Speculation: Ryzen 3000 series

Page 192 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

BigDaveX

Senior member
Jun 12, 2014
440
216
116
Assuming these leaks are legit, I'm guessing that AMD focused on giving the on-chip caches insanely high bandwidth in order to offset the losses from separating the memory controller and CPU die. And assuming the other leaks are legit, it looks like they succeeded spectacularly.
 

Carfax83

Diamond Member
Nov 1, 2010
6,841
1,536
136
The AIDA64 cache and memory benchmark scale with the amount of cores, so it's best to compare it to an Intel 7920x. Where the 3900x really crushes the 7920x is in L3 cache bandwidth it seems, and I'm sure that its performance plays a major role in Zen 2's gaming prowess.



Skylake-E's L3 cache is considerably slower than Broadwell-E's, and is a big reason why Skylake-E's gaming performance was impacted:

 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
Assuming these leaks are legit, I'm guessing that AMD focused on giving the on-chip caches insanely high bandwidth in order to offset the losses from separating the memory controller and CPU die. And assuming the other leaks are legit, it looks like they succeeded spectacularly.

1) L1 bandwith numbers look "Intel" like, 50+% more total L1 bandwith Is expected as it has 50% more cores, latency is in same ballpark, 4.4Ghz clocked 4 cycle cache ~0.9ns latency, needs 5Ghz to show 0.8ns as on Skylake.
2) L2 bandwidth looks like it has minor improvements, previuos gen had ~100GB/s per core, @4ghz, this one seems to have pushed that number to 120GB/s or so
3) L3 is almost the same as previuos generation, very comparable to 2 chip threadrippers like 1950x

So just as AMD disclosed, we are getting some epic L1 cache improvements, but the rest were not touched that much.

P.S. and Latency would be nice to see
 
Last edited:

IEC

Elite Member
Super Moderator
Jun 10, 2004
14,354
5,012
136
For comparison, here is an i7-8700K @ 5GHz, no AVX offset, 4700 cache clock with DDR4-3600 CL15 B-die memory:

 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
For comparison, here is an i7-8700K @ 5GHz, no AVX offset, 4700 cache clock with DDR4-3600 CL15 B-die memory:

The legendary 8.33ns stuff from G.Skill I have 3200CL14 @ 3733, but have been eyeing 3600CL15, as they have dropped to 180 eur per kit.
 

exquisitechar

Senior member
Apr 18, 2017
666
904
136
Looks good. "2700x is consistently beaten 20-30% by the 3700x" sums it up.

Oh man, compared to the 9900K even the 3700X is even winning in 4 out of 6 tested games in 720p low settings benchmarks, quite impressive.
 
Reactions: thigobr

exquisitechar

Senior member
Apr 18, 2017
666
904
136
Oh, it actually isn't. That was just from a misleading comment from someone who read the review before it was taken down...

3700X is also pulling quite a bit of power in this review?
 
Reactions: guachi

Hans Gruber

Platinum Member
Dec 23, 2006
2,214
1,152
136
It will be interesting to see reviews using x570 motherboards as well as X470 motherboards. Memory speeds of 3200mhz, 3600mhz, 3800mhz and 4000mhz so we can see how the memory scales. I read on tomshardware the sweet spot is 3600mhz with the low latency memory or 3733-3800mhz for performance value. Going higher is not worth the price. Since these are high end parts, value goes out the window. So we will probably see review with beyond 4000mhz memory.
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,655
136
Oh, it actually isn't. That was just from a misleading comment from someone who read the review before it was taken down...

3700X is also pulling quite a bit of power in this review?
Comparing it to the 2700x, my only working theory is PB was on basically turning it into a 3800x.
 
Last edited:

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,655
136
that is not too practical to say that, because what does then, a 3800X turn into?
There is probably a power cap to some degree. But I think that the 3700x is meant more for users looking for low power usage and therefore unlikely to turn PB on. Outside better binning I am guessing the 3700x and 3800x unleashed are relatively close to each other.
 

exquisitechar

Senior member
Apr 18, 2017
666
904
136
Comparing it to the 2700x, my only working their is PB was on basically turning it into a 3800x.
Another hint is that the 3700x got the same CB R15 ST score as the 3900x. And don't forget that the Intel parts are running 2667 memory. Sure, both sides are running their max officially supported memory, but it's something to keep in mind.
 

Gideon

Golden Member
Nov 27, 2007
1,709
3,927
136
It will be interesting to see reviews using x570 motherboards as well as X470 motherboards. Memory speeds of 3200mhz, 3600mhz, 3800mhz and 4000mhz so we can see how the memory scales. I read on tomshardware the sweet spot is 3600mhz with the low latency memory or 3733-3800mhz for performance value. Going higher is not worth the price. Since these are high end parts, value goes out the window. So we will probably see review with beyond 4000mhz memory.

AMD has stated publicly, that you have to set the Infinity Fabric multiplier to 1/2 if you want to run memory faster than 3733 MHz. That pushes latency back down to about ~2666 MHz levels. This means you have to clock the memory insanely high (e.g. 4400 Mhz +) just to get the same latency back you had at say 3200 MHz.

This means that running memory faster than 3733 MHz is only relevant for professional workloads that need more bandwidth. For latency-sensitive workloads (e.g. games) 3600-3733 is optimal, everything above isn't just diminishing returns, it will perform noticeably worse. If you have say 4400Mhz memory and you only game, you should run it @ 3733 and the lowest CL latency you can manage.

This is the relevant slide (Notice the worse latency even when at 4400Mhz):

https://i.redd.it/d24hymdbbm331.png
 

extra

Golden Member
Dec 18, 1999
1,947
7
81
Yeah that's what it looks like. Though I can't read that review. It's definitely (apparently much) more power efficient than Intel but there's no way it's at it's rated TDP. Then again, maybe that power is the entire system and the 570 chipset is a huge power hog like people have said. I can't read german.

Either way, it looks good, but not miraculous. Will definitely be my next purchase!

That review (thanks for posting it though) doesn't seem too useful, lol. We don't really know any juicy details like what ram speeds were used, etc. I'm excited though. Two more days. I got a big bottle of vodka and I'm going to go into a coma, wake me up when reviews are out! (kidding, don't worry!)

Also, what others have said: those cache improvements. The L3 bandwidth is insane. Of course, we can't see the latency. Two more days.

Would any of you incredibly technically inclined folks like to take a guess at how they got the L3 bandwidth so high? Those are L2 cache numbers.
 

SK10H

Member
Jun 18, 2015
117
50
101
There is probably a power cap to some degree. But I think that the 3700x is meant more for users looking for low power usage and therefore unlikely to turn PB on. Outside better binning I am guessing the 3700x and 3800x unleashed are relatively close to each other.

Look like if enabling PBO on 3700x can break the power cap to >105W <~160W without fixed oc, being so close to 3900x at gaming, I will cheap out and get the 3700x on my Asus C6H from 2700x. Hopefully it's still compatible with Zen3 so I can do a swap replacement again.
 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
Would any of you incredibly technically inclined folks like to take a guess at how they got the L3 bandwidth so high? Those are L2 cache numbers.

It has 4CCX's, just like TR 1950x and has ~same cache L3 bandwidth as TR 1950. What remains to be seen, if L3 bandwidth improves with 16 cores, or it stops scaling.

AMD's L3 cache numbers are so good cause each CCX has private 16MB of L3 that's what they get to use, and total bw is probably 250GB/s for each 4C ( was good and is good). On 9900K, Intel has 16MB of L3, but it is on ring bus and obviously all cores compete for it and it scales with uncore speeds only. From results in this thread it's obviuos that 8 Cores have acccess to ~400GB/s read, lower, but completely comparable numbers per core.
 

Arkaign

Lifer
Oct 27, 2006
20,736
1,377
126
It seems like Zen2 will be a lot of fun for tinkerer types. Settings and matching with the right components will make profound differences.

As opposed to socket 115x, which is basically what I'd term 'plug-n-meh' lol. Yes, going from DDR4-2667 to a nice Bdie @ 3733-4000 with tight latencies can pay off, but not with nearly the complexity or potential potholes you get from Intel HEDT or these Zen2s, TRs and the like.

I bet it will be really important to have competent OEMs to make decent Zen2 configs. I've seen some shockingly terrible OEM systems with superficially decent specs, but falling short for the dumbest reasons.
 
Reactions: lightmanek

Tuna-Fish

Golden Member
Mar 4, 2011
1,419
1,749
136
AMD has stated publicly, that you have to set the Infinity Fabric multiplier to 1/2 if you want to run memory faster than 3733 MHz. That pushes latency back down to about ~2666 MHz levels. This means you have to clock the memory insanely high (e.g. 4400 Mhz +) just to get the same latency back you had at say 3200 MHz.

Not quite. The IF: DDR multiplier is user-settable, and has three choices in your bios: auto, 1:1 and 1:2. By default, every bios ships with it in auto. What auto does, is sets it to 1:1 if ram is no faster than 3733, and 1:2 if it is faster than that. This was confirmed by a Finnish IT journalist asking it from AMD.

That is, at least from purely settings perspective, it's possible to run faster than 3733MHz ram at 1:1. However, this is running it at a faster speed than what AMD intended, and is therefore overclocking. How high the IF clock can actually go is still unknown, but there is likely at least some margin on there, AMD isn't likely to ship every chip at the very highest speed they can possibly wring out of them. So finding out the actual limits is something I expect to happen quite quickly after release.
 

DrMrLordX

Lifer
Apr 27, 2000
21,794
11,143
136
that is not too practical to say that, because what does then, a 3800X turn into?

We don't know yet. See below.

There is probably a power cap to some degree. But I think that the 3700x is meant more for users looking for low power usage and therefore unlikely to turn PB on. Outside better binning I am guessing the 3700x and 3800x unleashed are relatively close to each other.

Remember, PB != PBO. Precision Boost and XFR (when available) are enabled by default as a part of Core Precision Boost (CPB). PBO has to be user enabled. Bear with me here.

AMD_Robert already said that (and I paraphrase here) the current limit on Matisse is capped at 95a when the VRMs are thermally constrained. That's with PB + XFR (not PBO). If the VRMs are not thermally constrained, then the current limit is 140a. He did not say how close each individual CPU would get to the 140a limit based on their SKU. In theory, an 8c Matisse chip should pull 140a somewhere between all-core speeds of 4.5-4.6 GHz, or maybe a little faster on a golden sample. PBO is not going to get you that much higher thanks to process limits.

I don't honestly think the 8c chips are going to pull the full 140a on a board with cool VRMs. A 3600 or 3600x definitely won't. 3950x? Sure, easy. But I think we are going to see a situation where any chip can break past the 95a/142W limit set for stock operation. It depends on your VRM temps.

Activate PBO and there is no current limit. Or if there is one, it will be higher than 140a. You will still throttle with overheating VRMs.

That is, at least from purely settings perspective, it's possible to run faster than 3733MHz ram at 1:1. However, this is running it at a faster speed than what AMD intended, and is therefore overclocking. How high the IF clock can actually go is still unknown, but there is likely at least some margin on there, AMD isn't likely to ship every chip at the very highest speed they can possibly wring out of them. So finding out the actual limits is something I expect to happen quite quickly after release.

It's going to require some SoC voltage increases to explore that possibility. Fortunately, all the SoC functions are on 12nm. They should have the same voltage limits as Pinnacle Ridge.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |