Speculation: Ryzen 3000 series

Page 106 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

What will Ryzen 3000 for AM4 look like?


  • Total voters
    230

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Yeah I'm looking at the 12 core also. Really depends on how well the infinity fabric holds up though. That's the key.

In some cases an 8c Zen2 might be faster due to intercore latency. So yes, IF latency is going to be something to watch.

Interesting, looks like Wikipedia is wrong. I had to read a bit to find out why. DDR4 still uses an 8n prefetch but it can use groups of banks (2 or 4) - so that’s how the IO rate is doubled with the same base clocks on the sdram chips.

Yup, them sneaky JEDEC buggers have been upping throughput without actually increasing RAM clocks by all that much. Sort of galling, really, when you think of the memory performance we could have if they would only let memory clocks go higher than that.


Seagate Momentus? Wow, really? Couldn't someone at least have hooked up a SATA SSD or something? No love for that ES machine. Interesting results otherwise. My 1800x with DDR4-3200 14-14-14-28 got 70.1ns latency in AIDA64. I suppose if I had relaxed timings a bit, I'd be close to that Zen2.
 
Last edited:

PotatoWithEarsOnSide

Senior member
Feb 23, 2017
664
701
106
Interesting that turbo is listed as less than base. :p

Does anyone recall the latency graphs for the earlier ESs that were benched? Pretty sure I recall much higher latencies.
 
  • Like
Reactions: Olikan

dnavas

Senior member
Feb 25, 2017
355
190
116
So it's running at 3200MHz out of the box. Latency is 80ns for this early sample, not too bad, in my opinion.

Well, but, it's not fabulous either. 60ns would have been nicer to see this gen considering it's up against the 9900k.

ed: FWIW, I just downloaded and ran these tests locally. Obviously I don't have a gaming rig :> https://www.userbenchmark.com/UserRun/16928302
Anyway <90ns to memory and 10ns L3.

[ed2: oh, 15ns is L3, not inter-die w/ corresponding edits and implied head-slaps]

I thought the Zen2 was supposed to double L3 to 32MB? So, is what I'm seeing in the original link higher L3 latency and terrible inter-die latencies? As in, it's practically cheaper to go back to RAM latency?
 
Last edited:

amrnuke

Golden Member
Apr 24, 2019
1,181
1,772
136
Well, but, it's not fabulous either. 60ns would have been nicer to see this gen considering it's up against the 9900k.

ed: FWIW, I just downloaded and ran these tests locally. Obviously I don't have a gaming rig :> https://www.userbenchmark.com/UserRun/16928302
Anyway <90ns to memory and 10ns L3.

[ed2: oh, 15ns is L3, not inter-die w/ corresponding edits and implied head-slaps]

I thought the Zen2 was supposed to double L3 to 32MB? So, is what I'm seeing in the original link higher L3 latency and terrible inter-die latencies? As in, it's practically cheaper to go back to RAM latency?

These are engineering samples, and it had a boost clock lower than base clock. I think it's a little early to be drawing conclusions.

But... if it remains that way, AMD has a huge problem, IMO.
 

dnavas

Senior member
Feb 25, 2017
355
190
116
These are engineering samples, and it had a boost clock lower than base clock. I think it's a little early to be drawing conclusions.

But... if it remains that way, AMD has a huge problem, IMO.

Well, my system runs just fine, and if I can get a 48 core that runs at slightly higher speeds to my 16 core, I'm going to be pretty doggone happy, so it's a long way away from "huge problem". But 13ns (or whatever it was actually) L3 is a definite regression and could cause problems. It's an area to keep an eye on which may represent an ongoing source for potential future improvement :>
 

Tuna-Fish

Golden Member
Mar 4, 2011
1,324
1,462
136
I thought the Zen2 was supposed to double L3 to 32MB? So, is what I'm seeing in the original link higher L3 latency and terrible inter-die latencies? As in, it's practically cheaper to go back to RAM latency?

L2 in Zen1 is 8MB per CCX. Zen2 doubles that to 16MB, and with 2 CCX per chip, that means 32MB per chip, or 64MB for any 2-chip Zen2. The usable cache for a single thread is only the cache of the local CCX, as with previous Zens.
 
  • Like
Reactions: dnavas

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
If Zen2 can run DDR4-5000 with "reasonable" timings, I would expect much lower latency. The potential for tuning will be there.
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,744
3,077
136
In some cases an 8c Zen2 might be faster due to intercore latency. So yes, IF latency is going to be something to watch.
Probably not going to be much worse than it already is, so much of the latency is taken up in the actual coherency, thats one of the big changes for Rome only one big home agent taking care of all of this not 4 small ones.

As in, it's practically cheaper to go back to RAM latency?
Thats how cache coherence protocols work for the most part, its far safer to flush back to memory and then reread to ensure you have the correct data.

AMD using some version of MOESI with some extra states ( they haven't really detailed from what i know)
https://en.wikipedia.org/wiki/MOESI_protocol
While MOESI can quickly share dirty cache lines from cache, it cannot quickly share clean lines from cache. If a cache line is clean with respect to memory and in the shared state, then any snoop request to that cache line will be filled from memory, rather than a cache.

If a processor wishes to write to an Owned cache line, it must notify the other processors that are sharing that cache line. Depending on the implementation it may simply tell them to invalidate their copies (moving its own copy to the Modified state), or it may tell them to update their copies with the new contents (leaving its own copy in the Owned state).
 
  • Like
Reactions: dnavas

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
No, it's 3200 MHz but according to Looncraz it might have a really high CAS latency of CL22 or CL24. If that's true the actal latency would be between zen and zen+
Yup, the other reddit poster did a great job with finding the exact spec data for the new Micron memory: its CL22/24 for operating at 3200MT/s.

If Zen2 can run DDR4-5000 with "reasonable" timings, I would expect much lower latency. The potential for tuning will be there.
I would stay a bit more conservative about that, be confident about DDR4-4000 and hope for incremental improvements towards something like 4200-4400.
 
Last edited:

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Well, but, it's not fabulous either. 60ns would have been nicer to see this gen considering it's up against the 9900k.

ed: FWIW, I just downloaded and ran these tests locally. Obviously I don't have a gaming rig :> https://www.userbenchmark.com/UserRun/16928302
Anyway <90ns to memory and 10ns L3.

[ed2: oh, 15ns is L3, not inter-die w/ corresponding edits and implied head-slaps]

I thought the Zen2 was supposed to double L3 to 32MB? So, is what I'm seeing in the original link higher L3 latency and terrible inter-die latencies? As in, it's practically cheaper to go back to RAM latency?

Didn't AMD say the latencies are only better on x570? And actually worse on x470? At least that is what I remember.

EDIT: I mentioned it because I got confused with the links and thought they ran the ES on an older board but that was actually your link. In the one of the ES the chipset is actually unknown.
 

PotatoWithEarsOnSide

Senior member
Feb 23, 2017
664
701
106
Slightly off this on a tangent.....that Reddit look link shows an image of the 12c ES, purportedly. What strikes me as odd is that it doesn't have the same layout as the single chiplet CPU that Lisa Su held up at CES; here chilets are either side of the IO die, but at CES the chiplet was to the right and with a space for a second beneath it. Not only that, the dimensions seem to be different.
Am I missing something here?
 

zinfamous

No Lifer
Jul 12, 2006
110,512
29,099
146
Looks to me the x570 is hot as hell.

Surely it could be effectively cooled with a very robust heatsink and solid case cooling, right? seems to me if you have a front-mounted radiator and proper rear exhaust, there should be enough cool air passing through and over the chipset? Or just top-mount the radiator to isolate CPU/GPU heat, and slap 2 very large case fans on the front?
 

realibrad

Lifer
Oct 18, 2013
12,337
898
126
Just wondering your opinions, I understand the fan is another thing that could fail and another thing that generates noise but isn’t more cooling what’s needed on the x470’s?

My worry is the fan failing. I'm sure its not going to be very loud, but I would bet that they configure the board in a way that a fan replacement would not be easy. I would rather use my casefans in a well cooled case to blow over a heat sink.