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!

 

[Hitman928]

I guess Zen 4 gets a bad rep due to "stock" PPT levels AMD set. The Eco mode and manual tuned results are really really good. This bodes extremely well for mobile chips (both Phoenix Point and Dragon Range).

Edit:

der8auer has some great results with de-lidding :

Temps went down by more than 20C with direct die cooling! Crazy stuff.

That's pretty crazy. If AMD could figure out a better way to do heat spreader attach, they could actually get a little bump in performance and a significant increase in efficiency. They're already using solder so not sure what more they could do at high volume except maybe tighten up tolerances and improve their soldering to decrease voids. Nothing will beat direct die, but maybe they could figure out a way to improve things.

 

[igor_kavinski]

W T H???

AMD made a big booboo with their TIM. They could have improved the thermals for their flagship at the very least. Don't like this at all.

 

[Hitman928]

W T H???

AMD made a big booboo with their TIM. They could have improved the thermals for their flagship at the very least. Don't like this at all.

What do you suggest they use as a substitute TIM?

 

[nicalandia]

AMD made a big booboo with their TIM. They could have improved the thermals for their flagship at the very least. Don't like this at all.

Hey more power to you as a Custom Builder. Lower the thermals 10-15 degrees should be doable pretty easy.

 

[igor_kavinski]

I bought AMD Shares right after the Bulldozer Fiasco. I've been sitting them until now. I will wait until Zen5 and Sell.

And then you will never have to work again?

 

[Hitman928]

Hey more power to you as a Custom Builder. Lower the thermals 10-15 degrees should be doable pretty easy.

Yeah, I'm perfectly happy with my 5900x for now, so I'll probably skip Zen 4, but if Zen 5 (or whatever Intel has if they get something competitive out) has the same results, I'll be very tempted to run bare die for the first time in a long time.

 

[igor_kavinski]

Nothing will beat direct die, but maybe they could figure out a way to improve things.

They better. 7950X users deserve the best!

 

[igor_kavinski]

Hey more power to you as a Custom Builder. Lower the thermals 10-15 degrees should be doable pretty easy.

I don't wanna risk ruining a CPU through delidding.

 

[igor_kavinski]

What do you suggest they use as a substitute TIM?

They have people who get paid more than me to research that. Seems they didn't have enough time. Or weren't given enough time.

 

[inf64]

They better. 7950X users deserve the best!

I think that manual undervolting will do the trick in most cases coupled with great AIOs. 7950X thermals react well to undervolting so it's an easy fix for potential 7950X buyers.

 

[In2Photos]

I think that manual undervolting will do the trick in most cases coupled with great AIOs. 7950X thermals react well to undervolting so it's an easy fix for potential 7950X buyers.

Agreed. Delidding is a bit extreme for most users even with the extreme results. If it were as simple as changing out the retention mechanism like with 12th gen for 5-7C it might be more mainstream.

 

[Hitman928]

They have people who get paid more than me to research that. Seems they didn't have enough time. Or weren't given enough time.

This is a naive statement, especially in light of your original response that AMD made a mistake with their TIM. Is anyone using a better TIM? What research is out there that's even close to mass market viability? How long should AMD delay their launch waiting for a better TIM that is viable? What are the opportunity costs related to doing so? Can you answer any of these questions to support that AMD made a mistake with their TIM?

 

[nicalandia]

And then you will never have to work again?

I bought, a little more than 1,000 shares at $1.86

 

[Topweasel]

They have people who get paid more than me to research that. Seems they didn't have enough time. Or weren't given enough time.

Or solutions that might work better are cost prohibitive to include in a volume solution. Keep in mind Intel went thermal paste for a long time to reduce costs. AMD improved their solder solution, tightened the heat spreader. Eventually what we can do, isn't a reasonable solution for a manufacturer. Some of it is cost, some of it is validation, some of it is on the failure rate of implementation.

It would be one thing with the paste and us knowing it was a limiting factor. If you can't come up with a reliable solution with little effect on price, we can't really say there is a solution.

 

[H T C]

A couple of stupid questions:

1 - Any chance one can change the temp for TJmax in BIOS? NOT to a higher number, but to a lower one. For example: 90ºC, instead of 95ºC. The idea is to save power and have better efficiency.

2 - Efficiency will be the same regardless of the cooler used? For example: if i use a Noctua NH-C14 instead of an AIO behemoth like the one Steve from GN used. I know the temps the CPU will reach will be the same but, since the CPU is supposed to boost to lower numbers due to the cooler having much worse capability, won't it also be much more efficient like this?

 

[inf64]

A couple of stupid questions:

1 - Any chance one can change the temp for TJmax in BIOS? NOT to a higher number, but to a lower one. For example: 90ºC, instead of 95ºC. The idea is to save power and have better efficiency.

2 - Efficiency will be the same regardless of the cooler used? For example: if i use a Noctua NH-C14 instead of an AIO behemoth like the one Steve from GN used. I know the temps the CPU will reach will be the same but, since the CPU is supposed to boost to lower numbers due to the cooler having much worse capability, won't it also be much more efficient like this?

1) I don't think so, but who knows, maybe some motherboards will have this option.

2) The CPU will hit lower clocks due to more frequent throttling, and this will result to possibly lower power draw (and somewhat lower performance of course). I suspect that it will be much easier for you to simply play with undervolting as this will be a go-to method to even improving the default boost behavior due to lower power draw and lower temps.

 

[naad]

A couple of stupid questions:

1 - Any chance one can change the temp for TJmax in BIOS? NOT to a higher number, but to a lower one. For example: 90ºC, instead of 95ºC. The idea is to save power and have better efficiency.

2 - Efficiency will be the same regardless of the cooler used? For example: if i use a Noctua NH-C14 instead of an AIO behemoth like the one Steve from GN used. I know the temps the CPU will reach will be the same but, since the CPU is supposed to boost to lower numbers due to the cooler having much worse capability, won't it also be much more efficient like this?

No need for TJmax adjustment, and likely no motherboard will support that, just lower the PPT by a bit and you won't be hitting 95C, assuming your cooling isn't anemic.

 

[DiogoDX]

I think that manual undervolting will do the trick in most cases coupled with great AIOs. 7950X thermals react well to undervolting so it's an easy fix for potential 7950X buyers.

Brazilian review did this (~13 min mark):

7700X
Stock - 95 C - 5245Mhz all core - 140W
-0.100Mv - 87 C - 5360Mhz all core - 130W

But for the temps changing the TDC,EDC and PPT was better (~12 min mark).

76 C - 5200Mhz all core - 112W

 

[nicalandia]

Anandtech doing Gods Work and bringing 480P gaming performance.


From Anandtech: "In Civilization VI at 480p, we see the Ryzen 7950X and 7600X decimate the competition."

 

[gdansk]

Gamers with 5800X3D probably feel pretty good right about now.

 

[TESKATLIPOKA]

Reviews just confirmed what I was saying after the announcement about worse power efficiency at stock settings.
Now, I won't just say I am not impressed, but I will directly say I am disappointed by the power efficiency of these Zen 4 SKUs at stock settings.
On the other hand, I am very impressed with power consumption at limited TDP.
By using ECO mode, both 7700x and 7950x will lose only 4-5% of performance, but the power limit is back to what their predecessors had(88W and 142W).
I really have to question, If this tiny increase in performance was worth It. In my honest opinion, It was not.

Thanks to Computer Base, we can see some interesting observations about future mobile SKUs.

Model	R9 6900H	R7 7700x	R9 7950x	i9-12900k	R7 7700x	i9-12900k	R7 7700x	R9 5900x	R9 5950x	R9 7950x	R9 7950x
Power limit	45 W	45 W	45 W	65 W	65 W	88 W	88 W	88 W	88 W	65 W	88 W
Performance	100 %	122 %	126 %	128 %	145 %	146 %	155 %	157 %	170 %	194 %	229 %

If you want to buy a 16C32T Dragon Ridge, then you should choose a laptop with at least 65W sustained, at 45W It's barely faster than 7700x at 45W.
7700x at 45W is "only" 22% faster than R9 6900H, but It's 45% faster at 65W, while the energy efficiency is still the same as Rembrandt, that's very good.
Keep in mind that this is based on desktop chips, which have a separate IO die, and are not binned chips.

Now the biggest question is what to buy next year. Phoenix or Dragon Ridge? A very hard question.

 

[nicalandia]

Gamers with 5800X3D probably feel pretty good right about now.

The 7600X beats them overall... Only on certain games like MS Flight Simulator is where the 5800X3D is untouched and will remain untouched until the 7800X3D

 

[deasd]

7950X only 208 watts peak in Prime95??

https://twitter.com/x/status/1574390102490505217

too much reviews to read. what amaze me is the AVX512 perf, and here's one of AVX512 implemented realworld productivity.

https://twitter.com/x/status/1574426064415334401

AMD IN TROUBLE!!!

 

[moinmoin]

Sure, but they are getting a lot of bad publicity because of the high temperatures and how hard Zen 4 is to cool. I don't think the single digit % performance improvement is worth that.

Yeah, anything at sane per core power levels is going to be awesome, and obviously completely dominant in MT. They just really stretched for that last 5%.

That's pretty crazy. If AMD could figure out a better way to do heat spreader attach, they could actually get a little bump in performance and a significant increase in efficiency.

I'm getting the feeling AMD just increased the TDP for the heck of it, as Intel creates the market and AMD like to make itself confortable in there. And for better or worse DIY desktop is all about absolute performance regardless of consumption. So for AMD all the downsides of Zen 4 equal opportunities and headroom for improvements in future gens ("less heat!", "more perf/watt!" etc.).

I for one will just use U series chips and ECO mode by default and be done with all that.

 

[naad]

Reviews just confirmed what I was saying after the announcement about worse power efficiency at stock settings.

Now the biggest question is what to buy next year. Phoenix or Dragon Ridge? A very hard question.

Ask yourself this: "do i want a 12 pounder desktop replacement or do I want a 900 gram bread knife $1400 starting"