Possible 12-Core 24-Thread Third-Gen Ryzen 'Matisse' CPU Pops up in UserBenchmark Database

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
858
315
19,360


The clocks are listed in the 2D3212BGMCWH2_37/34_N product identifier. I clarified that in the article, thanks for the heads up.
 

PapaCrazy

Distinguished
Dec 28, 2011
311
95
18,890
Rumored core counts are beginning to make sense. The yields on the 8 core chiplets must be too good to even mess with quad cores. So 1 'bad' chiplet=a 6 core 3300. 1 'good' chiplet=8 core 3600. 1 good+1bad (or even 2 bad)=12 core 3700. 2 good chiplets=16 core 3800. Incredibly efficient method of packaging chips if true.
 

rantoc

Distinguished
Dec 17, 2009
1,859
1
19,780
AMD really seems to have a gem upcoming, ipc thats on pair or better than intel and finally clock rates that's comparable along with more cores and clever core packaging = Better use of the silicon and that also use to turn down to better consumer prices...

While on the subject of prices... intel's latest stunt with non-igpu chips (likely failed gpu's just cut out) for the same price is just greed, faulty chips for the same price - Just bloody no. Each day intel works hard to make me avoid them more and more!
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995
"The UserBenchmark System Memory Latency Ladder quantifies the latency of L1, L2, and L3 caches of the test sample, and the decline at 32MB indicates the engineering sample comes armed with 32MB of L3 cache."

Unless I'm missing something, if latency declines at 32MB, i would say L3 is smaller than 32MB. If L3 was exactly 32MB there would be no decline.
 

s1mon7

Reputable
Oct 3, 2018
96
4
4,635
I'm not sure if you guys caught it, but a 13% IPC boost over Zen+ would top Intel's Coffee Lake IPC by a healthy margin. That would be just incredible for a gen-over-gen boost. The core count increase is the icing on the cake.
 
As a staunch defender of the leak due to just how reasonable it seems... I'm feeling pretty pleased. The Ryzen 3000 series is going to be a heck of a product stack, even if I'm wrong and this is the top core configuration... but as I've said many times, "If AMD can, why wouldn't they?". So I very much expect to see a 16 core sku with a 3000 series designation. Will it happen at launch? I dunno, but I expect that it will happen.

Also, this is really our first hard look at IPC gains... and 13% is great, but in context looks freaking outstanding. AMD was within 5% of the Skylake based chips with Zen+. This puts them over Intel in IPC. More amusingly, this is done with a modest DDR4 2666 memory kit... so... either RAM speed doesn't matter as much anymore (which is wonderful), OR we could extract even more performance with a DDR4 3200 kit (which is more wonderfuller)... and if it behaves like Zen and Zen+ that could be a significant performance boost.

So, this means that people like me, with Haswell to Kabylake i5's and i7's will FINALLY have something WORTH upgrading to at a reasonable price. I mean, if a Ryzen 5 3600 is going to be in the same performance range as an i9 9900K (remember AMD's demo, the 8c/16t chip would be a 5 if the 12c/24t is a 7) and cost less than my old 4590 did when it was new... heck yeah! I'm upgrading, and I suspect a lot of other people will be as well.

Clear the lines, the hype train is coming through.
 

salgado18

Distinguished
Feb 12, 2007
928
373
19,370


It could have 16MB, 8MB per chip. Or 24MB, 16MB per chip, with one cut in half.
 
Jan 24, 2019
1
0
10
Still going to wait for the full release numbers. This engineering CPU is low on all the Mixed scores...

Mix 8700k 2700x 12c/24t Zen2

SC 138 120 104
QC 525 468 359
MC 1078 1311 1483

For me it doesn't have to beat intel, just tie them or close that gap on the SC and QC and these chips will be the winner. I am not a fanboy of either brand I use for games so whatever hits my budget for performance is where I go.
 

InvalidError

Titan
Moderator

The engineering sample at CES may not be representative of Ryzen 3's final form(s), it is entirely possible that the chip used in the demos was picked for best possible results. Actual yields may make the lower-end 8C16T part 4+4 instead of 8+0, so we'd end up with the 3600 being 4+4 and 3600G/X being 8+IGP/8+0. After all, AMD needs at least one SKU to use chiplets with some defective L3$ on and this is the one that makes the most sense to me.
 
@DGPROPHET
Still going to wait for the full release numbers. This engineering CPU is low on all the Mixed scores...

Mix 8700k 2700x 12c/24t Zen2

SC 138 120 104
QC 525 468 359
MC 1078 1311 1483

You have quoted the INT scores instead of Mixed Scores.
Mixed are: SC - 115 QC - 373 MC - 1863
 

salgado18

Distinguished
Feb 12, 2007
928
373
19,370

Also should account for differences in clocks. The engineering sample is supposedly at 3.6GHz, and probably without SMT enabled (QC should scale above 4 times the SC scores).
 
since zen+ is already within (+)(-)3% of intel's current gen IPC numbers (just much lower clocked) this would be a pretty big deal if AMD increased IPC by 13%. Just a little scribbling here tells me that a similar core'd intel CPU running at 5GHZ would be matched/bested by an AMD at 4.6GHZ if that's true, and apparently AMD is claiming atleast one of their cpus will hit 5ghz out of the box. If that's the case then the new performance king will be AMD by a comfortable margin.
 

hannibal

Distinguished
Most likely best 8 core 16 threath Are gonna be Ryzen 3700x or 3800x and 12 core and 16 core comes above them. My prediction is that 8/16 will be 3700 12/24 3800 and 16/32 3900... or we get classic 3800 x2 for 16/32 and 3600 X2 for 12/24. So the core count from 2600 to 3600 will remain the same and so on. And those multi chip version will become new flagship models above old ”singlel chip versions.
Lets see if we get 3600 X2 or what ever it will be called at the same time as normal 3600 version or just later at the end of the year?

In anyway this is very interesting year for CPU... not so interesting considering GPUs... I am much more interesting 2020 considering GPUs.
7+nm versions of Nvidia RTX cards and highend next gen version Of Navi architecture that should Also become 2020.
 

InvalidError

Titan
Moderator

8C16T has been the x700 model for two generations already and were competing against Intel's 4C4T/4C8T at the time. Now that Intel has stepped its mainstream up to 6C6T and 8C8T, the x600 and below need some sort of promotion to maintain a compelling value position, especially if AMD remains at a ~10% clock disadvantage.

I'll be disappointed if the 3600 turns out 6C12T, probably means I'll be holding on to my i5-3470 for at least another year..
 


What do you mean? Yea, Faildozer was like 50% behind in IPC, but Ryzen has already had 1 generation and a refresh out for 2 years now. It's well known, tested and document how their performance compares. Clock for clock, it's 5-15% behind depending on the application. It's really doubtful they'll go backwards in IPC, so it'll for sure close the gap with intel, or get close.
 

joeblowsmynose

Distinguished


I actually don't recall AMD ever claiming that anywhere ... but there have been a few enthusiasts who have been circulating that rumour.

If 13% IPC pans out, I'll be more than happy with a 4.6-4.8 boost / full time overclock for my next upgrade - granted that it doesn't suck 250w through the socket under load like a 9900k ... which at 7nm it better not.
 

joeblowsmynose

Distinguished


Have you not noticed Ryzen yet, or the freshly renewed AMD in the last 2.5 years?

Or are you referring to the hype train being loaded up so much with the hype contained on it too heavy for delivery? That would be potentially valid.

I think its best to watch AMD and see what they claim, as opposed to rumours. AMD under Lisa seems to be taking the "under promise and over deliver approach" so I think in that light, watching what AMD says and does not say is anyone's best bet at ensuring the hype train has enough horsepower to reach its destination ...

So far things are looking pretty good though ... I expressed strong doubts and the initial AdoredTVs claims leak, but I have been pleasantly able to appease some of those doubts an some of AMDs actual announcements.
 

InvalidError

Titan
Moderator

IIRC, the 13% IPC improvement figure came out of a workload that specifically hammered Zen 2's improvements. For more normal workloads, it is more likely to be somewhere in the 5-10% range.
 

1_rick

Distinguished
Mar 7, 2014
104
44
18,620


Yeah, it would be awful to only get 2 more cores and 6 more threads, all of which will probably be faster than your current speed.