***OFFICIAL*** Ryzen 5000 / Zen 3 Launch Thread REVIEWS BEGIN PAGE 39

Page 33 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

inf64

Diamond Member
Mar 11, 2011
3,759
4,213
136
I know this is about Zen 3 but… is there any word/rumor on Zen 4? It's a new/competitive design like Zen 3 or evolutionary from it?
No doubt they'll release soon, just I'm not sure if another 20% is happening this close.

Zen 2 and older would become way too depreciated, literally sold for nothing then, Zen 3 too... old tech in a year? They can't push too much on that front: designing chips has a price that must be payed back, same for dices (look how they reused the interconnect one).
I think it will be similar to Zen1->Zen2 jump. Around 15% IPC (average), possibly a 2x jump in AVX stuff ( versus Zen3). Chiplets might be 2x8 cores or even 1x16 configuration, caches will be larger with a possibility of a massive L4 (die stacking?). If they achieve ~15% IPC jump versus Zen3 then Zen5 needs ~21% over Zen4 to get to the same 40% goal AMD has stated in the past and managed : EX->Zen1, Zen1->Zen3
 

DrMrLordX

Lifer
Apr 27, 2000
21,794
11,143
136
They can't push too much on that front: designing chips has a price that must be payed back, same for dices (look how they reused the interconnect one).

They'll get their money back. Just because something's depreciated doesn't mean it's a total loss. AMD has long-term contracts to fulfill, they still have Renoir chips flying off the shelves, and sometimes there are shortages of Matisse chips as well. It is doubtful that there will be significant Zen2 supply remaining by late 2021 unless it is AMD's intention to flood the market with older product. AMD also certainly has the option to flog Zen3 for awhile. I'm not sure that even Alder Lake will amount to strong competition.

What do we know about Zen4? Not much. PCIe5, DDR5, USB4? Those are my assumptions. And it will be a new chipset. We have a thread about it around here somewhere . . .
 

HurleyBird

Platinum Member
Apr 22, 2003
2,725
1,342
136
I know this is about Zen 3 but… is there any word/rumor on Zen 4? It's a new/competitive design like Zen 3 or evolutionary from it?
No doubt they'll release soon, just I'm not sure if another 20% is happening this close.

I think it will be evolutionary (in terms of architecture) but I'm expecting a 20+% jump, easily.

Not because of the cores themselves, but because of everything else.

The new platform. DDR5. 5nm and everything that entails. Almost certainly more FP resources ala. Zen 2... but the big kicker is finally updating that terribly performant and power hogging IO die which Renoir showed was already a significant bottleneck in zen 2 and therefore can only be an even more substantial bottleneck irt the more powerful zen 3. Hopefully with some manor of advanced packaging this time.
 

HurleyBird

Platinum Member
Apr 22, 2003
2,725
1,342
136
They'll get their money back. Just because something's depreciated doesn't mean it's a total loss. AMD has long-term contracts to fulfill, they still have Renoir chips flying off the shelves, and sometimes there are shortages of Matisse chips as well.

Zen 3 will coexist alongside Zen 4 for a long while. As you pointed out, Intel is unlikely to convincingly beat Zen 3 for some time, but besides that the two designs aren't competing for fab space (in a capacity constrained context), and the platforms and memory standards differ to boot.
 
Reactions: prtskg

Makaveli

Diamond Member
Feb 8, 2002
4,758
1,148
136
Zen 3 will coexist alongside Zen 4 for a long while. As you pointed out, Intel is unlikely to convincingly beat Zen 3 for some time, but besides that the two designs aren't competing for fab space (in a capacity constrained context), and the platforms and memory standards differ to boot.

what i'm curious about is if we will see a Zen 3+ but on a AM5+DDR5+PCIe 5.0 platform to get the waters ready before they drop Zen 4.
 
Reactions: Tlh97 and Drazick

zinfamous

No Lifer
Jul 12, 2006
110,802
29,553
146
They really don't need to hurry. 10nm is still a problem at Intel, 7nn seems equally dubious. As long as Zen 4 arrives in the 15 - 18 month window AMD should be more than okay. As long as they don't go into stasis while in the lead haha.

yeah, exactly. While AMD doesn't need to hurry, they certainly shouldn't be pulling what Intel did over the last 10 years.
 
Reactions: Tlh97

moinmoin

Diamond Member
Jun 1, 2017
4,994
7,765
136
but the big kicker is finally updating that terribly performant and power hogging IO die which Renoir showed was already a significant bottleneck in zen 2 and therefore can only be an even more substantial bottleneck irt the more powerful zen 3. Hopefully with some manor of advanced packaging this time.
I think you need to lower your expectation a tad. A new IOD won't see much "better performance and less power hogging" all while introducing DDR5 and potentially PCIe 5 as well. Renior didn't really show what you claim, it only supports PCIe 3 instead 4 and not the same amount of lanes either. Uncore power usage is a real issue, but without cutting I/O it will be very interesting to see what improvements are possible.

One thing that Renoir did bring (though that's in a decidedly mobile context for now): The biggest power hog of the whole uncore is the IMC, especially the higher the RAM clock. Unlinking RAM clock from the IF clock allows the IMC to clock down when not required and saving power that way. Possibly DDR5 and PCIe 5 introduce new powers saving features as well?

what i'm curious about is if we will see a Zen 3+ but on a AM5+DDR5+PCIe 5.0 platform to get the waters ready before they drop Zen 4.
That's what the rumored Warhol appears to be about. Being wider Zen 3 appears to be bottlenecked by RAM bandwidth, so a Zen 3 on AM5 may unlock some more power at the high core end.

"About the only issue is that it is still stuck with 2-channel DDR4 memory that even at high speeds (e.g. expensive 4266Mt/s) cannot feed 16-cores / 32-threads in streaming algorithms despite the absolutely massive 64MB L3 cache."
 
Last edited:

HurleyBird

Platinum Member
Apr 22, 2003
2,725
1,342
136
Renior didn't really show what you claim, it only supports PCIe 3 instead 4 and not the same amount of lanes either.

Renoir showed that you could remove three quarters of the L3 cache and achieve roughly comparable performance. Not a single person expected it to perform so well. Slightly slower than Mattise overall, but with the ability to eek out wins here and there too.

Mattise and Renoir are both effectively "Celeron" implementations of the Zen 2 architecture (ditto Vermeer and Zen 3), albeit hobbled and cost-reduced in different ways. A true, premium Zen 2 implementation wouldn't lose to Skylake in anything except extreme outliers. It just goes to show how good the architecture really is.

A new IOD won't see much "better performance and less power hogging" all while introducing DDR5 and potentially PCIe 5 as well..

If I had to guess, I'd guess that Zen 5 will be the one with PCIe 5. DDR5 is more energy efficient than 4, although if that's true for the memory controller as well I have no idea. X570 is a good indicator of the IOD's excessive power consumption though.

New IOD should be on a better process, and will benefit from the lessons learned from the first IOD. But, I don't expect that lowering IOD power (and therefore raising CCD power allowance) is that big of a performance boost vs lowering latency and increasing bandwidth between the IOD and CCDs. My expectation is for desktop Zen 4 to use an interposer or some other advanced packaging. If it doesn't, it's still a "Celeron" implementation.
 
Reactions: Carfax83 and Tlh97

moinmoin

Diamond Member
Jun 1, 2017
4,994
7,765
136
Renoir showed that you could remove three quarters of the L3 cache and achieve roughly comparable performance.
No, it didn't do that. It did show which workloads profit from the larger L3 cache where Renoir performs significantly worse as a result, and that plenty workloads remain that aren't affected by that.

X570 is a good indicator of the IOD's excessive power consumption though.
Only if PCIe link state power management is not enabled. With it enabled it should be comparable with 400 series chipsets.

New IOD should be on a better process
Which by itself may mean nothing. Uncore logic is notoriously hard to scale down with power usage improvement as the target. Maybe AMD finds and uses a process that is more suitable for that than GloFo's 14/12nm and TSMC's N7.

But, I don't expect that lowering IOD power (and therefore raising CCD power allowance) is that big of a performance boost vs lowering latency and increasing bandwidth between the IOD and CCDs. My expectation is for desktop Zen 4 to use an interposer or some other advanced packaging.
This I agree with, with a new platform coming I also expect changes to how the uncore is organized with potential usage of an interposer and "X3D" stacking.
 
Reactions: Carfax83

HurleyBird

Platinum Member
Apr 22, 2003
2,725
1,342
136
No, it didn't do that. It did show which workloads profit from the larger L3 cache where Renoir performs significantly worse as a result, and that plenty workloads remain that aren't affected by that.

There are always outliers, but I've seen very few benchmarks where Renoir significantly suffers from its lack of L3.

But even if you can provide some, you would also need to look at Mattise with 75% of its cache disabled to draw any real conclusion from that.

A big issue here is that there is a severe lack of clock-normalized Renoir-vs-Mattise benchmarks out there. What seems obvious enough is that Renoir is closer to Matisse than Raven was to Pinnacle Ridge, even though the L3 cache size penalty is doubled. That's the whole reason Renoir took us all by surprise, because we were extrapolating from the last generation and therefore didn't expect it to perform as well as it did.

It's also worth mentioning that Renoir has double the memory write bandwidth as Matisse. The difference isn't just latency.

Only if PCIe link state power management is not enabled. With it enabled it should be comparable with 400 series chipsets.

Wrong.


 

CHADBOGA

Platinum Member
Mar 31, 2009
2,135
832
136
Now imagine what it would be like if AMD could get Zen4 out at about this time next year.
I think and hope AMD brings out a Zen 3+ on 5nm as the final hurrah for AM4.

That will give them time to make sure that their first DDR5 based CPU is as good as it can reasonably be, in late 2022.
 

DrMrLordX

Lifer
Apr 27, 2000
21,794
11,143
136
what i'm curious about is if we will see a Zen 3+ but on a AM5+DDR5+PCIe 5.0 platform to get the waters ready before they drop Zen 4.

There's this thing called Warhol . . .

I think and hope AMD brings out a Zen 3+ on 5nm as the final hurrah for AM4.

Actually

I don't see the point of any more releases for AM4 to be honest.

5nm Zen3+ on AM5 seems more likely.

Yeah, I agree with that. If AMD is going to rehash Zen3 on N5P (or similar; AMD may get their own custom node again), then it'll be AM5.
 
Reactions: french toast

CHADBOGA

Platinum Member
Mar 31, 2009
2,135
832
136
I don't see the point of any more releases for AM4 to be honest.

5nm Zen3+ on AM5 seems more likely.
It would give AMD more time to make sure that their first DDR5 platform is in tip top shape, if they need that time, plus I am assuming that there may be valuable lessons they can learn by going with a known quantity like Zen 3 and pumping it out on 5 nm.

Also DDR5 prices may not be where AMD would like them, in a year's time.

With Intel being so uncompetitive right now, AMD has the luxury of taking this approach.
 
Reactions: Tlh97

cellarnoise

Senior member
Mar 22, 2017
728
399
136
Chiplets... I know that it is not the same and had been done before this, but look up Pentium Pro for 1st workstation cost level... I worked on a top end Pro workstation at the time, like 200 mghz... Not bad for a single desktop top notch workstation then and really pretty fast. Sparks and whatever were behind, as well as anything AMD had. Around the time of the 1st consumer / workstation 2 cpu motherboards. Clunky.

I am most impressed with Zen 3 as AMD seams to have really learned from and optimized on this node. So cool to see what seems around same die size? and increased frequency from only manufacturing refinements, layout and optimizations! Making my current AMD cpus depreciate at a fast pace Now worth nothing?

Loving what AMD and the graphics side appears to be doing! I want!

Just trying to determine when to buy Intel stock again, as the Gorilla will likely come out of hibernation soon and I foolishly did not jump on the AMD ride!?

Though I plan on either a 5900 or 5950. My b350 board can only go to 3950x? Whatever...

Need more ram, power supply, motherboard, and new GPU. Then up and running depending on O.S. and other software.
 

Martimus

Diamond Member
Apr 24, 2007
4,488
153
106
Chiplets... I know that it is not the same and had been done before this, but look up Pentium Pro for 1st workstation cost level... I worked on a top end Pro workstation at the time, like 200 mghz... Not bad for a single desktop top notch workstation then and really pretty fast. Sparks and whatever were behind, as well as anything AMD had.
Not true. The K6 had higher IPC and higher clocks on integer workloads, which were the primary workloads at the time. It wasn't until the Pentium 2 that Intel took back the performance crown, but they had better marketing so it didn't affect them.
 

lixlax

Member
Nov 6, 2014
184
158
116
Zen 2 and older would become way too depreciated, literally sold for nothing then, Zen 3 too... old tech in a year? They can't push too much on that front: designing chips has a price that must be payed back, same for dices (look how they reused the interconnect one).
True, but isn't that what we, the tech enthusiasts, really want? The more AMD (or any other company) sells, the more agressive they can be (design costs etc are being paid quicker).
 

coercitiv

Diamond Member
Jan 24, 2014
6,384
12,802
136
Zen 2 and older would become way too depreciated, literally sold for nothing then, Zen 3 too... old tech in a year? They can't push too much on that front: designing chips has a price that must be payed back, same for dices (look how they reused the interconnect one).
Apple makes 1+ chips per year - with arguably strong performance gains per gen.

AMD is not in the same situation as Apple, but they're not in the same situation as Intel either: they're only making 2 dies that need frequent refresh: mobile monolithic and server chiplet. The IO dies have longer lifespan, so do other chips intended for secondary markets. Compare this to the slew of designs Intel normally dished out per year, even if their architecture changed less often in terms of big performance jumps: they had 2+ dies in mobile, 2+? in desktop, 2+ in servers.

As long as they have the engineering prowess to increase performance, logistics won't be a problem for the foreseeable future.
 

scineram

Senior member
Nov 1, 2020
361
283
106
what i'm curious about is if we will see a Zen 3+ but on a AM5+DDR5+PCIe 5.0 platform to get the waters ready before they drop Zen 4.
Probably. I think they will want to launch something in 2021, if only to introduce AM5 and DDR5.
 

Gideon

Golden Member
Nov 27, 2007
1,709
3,927
136
Probably. I think they will want to launch something in 2021, if only to introduce AM5 and DDR5.
Warhol seems to fit thet role, though it's still somewhat unclear what it entails. On top of the new I/O die I also would have thought they'd want to use a 6nm versions of Zen 3 Compute die on at least some higher-end SKUs. (as the masks are compatible it should be a minimal effort) but it looks like it's not the case. Perhaps 6nm isn't ready by that time or is too expensive.
 
Reactions: Tlh97 and scineram
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/    |