Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[TESKATLIPOKA]

What about possibility of higher number of cores in the future?
With 105W TDP it would be effectively impossible, at 170W there's at least a chance for 24 or maybe even 32 cores (Zen 5 + Zen 5c).

If they release a 24C48T Zen 4 then It would be fantastic. Then they could lower noticeably the clocks and still be the winner in MT.
The problem for Zen4 is that there is not enough cores in desktop, so they have to increase the clocks to stay competitive or win.

 

[TESKATLIPOKA]

Keep in mind nobody is forcing you to run 105W and 170W TDP. Just like with Intel, you have multiple levers to fully customize your build.

I keep my 12700K @ PL1=PL2=125W and it's perfect for me. Back when I configured my 1600X build using semi-passive cooling, I didn't even adjust power limits but went for custom temperature target instead. I liked it so much than even with my Intel build I'm attempting to emulate that behavior by lowering max temp setting to 90C instead of auto.

The non-X SKUs will be great for value builds, but if you REALLY value efficiency like you say, then you want the better binned SKUs with custom power limits.

Of course you are right and I could change It to my preferences, but reviews will compare these SKUs at stock.

BTW I will buy mobile Zen 4 next year. Phoenix Point is the most likely candidate for me.

 

[ryanjagtap]

If they release a 24C48T Zen 4 then It would be fantastic. Then they could lower noticeably the clocks and still be the winner in MT.
The problem for Zen4 is that there is not enough cores in desktop, so they have to increase the clocks to stay competitive or win.

In a video by Dr. Ian Cutress, he speculates that there is no space available to put another CCX in the traditional (horizontal) way on the substrate, he says they'd have to place the CCX vertically oriented and move the IOD well as the routing et al to make that viable. ( On that same speculation, if they place the CCX vertically, the L3 cache are adjoined and the V-cache model of a 16 core variant can have the full access of it for all the cores and not just for their respective CCX only)

It's near 8:10 or something

 

[TESKATLIPOKA]

So you are arguing about nothing….I see.

Zen 4 is far more efficient than Raptor Lake or Alder Lake. AMD raised power limits, which raised TDP, in order to unlock multicore performance. There is nothing wrong with this. Intel has high power limits because they are using a design that would not otherwise be competitive.

Just because of you, I am putting a 7950x in a mini ITX build.

I see, so increasing TDP is nothing for you.
When Intel raised TDP(power limit) to stay competitive in MT It was bad, very bad.
When AMD does It to stay competitive in MT, there is nothing wrong with It, because Intel already did It.

Just because of me you don't need to buy 7950x for a mini ITX build. You can buy even Intel, It's your choice.

 

[leoneazzurro]

I am talking about performance/W here for Desktop SKUs they just announced.
These models won't have 62% higher performance in MT. Maybe some extreme case, but I am pretty certain not on average.
There are no such claims for 105 or 170W TDP. They claimed 74% higher performance at 65W TDP 5950x vs 7950x, but I am talking about SKUs with higher TDP.

Yes but you are focusing only on the low end where the core count deficit compared to Intel has forced AMD to push up frequencies in order to stay competitive while you are not considering those SKUs (7900X and 7950X) where the architecture/process advantages can be shown. In other words, we could check a 7700X compared to a 5800X (another notoriously not very power friendly SKUs in the Vermeer line) and we would probably find similar results to what AMD claims for the top SKUs, that is +35% @170W TDP 7950X/5950X. Or the user could run these CPUs in Eco mode (a standard feature in many AMD boards) and enjoy 90%+ of the peak performance. This is the result of the competitive environment set by Intel and by the cost/benefit analisys done by AMD. Lower end SKUs are in a tough position but 7900X/7950X are in a way different spot. But this has very little to do with problems in the architecture in itself (which, on the opposite side, is showing great promise for the server market). AMD could chose tomorrow to release 10 and 14 core CPUS (7800X? 7930X?) and scale down the TDP/price of the lower end SKUs and become way very "attractive" in the perf/W at all performance points. If this is not happening is because AMD is focussing more on the very high end of the market and servers and wants to keep high margins. Lower end will be probably addressed way better by monolithic N4 Phoenix APUs which are not so far away, as it will happen in mobile with the Phoenix/Dragon Range split.

 

[TESKATLIPOKA]

Yes but you are focusing only on the low end where the core count deficit compared to Intel has forced AMD to push up frequencies in order to stay competitive while you are not considering those SKUs (7900X and 7950X) where the architecture/process advantages can be shown. In other words, we could check a 7700X compared to a 5800X (another notoriously not very power friendly SKUs in the Vermeer line) and we would probably find similar results to what AMD claims for the top SKUs, that is +35% @170W TDP 7950X/5950X. Or the user could run these CPUs in Eco mode and enjoy 90%+ of the peak performance. This is the result of the competitive environment set by Intel and by the cost/benefit analisys done by AMD. Lower end SKUs are in a tough position but 7900X/7950X are in a way different spot. But this has very little to do with problems in the architecture in itself (which, on the opposite side, is showing great promise for the server market). AMD could chose tomorrow to release 10 and 14 core CPUS (7800X? 7930X?) and scale down the TDP/price of the lower end SKUs and become way very "attractive" in the perf/W at all performance points. If this is not happening is because AMD is focussing more on the very high end of the market and servers. Lower end will be probably addressed way better by monolithic N4 Phoenix APUs which are not so far away, as it will happen in mobile with the Phoenix/Dragon Range split.

1. The architecture of Zen4 is great, I never disputed that. The problem is low core count especially in low-end, which was compensated by much higher clocks.
2. You can compare 5800x(105W) vs 7700x(105w) and It would look great for Zen4, but we don't know If 7800x won't have 170W for example.
3. I have no worries about mobile Zen4. AMD will release 16c32T there, Intel will have a big problem, even If they release 24C32T Raptor.

edit: fixed the wrong TDP value.

 

[Abwx]

When AMD does It to stay competitive in MT, there is nothing wrong with It, because Intel already did It.

That s it, Intel did set a precedent.

Perf/watt doesnt seem to be an as good selling point that pure perf, AMD knows that their TDP limited AM4 plateform allowed Intel to seemingly compete against the 5950X with a chip that was a contender for the 5900X should it set at comparable TDP.

That being said there will still be eco modes for the 7950X to limit power at 142W and 88W for the power aware public.

 

[leoneazzurro]

1. The architecture of Zen4 is great, I never disputed that. The problem is low core count especially in low-end, which was compensated by higher clocks.
2. You can compare 5800x(65W) vs 7700x(105w) and It would look much better, but we don't know If 7800x won't have 170W for example.
3. I have no worries about mobile Zen4. AMD will release 16c32T there, Intel will have a big problem, even If they release 24C32T Raptor.

1) The core count/TDP choice for SKUs is purely commercial, nothing to do with flaws in the architecture by itself. As told, AMD could choose to use different core counts and achieve better efficiency. But it would mean to lower margins, so they will do only if needed.
2) 5800X is 105W part. I would not advice to run either part @170W
3) fully agree, especially because of the split. U parts will be addressed by Phoenix, while DGPU laptops will receive Dragon, and that alone will set up a completely different scenario respect to current situation where Rembrandt tries to cover everything.

 

[inf64]

OneRaichu posted a few interesting tweets:

https://twitter.com/x/status/1565648989784985600

"1/2 1. Zen 4 is better than a lot of people guess, and the accumulated heat question is not fatal for it. 2. Because the manufacturer still adjusts the bios setting, the power consumption of the new product is not accurate. Two examples: "

"2/2 (1) In July, Zen 4 ES needs 260w in the FPU test and only 4.8 GHz, but now it can get almost 220w and 5.0Ghz. (2)In the last test of my test, RPL-S need almost 350w to finish the CB benchmark and stress test. After two version microcode updates, it is less than 300w in now. "

I expect that retail SKUs perform just like AMD shown in the slides (ie. +48% R23 MT increase Vs 5950X). They had to be 100% sure or they would not post such claims . Just relax and wait for the 20th embargo lift.

That s it, Intel did set a precedent.

Perf/watt doesnt seem to be an as good selling point that pure perf, AMD knows that their TDP limited AM4 plateform allowed Intel to seemingly compete against the 5950X with a chip that was a contender for the 5900X should it set at comparable TDP.

That being said there will still be eco modes for the 7950X to limit power at 142W and 88W for the power aware public.

Pure performance is what majority of potential customers want to buy, even if they don't go for the top SKU. The halo effect is real, so it's understandable why AMD followed suit with the power increase. As long as it works as intended and can be tweaked by curve optimizer and similar tools, all will be fine.

I expect great things from tweaked Zen 4 chips, especially when we have leaks that manual all-core OC can be combined with PBO2 so we could get best of both worlds - highest possible ST performance and tweaked/best performance for MT apps (with lowered Vcore). Premium cooling is a must in any case.

 

[TESKATLIPOKA]

1) The core count/TDP choice for SKUs is purely commercial, nothing to do with the architecture by itself. As told, AMD could choose to use different core counts and achieve better efficiency. But it would mean to lower margins, so they will do only if needed.
2) 5800X is 105W part. I would not advice to run either part @170W
3) fully agree, especially because of the split. U parts will be addressed by Phoenix, while DGPU laptops will receive Dragon, and that alone will set up a completely different scenario respect to current situation where Rembrandt tries to cover everything.

1) I know and agree with you
2) During cold winters, It could be a good choice
3) I have to wonder If Phoenix will go down to 15W, I heard(read) only about 35-45W. Maybe with a weaker IGP we will see a 15W model, but for 1536SP(6WGP RDNA3) It looks too low.

 

[leoneazzurro]

I am quite sure Phoenix will go down to 15W/28W as Rembrandt today, but it will be OEMs to choose. There will be U parts for thin and light for sure, as Intel has its own parts for competing in that market. 35-45W is the usual range for cheaper "all in one" laptops with integrated graphic.

 

[In2Photos]

3. AMD already have made price cuts and introduced new cheaper AM4 models like the $200 5600 and $300 5700X. Since 7600X is at $300 they will not price 5800X3D lower than $400 because it will cannibalize the ZEN4.

I already mentioned this but the 5800X3D is currently $385.

AMD Ryzen™ 7 5800X3D 8-core, 16-Thread Desktop Processor with AMD 3D V-Cache™ Technology https://a.co/d/9f8tCL8

 

[nicalandia]

Guys. Lets not make a big deal out of unreleased CPU with Beta Bios. Even after release there is room for improvement.

 

[nicalandia]

I was very accurate with my prediction on die Size and FPU size.

 

[moinmoin]

These models won't have 62% higher performance in MT than their predecessors. Maybe some extreme case, but I am pretty certain not on average.

These models won't have 62% higher power consumption than their predecessors. Maybe some extreme case, but I am pretty certain not on average. These higher TDPs give more headroom, that's all there is.

I'm personally most interested in U-series processors since the above is a nonsense competition, a nonsense debate, a whole lot of waste of time and energy for everyone and everything involved. Just for somebody to be able he got the most fastest most wasteful piece ever at that point in time. And even be proud of that for some reason.

Lower end will be probably addressed way better by monolithic N4 Phoenix APUs

Phoenix will use "AMD chiplet architecture", see

Page 10 - Question - AMD Phoenix/Zen 4 APU Speculation and Discussion

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[leoneazzurro]

Phoenix will use "AMD chiplet architecture", see

Page 10 - Question - AMD Phoenix/Zen 4 APU Speculation and Discussion

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Yeah, I saw those articles but it was very unclear what in the APU would have been split in the chiplet, if the whole GPU or only the new AI accelerator part, if there will be IC and so on. As early leaks gave a monolithic block for CPU+GPU, and a 4N product should have significant area reduction compared to 6N (so the new GPU will be not way bigger than the current one in RMB), I am still more inclined to think that only the AIE will be added as a chiplet on the package.

 

[eek2121]

I see, so increasing TDP is nothing for you.
When Intel raised TDP(power limit) to stay competitive in MT It was bad, very bad.
When AMD does It to stay competitive in MT, there is nothing wrong with It, because Intel already did It.

Just because of me you don't need to buy 7950x for a mini ITX build. You can buy even Intel, It's your choice.

The power limit of Zen 4 is 230W. The TDP is 170W. That means the chip absolutely will not be running at at constant 230W. More importantly, that is the total amount of power the entire chip will consume, not just the CPU cores. Intel no longer has a TDP, just a power limit of 241W. That means that for a given workload, the Intel chip will always consume more power than the AMD chip.

Even if you set the power limit of the Intel chip to 230W, it would still consume more energy.

That is because the AMD chip would eventually back off on power as it hits TDP limits, while the Intel chip would not.

It sounds like many of you have hurt feelings that AMD took a page from Intel’s playbook.

Is Zen 4 more power efficient than Zen 3? Yes. Is Zen 4 more power efficient than Alder Lake? Yes. Will Zen 4 more power efficient than Raptor Lake? Absolutely.

 

[Kaluan]

Oh my, so many of you having meltdowns over announced TDP and rumored heat dissipation woes...

Intel has essentially normalized high power consumption on their consumer products since 9th gen (~5 gens ago), they literally had to show downclocked Alder Lake in slides to proper emphasize it's PPW gains vs Rocket Lake (which had atrocious PPW to begin with) if I remember correctly. Now people are angry that AMD can't or doesn't want to compete with Intel with one hand tied behind it's back for the ~4th time in a row. I mean what? 😐

Someone said Ryzen 7000 might have problems if reviewers review day bench it on air... bruh, both AMD and Intel provide tech outlets with review kits, and those pretty much always contain AIOs.

Quite a few people were concerned about high temperatures on ADL i9 (info from actual launch day revews no less) but it turned out to be mostly overblown. We're also getting reports that Raptor Lake has a max junction temp of 115C(!) and leaks were it peaks over 100C on the die. But no one lost their marbles over it or thought it is/will be a terrible product.

Anyway, interestingly enough launch has been moved up by 12 days, but the NDA lifting only by 7.

Edit: Phoenix Point will obviously have SKUs at/tuned for 15/28W as well... that was the whole point of it wasn't it?
Steam Deck 2 hype is already getting it's first rumor legs, pretty sure it will be based on a Phoenix Point derivative. And the handheld PC makers will also surely want a piece of that Zen4/RDNA3 action.
so I'm not sure how this is even up for debate.

 

[eek2121]

Oh my, so many of you having meltdowns over announced TDP and rumored heat dissipation woes...

Someone said Ryzen 7000 might have problems if reviewers review day bench it on air... bruh, both AMD and Intel provide tech outlets with review kits, and those pretty much always contain AIOs.

Quite a few people were concerned about high temperatures on ADL i9 (info from actual launch day revews no less) but it turned out to be mostly overblown. We're also getting reports that Raptor Lake has a max junction temp of 115C(!) and leaks were it peaks over 100C on the die. But no one lost their marbles over it or thought it is/will be a terrible product.

Anyway, interestingly enough launch has been moved up by 12 days, but the NDA lifting only by 7.

Edit: Phoenix Point will obviously have SKUs at/tuned for 15/28W as well... that was the whole point of it wasn't it?
Steam Deck 2 hype is already getting it's first rumor legs, pretty sure it will be based on a Phoenix Point derivative. And the handheld PC makers will also surely want a piece of that Zen4/RDNA3 action.
so I'm sure how this is even up for debate.

It's interesting that people are doing this. Leaks from ES samples, prerelease BIOS, etc. and suddenly everyone is going nuts. Zen 4 is going to be a fantastic chip.

I doubt a successor to Steam Deck will use Zen 4, unfortunately, unless Valve scores a heck of a deal with AMD. More likely it will use some form of Rembrandt unless they release a new higher priced SKU rather than a successor. The whole reason valve went with the APU it did was cost vs. gaming performance at decent power. Zen 4 is a premium product. Rembrandt will end up being the next budget product.

I could be wrong, however. I would welcome being wrong, actually. I love my Steam deck, but I do wish it had a bit more horsepower.

 

[Markfw]

Oh my, so many of you having meltdowns over announced TDP and rumored heat dissipation woes...

Intel has essentially normalized high power consumption on their consumer products since 9th gen (~5 gens ago), they literally had to show downclocked Alder Lake in slides to proper emphasize it's PPW gains vs Rocket Lake (which had atrocious PPW to begin with) if I remember correctly. Now people are angry that AMD can't or doesn't want to compete with Intel with one hand tied behind it's back for the ~4th time in a row. I mean what? 😐

Someone said Ryzen 7000 might have problems if reviewers review day bench it on air... bruh, both AMD and Intel provide tech outlets with review kits, and those pretty much always contain AIOs.

Quite a few people were concerned about high temperatures on ADL i9 (info from actual launch day revews no less) but it turned out to be mostly overblown. We're also getting reports that Raptor Lake has a max junction temp of 115C(!) and leaks were it peaks over 100C on the die. But no one lost their marbles over it or thought it is/will be a terrible product.

Anyway, interestingly enough launch has been moved up by 12 days, but the NDA lifting only by 7.

Edit: Phoenix Point will obviously have SKUs at/tuned for 15/28W as well... that was the whole point of it wasn't it?
Steam Deck 2 hype is already getting it's first rumor legs, pretty sure it will be based on a Phoenix Point derivative. And the handheld PC makers will also surely want a piece of that Zen4/RDNA3 action.
so I'm not sure how this is even up for debate.

I just want to see what happens on review day, and then for myself. I believe they will be more energy efficient.

 

[Makaveli]

1. The architecture of Zen4 is great, I never disputed that. The problem is low core count especially in low-end, which was compensated by much higher clocks.
2. You can compare 5800x(65W) vs 7700x(105w) and It would look great for Zen4, but we don't know If 7800x won't have 170W for example.
3. I have no worries about mobile Zen4. AMD will release 16c32T there, Intel will have a big problem, even If they release 24C32T Raptor.

there is no 65watt 5800x its a 105watt part and you have to enable eco mode to get it to 65W do you mean the 5700x?

 

[gdansk]

I can't say I am impressed with Zen 4 Desktop SKUs.
Zen4 SKUs released by AMD are less energy efficient than their predecessors even though they are on 5nm. High clocks are simply killing any efficiency gains from the process.
Ryzen 5 5600X vs Ryzen 5 7600X -> 65W vs 105W (+62%)
Ryzen 7 5700X vs Ryzen 5 7700X -> 65W vs 105W (+62%)
Ryzen 9 5900X vs Ryzen 5 7900X -> 105W vs 170W (+62%)
Ryzen 9 5950X vs Ryzen 5 7950X -> 105W vs 170W (+62%)
I am more interested in model without X in their name and back to 65 and 105W TDP.

I understand what you mean but AMD seems pretty confident it is even better at low TDPs (based on their slides). And if you think about it in performance per watt, Zen 4 was designed for Genoa-like clock rates not what we'll see in desktop.

In performance per watt Zen 4 stands to be better than Zen 3. Just not the way AMD is shipping it in Raphael. You'll have to set your own power targets if you like efficiency. Or wait for non-X SKUs.

 

[lobz]

Yes, I am basing It on TDP and not actual power consumption tests, because there are none yet.
105-170W TDP (142-230W PPT) or 62% increase happened for a reason.
Do you think those values are only for AVX-512 and in everything else It will be a lot less? I am not so optimistic.

Then I can completely ignore your argument till launch, thank you for clarifying.

 

[nicalandia]

Going off by Locuza's diagram and what AMD might pull off in regards of Zen4c FP Register size(compact FPU as found on PS5) This is how smaller the Zen4c will be compared to Zen4





I just don't see how AMD will be able to Halve the Zen4 CCD

 

[Rigg]

1. Faster than 5800X3D ??? I highly doubt it.

That's fine. you can doubt whatever you want. it doesn't make your previous statement any less questionable.

2. Two things, they said up to 20% and all products will not see the same price increase.

OK?? this is literally verbatim what i wrote. Not sure what you are arguing with. i noticed you didn't mention how this equates to a lower MSRP for the 13700k than the 12700k though.

3. AMD already have made price cuts and introduced new cheaper AM4 models like the $200 5600 and $300 5700X. Since 7600X is at $300 they will not price 5800X3D lower than $400 because it will cannibalize the ZEN4.

The 5800x3D is a bit unique and may not see as aggressive of price cuts as other Zen3 SKU's. It's currently on sale for $385 on Amazon though so you're already wrong. I don't think AMD cares about cannibalizing the 7600x and 7800x. It's clear from their pricing strategy they want to entice you to buy Zen 4 Ryzen 9 CPU's. Their margins are probably much better on those higher end SKU's with a more expensive I/O die and packaging cost. My speculation is that the entire desktop zen 3 desktop lineup settles at $80-$400 w/ 5800x3D @ $300 by early next year if not even sooner. The 7600x should compare nicely to the 5800x3D while mostly outperforming it, so I'm not sure why the 5800x3D needs to be more expensive.

4. Using the 13700K at default there are no issues with motherboard VRMs . There is also the option for the 65W 13700 that is cheaper , at way lower TDP and only 2-3% slower vs 13700K in gaming. I bet it will still be at least as fast vs 5800X3D in gaming and way faster in MT.

I think your entire premise is deeply flawed. Why would a budget conscious buyer who grabbed a 5600(x)/B550 or 12400/DDR4 B660 in 2022 suddenly be upgrading to a new CPU in early 2023? A corresponding major (expensive) GPU upgrade seems like the only logical reason someone would do this. Why wouldn't the value conscious Intel buyer go for a 13600k or a discounted/used higher end ADL vs 13700K in this scenario?

Nobody with 2 brain cells to rub together bought a 12400 with expensive high freq low latency DDR5. I'm not confident Raptor Lake can even beat 5800x3D with pedestrian DDR4 in gaming.