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

[DisEnchantment]

Some systems are multi-purpose though. e.g. my current system with a threadripper is a threadripper for software development purposes (lots of cores does wonders for large C++ codebases) but at the same time I'd like to play the odd game on it. The higher max clocks for single-core are a major draw over EPYC(or even threadripper pro) for that case and even my current 2990wx I notice is getting slow-ish for gaming, in particular if one wants to upgrade to >60 fps on modern AAA games. In some cases you can get a whole bunch of extra perf by limiting the game to one CCX but it is all kinda messy.

2990WX is a bit behind though today. It has too many quirks.
Zen 3 TR should be much better.
Much higher ST clocks and 8 core CCX. You can get at least one core boosting high from each CCD for 8 high boosting cores in a 64 Core Zen 3 TR.
And inter core fabric is better in Zen2 and above.

Then there are likely the V Cache SKUs. Those will crush many engineering loads.
Once a lot of products move over to N5 in Q2 we can bet those V Cache SKUs will arrive.
Michael from Phoronix has some benchmarks and he was very stoked about the V Cache parts in dev loads but he was requested not to share by AMD.
In less than 10 days you will see Phoronix Linux benchmarks including dev loads for the V Cache parts.
Those will be no slouch in gaming either if AMD's Computex demo is to be believed.

Anyway, I am not sure HEDT SPR will be launched right off the bat in 2022.
But if they are then that is good, maybe we have a chance to see 96 Core 100Billion+ Xtor HEDT CPUs as a response at some point if not immediately.
But probably 8K a pop at least

 

[jamescox]

Yes, cause we're back in 2003 right? When quad cores were enterprise Xeon chips?

That's a silly argument since even servers moved away from that more than a decade ago. I remember when Xeons were ultra-low clocked Pentiums with large amount of caches, so per socket it heavily underperformed and it only made sense with transactional servers or when you wanted scalability with MP systems.

That's no longer the case, since Turbo allows clocks to be almost as high as low core count parts when low number of cores are active. So servers DO care since we know even in that space there are limits to scalability.

Besides, these are HEDT. I assure you rich gamers will be among the people who buy them. These will be tested by Anandtech, Techpowerup and the rest. Of course low thread count performance matters, because even Cinebench loses scaling pretty quickly. They will cry foul if with only 16 cores active it runs at 3GHz, rather than 4.5GHz it should be able to reach.

And that's if you consider 16 cores "low threads". Not that a 50 core 3.4GHz Golden Cove will lose to competitors.

The power usage figure is also irrelevant in this space if it also performs significantly higher.

Oh, so you are only talking about rich idiots who buy a 64-core processor for gaming only. I guess Intel can have that market. There processors will likely need the massively over spec cooling system that comes with such boutique gaming systems. They will be a lot more expensive than the competition, so they will fit right in with the status symbol buying characteristics of such “gamers”.

I have been working with some people trying to spec Epyc based systems and, guess what, they ignore the boost clock almost completely. All cores are expected to be in use and they have to be spec’d for the worst case. If you are buying a system for content creation and also a little gaming, then you might care, but any upcoming AMD or intel HEDT solution will likely be GPU limited anyway. If you want to pay more, I am sure Intel will take you up on that offer.

 

[andermans]

For Threadripper, maybe. For Epyc, and threadripper pro, probably unlikely. I have seen a lot of server systems where they actually lock it to the base clock for consistency. Do you really thing that a Threadripper 5000 is going to be a performance bottleneck in games? This gaming argument for 32+ core processors is ridiculous. I think Threadripper 5000 will do very well in games and if they make a version with stacked cache it will crush everything else, but still not worth the cost as a gaming only processor at all.

The 2990WX was marketed as a workstation processor for reasons. I think that is the one that had 4 cpu chips but only 2 quadrants connected to memory. You might be able to get more performance with some numa configuration. I have been playing with that on some Epyc processors. Anyway, that is around 4 years old now and wasn’t the best gaming processor from the start.

Yeah was talking purely Threadripper/HEDT here. Agree that gaming-like dual purpose is less likely for EPYC and agree that my 2990wx is actually getting old. Just wanted to offer some arguments why some people might also look at the perf with few cores active as a secondary constraint (i.e. gaming as the secondary purpose). At that point you have e.g. the EPYC 7763 or 7742 (I know one is Zen3 and the other is 225W, but AFAIU the 280W zen2 part doesn't have public specs) limited to ~3.5 GHz boost clocks while a 3990x is 4.3 GHz, so this is something I do have in my head for selecting HEDT vs. workstation vs. EPYC vs. mainstream.

 

[IntelUser2000]

Oh, so you are only talking about rich idiots who buy a 64-core processor for gaming only.

The "idiots" are also customers.

Marketing as an "HEDT" chip and assuming only people who max out 100+ threads you'll lose a lot of potential customers.

Again, you consider 20 cores + HT to be "small" amount of threads? What about 30? 40?

You want a chip that's the fastest in all scenarios and that's what will happen.

We all know Intel is totally sucking in servers now. That's not the point.

 

[jpiniero]

I have been working with some people trying to spec Epyc based systems and, guess what, they ignore the boost clock almost completely. All cores are expected to be in use and they have to be spec’d for the worst case. If you are buying a system for content creation and also a little gaming, then you might care, but any upcoming AMD or intel HEDT solution will likely be GPU limited anyway. If you want to pay more, I am sure Intel will take you up on that offer.

The workstation market (read: TR Pro) is somewhere in between. They want cores but they want frequency/per core performance too.

 

[igor_kavinski]

AMD’s Mark Papermaster Reveals Ryzen’s Stunning Success Secrets: 2022 Will Excite PC Enthusiasts (forbes.com)

Antony, with regards to the upcoming generation – I point to CES in January. We’re excited to be revealing some additional details on our new product launches that will deliver phenomenal experiences and as we’ve said, later in the year as it progresses we’ll share more detail on Zen 4 with some mentioned at CES and more announcements on it over the course of 2022.

The new year's going to start off good.

 

[soresu]

AMD’s Mark Papermaster Reveals Ryzen’s Stunning Success Secrets: 2022 Will Excite PC Enthusiasts (forbes.com)



The new year's going to start off good.

CES announcer:

".... Oh hi Mark!"

Sorry couldn't resist!

🤣🤣🤣🤣

 

[Hans Gruber]

I am just going to throw this out there. I think AMD need to consider either extending the life of the AM4 platform for Zen 4 or having AM5 motherboards with DDR4 memory. I know DDR5 is the future. There is a lot of high end DDR4 memory that can hold it's own until the DDR5 memory advances. Giving builders the option for DDR4 or DDR5 would make more who are watching and waiting on the DDR5 memory market.

This is just one idea that is different than the Intel tick-tock approach of two generations of CPU's until they force a motherboard upgrade.

 

[NTMBK]

I am just going to throw this out there. I think AMD need to consider either extending the life of the AM4 platform for Zen 4 or having AM5 motherboards with DDR4 memory. I know DDR5 is the future. There is a lot of high end DDR4 memory that can hold it's own until the DDR5 memory advances. Giving builders the option for DDR4 or DDR5 would make more who are watching and waiting on the DDR5 memory market.

This is just one idea that is different than the Intel tick-tock approach of two generations of CPU's until they force a motherboard upgrade.

With the separate IOD, I guess that they could fairly easily produce an AM4 version of the next gen chip just by packaging it with the old IOD. But that feels like a losing strategy- DDR5 is only going to get more mature and more readily available over the next year. It's time to let DDR4 go.

 

[NTMBK]

AMD’s Mark Papermaster Reveals Ryzen’s Stunning Success Secrets: 2022 Will Excite PC Enthusiasts (forbes.com)



The new year's going to start off good.

If CES 2022 even happens... a LOT of companies have backed out.

 

[jpiniero]

I am just going to throw this out there. I think AMD need to consider either extending the life of the AM4 platform

Zen 3D is that extension.

 

[igor_kavinski]

DDR5 also seems like a perfect match for Zen4 due to AMD's MT focus in CPU design and DDR5's more pronounced advantage in MT workloads.

 

[LightningZ71]

Zen3d should do well as an extension of AM4 because the larger cache should be able to hide a good portion of the bandwidth deficit of DDR4 while also enhancing the platform 's lagency advantage.

DDR5 will be a bigger boon for ZEN4 as its higher IPC will put greater bandwidth demands on the memory subsystem.

 

[Markfw]

If CES 2022 even happens... a LOT of companies have backed out.

Isn't that due to covid concerns ?

 

[soresu]

I am just going to throw this out there. I think AMD need to consider either extending the life of the AM4 platform for Zen 4 or having AM5 motherboards with DDR4 memory. I know DDR5 is the future. There is a lot of high end DDR4 memory that can hold it's own until the DDR5 memory advances. Giving builders the option for DDR4 or DDR5 would make more who are watching and waiting on the DDR5 memory market.

This is just one idea that is different than the Intel tick-tock approach of two generations of CPU's until they force a motherboard upgrade.

Are X670/B650 independent of the memory type though?

If not they will need more chipsets.

 

[Hitman928]

Isn't that due to covid concerns ?

Yes. I believe most, if not all, the major players are still planning on virtual sessions.

 

[Hans Gruber]

Are X670/B650 independent of the memory type though?

If not they will need more chipsets.

I guess all I am saying is for the earliest AM5 boards. They should have a DDR4 option board. Maybe producing the Zen4 on the AM4 platform might not make sense. By extending DDR4 support. It would press the prices for DDR5 down quicker. Based on everything I have read. The high clocking lower latency DDR4 ram does pretty well against DDR5. As the speeds on DDR5 increase over time, DDR4 will become a relic and afterthought.

After I upgrade to Zen3d next year, I am done for around 2 years.

 

[soresu]

After I upgrade to Zen3d next year, I am done for around 2 years.

To be honest I think that may be the most significant part of the driving force behind Zen3D in the first place given Zen4 is supposed to be coming within the same year.

 

[eek2121]

That makes sense anyway - I'd be very surprised to see Zen 5 before 2024.

So, what you are saying is that they want to fade into irrelevance? Genuinely asking here. Look at the long play of competitors. ALL of AMD's competitors are setting up for a huge sweep in 2022-2023, including Intel. Be as pessimistic as you want, but you might want to examine everyone and play devil's advocate before making such claims. A lot of folks here are skeptical of Zen 4 improving upon Zen 3 IPC by more than 20%. In order for AMD to even beat Raptor Lake, they'll have to hit big and hard. I have no doubt they'll absolutely kill it in the perf/watt market, but I have serious doubts that they will win the performance crown going forward. Let's rewind this chat to Athlon vs. P4 and then Intel rolling out Core/Core 2....

I hope I'm wrong, otherwise I'm gonna be broke. I'm up a BUNCH from AMD stock so far.

 

[HurleyBird]

So, what you are saying is that they want to fade into irrelevance? Genuinely asking here. Look at the long play of competitors. ALL of AMD's competitors are setting up for a huge sweep in 2022-2023, including Intel.

Chances of RL beating Zen 4 are about 1-in-100 (for desktop). And even if it does, AMD still has a bunch of levers they can pull including V-cache and higher core counts. Laptops could be another story completely thanks to the little cores, and we should have a better idea when mobile Alder Lake comes out.

RDNA 3 will almost certainly wreck whatever Nvidia puts up against it in Perf/W, and will probably win at raw rasterization, but is less likely to win in Ray-Tracing (will probably narrow the gap) plus DLSS is still a big advantage. But even if RDNA 3 just obviously wins at just about everything, AMD probably needs to win a few generations in a row to make a noticeable dent. Nvidia can afford to lose a single generation.

The next generation of products isn't what AMD needs to worry about. If anything AMD is poised to improve their position. It's the generations that come afterwards that might be able to knock AMD down a peg. Neither Nvidia nor Intel are asleep at the wheel.

 

[DisEnchantment]

Chances of RL beating Zen 4 are about 1-in-100 (for desktop). And even if it does, AMD still has a bunch of levers they can pull including V-cache and higher core counts. Laptops could be another story completely thanks to the little cores, and we should have a better idea when mobile Alder Lake comes out.

RDNA 3 will almost certainly wreck whatever Nvidia puts up against it in Perf/W, and will probably win at raw rasterization, but is less likely to win in Ray-Tracing (will probably narrow the gap) plus DLSS is still a big advantage. But even if RDNA 3 just obviously wins at just about everything, AMD probably needs to win a few generations in a row to make a noticeable dent. Nvidia can afford to lose a single generation.

The next generation of products isn't what AMD needs to worry about. If anything AMD is poised to improve their position. It's the generations that come afterwards that might be able to knock AMD down a peg. Neither Nvidia nor Intel are asleep at the wheel.

Currently, nobody outside of AMD besides close partners knows the performance of Zen 4. No leaked benchmarks, no leaked SKUs, no logs or even a hint of the SKUs are out.
That is for a part that has been sampling for almost a quarter.
Gigabyte Leak was a blemish on an otherwise stellar attempt at securing competitive advantage but still not much can be gleaned from it.

You can rest assured that board makers will only get RMB DDR5 (Splinter) to test AM5 until very last moment before they get Zen 4 .
All that is known is that the process offers 2x density, 2x efficiency and 1.25x perf (from AMD Nov 8 presentation)
The other thing that is known is that the CCD is 72.225 mm2 which brings a double of transistor count for Core+L2 (from Gigabyte and Execufix leaks, coming from SP5 design manual)

From performance perspective it's anybody's guess what AMD will get from 1.25x process frequency gain (obviously not across the board) and 2x jump in transistor count for Core+L2.
Zen iterations have delivered decent gains for the amount of XTor resources consumed.
This is about the only reason I am personally optimistic otherwise everything else is blind speculation.

However, Zen 4 is not the big leap though.
Mike Clark/Papermaster has alluded many times to a grounds up redesign every 3 years (i.e. every alternate generation with 18 months gap between generations) and Zen 5 is the next big Zen moment for AMD.
And late 2023 is not an exception but rather the expectation.
AMD leadership has been insisting on a constant cadence of 15-18 months, if we go by what we said we would be having something like this below



I find it hard to believe Zen 4 launches late 2022, that means 1 year of sampling to customers which is ridiculous it means they just sit around and let the competition knows about it and wait for one whole year.

 

[igor_kavinski]

I guess all I am saying is for the earliest AM5 boards. They should have a DDR4 option board.

Details on the Gigabyte Leak – Chips and Cheese

No mention of DDR4 there. Looks like AM5 will go all in with DDR5.

 

[jpiniero]

I find it hard to believe Zen 4 launches late 2022, that means 1 year of sampling to customers which is ridiculous it means they just sit around and let the competition knows about it and wait for one whole year.

By Late 2022, people are talking about Rapahel's release to DIY. Not Zen 4's launch. Unless AMD decides it needs to do a paper launch it's going to take months to build up enough Epyc rejects to have enough supply for a release.

 

[uzzi38]

By Late 2022, people are talking about Rapahel's release to DIY. Not Zen 4's launch. Unless AMD decides it needs to do a paper launch it's going to take months to build up enough Epyc rejects to have enough supply for a release.

Well I did whilst stating an extremely conservative estimate. I think it's more likely we'll see Genoa launch in Q3, but again Q4 is very much a conservative estimate assuming it'll be 2Q after initial shipments.

Ultimately that's based off of Milan, even though Milan could have been launched far earlier if AMD wanted to (I remember STH Patrick confirming this in an article a bit ahead of launch - that AMD could have launched Milan much closer to when Vermeer dropped and the platform would have been as ready as Rome was on it's own launch, but they decided against it).

 

[nicalandia]

Anyone that thinks that a 50C/100T Golden Cove will beat a 64C/128T Zen3 ThreadRipper Pro in Multi Threaded workload is in for a rude awakening. These are Full featured(8 Channel) EPYC with a boost with full fat L3, Golden Cove will be power constrained and will not boost past 4 Ghz on all cores.