MiNo is offline
Member
 
Join Date: Jul 2013
Location: Denmark
Posts: 164
Ooh, look at those cinebench scores! Said no gamer ever.


JK, awesome performance - but COME ON with some top shelf gaming stuff already. Please. Don't leave us alone with Intel and Nvidia.
Reply With Quote
  #6  
Old 24-05-19, 01:59 PM
Avet's Avatar
Avet Avet is offline
OC3D Elite
 
Join Date: Dec 2016
Posts: 1,411
Quote:
Originally Posted by tgrech View Post
Dual-mobo systems would likely become kinda redundant if these numbers are true(Since this would mean Threadripper has performance that matches Intel even in low core workloads). Essentially these are showing AMD matching Intel in IPC(Roughly same per core performance as the 9900K) in a benchmark that generally favours Intel. This would probably mean AMD surpassing Intel in a wider benchmark suite if true, unless the nature of Zen and its strong points has changed significantly.
The biggest limitation on Ryzen CPUs is their architecture. First of them frequency, and second latency. Even if you match the frequency Infinity fabric will still be way slower than Ring Bus, or Mesh Interconnect. And i don't see chiplet design reducing the latency. It may be the opposite. AMD compensates for this by offering more cores for lower price. If you have task that can be perfectly multithreaded (rendering) more cores beat frequency, and latency in most cases. Apart from rendering, pretty much all other tasks can't be spread perfectly to all cores. For anything that requires tasks to be competed in sequence Intel architecture wins. That is why Gaming, scrubbing in Premiere, Music production work much better on Intel. In this tasks with Threadripper, for example, you have bunch of cores waiting on each other, and for the most of the time doing nothing.

That is why two mid to high range systems which are tailored for tasks that they do best (editing Z390, rendering X399) can give you much better performance than a massive Intel W-3175X system that does both tasks really good.
Reply With Quote
  #7  
Old 24-05-19, 02:19 PM
tgrech tgrech is online now
OC3D Elite
 
Join Date: Jun 2013
Location: UK
Posts: 1,855
Nothing to indicate Zen2 will (necessarily) have these problems imo, the main issues of latency were caused by the fact inter-CCX communication had to go via the memory controller to access the L3 cache of another CCX, which was far weaker and higher latency than even Bulldozer's, the chiplet design makes it incredibly likely that inter-CCX communication within the same chiplet has paths for direct communication without having to go via the memory controller(Because that's now off chip which incurs a power efficiency penalty), so upto 8 cores in use I think there's a good chance latency has at least dropped to Bulldozer levels, which puts it within nanoseconds of Intels latest. Improvements to cache size and prediction rate would also make this significantly less of an issue. Not using the L3 in a victim cache configuration could pay dividends if they've improved the branch predictor.

Secondly of course, even with inter-chiplet communication, the memory controller is now far stronger, and it's likely infinity fabric will be operating at much higher clock speeds, and latency in this case is obviously the direct reciprocal of clock speed. The latency induced by the longer paths would be essentially insignificant, as we're talking about data travelling at ~1/3rd the speed of light over centimetres, issues that arise are from the signalling changes between the interposer, but they've been using interposers for extremely high bandwidth connections since Fury now so I'm sure that's an area they've been attempting to progress in.

[Edit to add] In terms of frequency, that's much more of a transistor node issue than an architectural one nowadays, you can slice up the pipeline stages fairly liberally with modern microprocessor designs to match your nodes target frequencies, that's the only major change required architecturally to make a processor work in frequency bands higher than intended(And all the work to rebalance those pipelines to actually gain a perf improvement from it but that's heavily automated nowadays with modern HDL techniques).
Reply With Quote
  #8  
Old 24-05-19, 07:25 PM
Gothmoth Gothmoth is offline
Advanced Member
 
Join Date: Jun 2017
Location: Germany
Posts: 280
who is interested in gaming scores of cpu´s these days?
the plebs who still game at 1080p?



at 4K it doenst matter if you have an intel or amd.
Reply With Quote
  #9  
Old 25-05-19, 01:19 AM
Emu Emu is offline
Member
 
Join Date: Jan 2015
Location: Australia
Posts: 125
People seem to be missing the point here. If this score is correct then AMD has made some huge improvements to the Zen architecture. A 9900k overclocked to 5GHz on all cores scores 2166 in Cinebench, the 16 core/32 thread Ryzen 3rd gen which is apparently running at 4.2GHz on all cores scored just under twice the score.

In other words, if this rumour is true then Zen2 has the same performance at 4.2GHz per core as a 9th Gen Core CPU running at 5GHz (assuming that Cinebench has a linear scoring system). To be quite honest, as nice as this would be I doubt that this is the case given all the rumours that we have heard up till now. Performance parity with Intel's 9th gen alone would be a huge win for AMD, especially if they can do that with twice the cores and half the power draw.
Reply With Quote
  #10  
Old 25-05-19, 03:50 AM
NeverBackDown NeverBackDown is offline
AMD Enthusiast
 
Join Date: Dec 2012
Posts: 17,052
Will also take into account all these vulnerabilities from Intel means their CPUs keep taking performance hits. It's helping AMD.
Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump










All times are GMT. The time now is 10:08 AM.
Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2020, vBulletin Solutions, Inc.