Speculation: Ryzen 3000 series

Page 112 - 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

JDG1980

Golden Member
Jul 18, 2013
1,663
570
136
competition does wonders...I think we will see Intel 7nm product in 2021, I predict some big bang conroe jump

I doubt it. You only get a Conroe/Ryzen type jump when the previous architecture (Netburst/Bulldozer) sucks. Even Sandy Bridge was only 10-15% IPC over Nehalem; the rest of the gains were due to clock speed increases enabled by the 32nm process. Those kind of easy process gains are a thing of the past now.

what worries me about ryzen 3k is now the stupid fan on x570....

i will never buy a board with fan until I absolutely must

Allegedly, the new chipset has a TDP of 15W. If that's the case, then it shouldn't be that hard to cool with a good heatpipe-enabled passive sink. Hopefully premium boards will do this instead of resorting to cheap 40mm fans. If not, then aftermarket fanless coolers will probably show up to fill the gap.
 

TheGiant

Senior member
Jun 12, 2017
748
353
106
I doubt it. You only get a Conroe/Ryzen type jump when the previous architecture (Netburst/Bulldozer) sucks. Even Sandy Bridge was only 10-15% IPC over Nehalem; the rest of the gains were due to clock speed increases enabled by the 32nm process. Those kind of easy process gains are a thing of the past now.



Allegedly, the new chipset has a TDP of 15W. If that's the case, then it shouldn't be that hard to cool with a good heatpipe-enabled passive sink. Hopefully premium boards will do this instead of resorting to cheap 40mm fans. If not, then aftermarket fanless coolers will probably show up to fill the gap.
no sorry, aftermarket cooler for chipset...no thanks
not buying 200 EUR+ board to handle some redacted 4cm vrrrrrr fan....I hope they don't mess things that way

Again with the profanity (in the same thread no less)?
Profanity is NOT allowed in the tech forums.


AT Mod Usandthem
 
Last edited by a moderator:

Abwx

Lifer
Apr 2, 2011
10,948
3,458
136
Could also be a lower end product, maybe a replacement for the 7281 (16 cores, 2.1 base/2.7 turbo, $650 "MSRP"). Take 4 of the lowest quality binned chiplets.

............

Quote from the link :

It seems obvious that the Dell PowerEdge R6515 mentioned in the record is a successor to the PowerEdge R6415 Rack Server that features 32-core AMD Epyc processors. It’s a step up, both for AMD and Dell.

SiSoftware posts a healthy score of 1544.41 Mpix/s for this AMD Rome/Dell PowerEdge combination. There has been some speculation that Dell could move toward an AMD-only server generation, which would not be entirely surprising considering the speed efficiency results of the 32-core 64-thread engineering sample in this case.


https://www.notebookcheck.net/SiSof...th-a-32-core-Rome-chip-from-AMD.420840.0.html
 
  • Like
Reactions: DarthKyrie

Abwx

Lifer
Apr 2, 2011
10,948
3,458
136
EPYC Rome 32c VS EPYC 7551 in SiSoftware Processor Multi-Media
I believe the speed is misreported and the the Rome proc is most likely OCd @ 3.36 ghz.
IPC is ~+15% Int and +10% Float
View attachment 6469

IPC numbers can hardly be extracted as the sample has 256bit exe units for AVX2, from the Quad-int throughput that does not seem to take advantage of the ISA it s likely that the chip is running at 2.4GHz and the 40% OC is in respect of the 1.7 base frequency, at 3.36 power would be increased 6-8x....

As for the real improvement methink that it s rather 15% in FP and eventualy 10% in Int, the latter is much more difficult to improve when new instructions are not used.
 

Asterox

Golden Member
May 15, 2012
1,026
1,775
136
............

Quote from the link :

It seems obvious that the Dell PowerEdge R6515 mentioned in the record is a successor to the PowerEdge R6415 Rack Server that features 32-core AMD Epyc processors. It’s a step up, both for AMD and Dell.

SiSoftware posts a healthy score of 1544.41 Mpix/s for this AMD Rome/Dell PowerEdge combination. There has been some speculation that Dell could move toward an AMD-only server generation, which would not be entirely surprising considering the speed efficiency results of the 32-core 64-thread engineering sample in this case.


https://www.notebookcheck.net/SiSof...th-a-32-core-Rome-chip-from-AMD.420840.0.html

Engineering sample or not, we already know the basic details for 64 core Epyc Rome server CPU.2.3ghz is stock clock, or expected when you show at least the basic CPU details.

https://www.anandtech.com/show/13598/amd-64-core-rome-deployment-hlrs-hawk-at-235-ghz
 
  • Like
Reactions: lightmanek

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
all things being relative, anyone's guess why quad int results so low, where everything else is so high?

Processors have native capability for 32bit / 64bit integers and 128 bit sized integer operations are usually done with a specialized codes ( think BigInteger classes in java or C#, but optimized for operations and operand sizes as required ). They are hard operations to vectorize, but can store temporaries and results in vector registers. So instead of processing 8 floats or 4 doubles at a time in 256bit AVX register, you are chewing over code that usually has a nasty dependency chain and high register pressure. Slowdown of two orders of magnitude is no big deal here compared to native sized ops.

Why INT results regressed - probably combination of lower clock speed and larger memory latency, ZEN2 does not improve that much in area of INT processing.

Without knowing what Quad-Float does, it is hard to know why it does not suffer the same as Quad-Int, but it benefits from ZEN2 improvements.
 

B-Riz

Golden Member
Feb 15, 2011
1,482
612
136

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136

According to Adored's Info, 16C CPU@4.2 GHz scored 4278 pts in Cinebench R15. Overall IPC of Zen 2 chips appears to be 12.5% higher than Zen+.
In CB r15 that is. Not what I would call ipc but anyways.

FMax rumours at at least 5GHz is sweet but I bet for 4.7. Perhaps we need a bet here as we had with zen :)?
 

Abwx

Lifer
Apr 2, 2011
10,948
3,458
136
On CB R15 a 4.2GHz 8C/16T Zen 2 will be on par with a 5GHz clocked 9900K at 1/3 of the TDP...

Improvement over Zen + is 15% or so, sure that R15 is FP but there s no AVX/FMA and it s somewhat representative of legacy SSE2-4.2 apps performances, possibly that there wll be a nice improvement in pure Integer like 7 Zip, partly due to bigger caches than Zen + that hide the MC latencies.
 
  • Like
Reactions: lightmanek

Asterox

Golden Member
May 15, 2012
1,026
1,775
136
In CB r15 that is. Not what I would call ipc but anyways.

FMax rumours at at least 5GHz is sweet but I bet for 4.7. Perhaps we need a bet here as we had with zen :)?

Yes it is only "air IPC", or tell that to any "Maxon Cinema4D" user who paid 3000euro for software licence.;)

 
  • Like
Reactions: lightmanek