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!

 

[igor_kavinski]

Intel Z790 motherboards listed with support for DDR5-6800 and DDR4-5333 memory - VideoCardz.com

Intel Z790 motherboards support up to DDR5-6800 memory Here are some first details on Intel Z790 memory support. We received documents for the upcoming MSI Z790 motherboard series, including their specifications. The documents confirm that these motherboards will support JEDEC DDR5-5600 specs...

videocardz.com

In fact, neither ASUS, Gigabyte nor ASRock list the memory specs, however Biostar does (thanks to dudulab for letting us know). According to X670E Valkyrie specs, it supports up to DDR5-6000 overclocked memory.

That seems pretty low and if that is the upper limit, it may even be not easily reachable. I hope Biostar just made a sucky mobo.

 

[DrMrLordX]

That seems pretty low and if that is the upper limit, it may even be not easily reachable. I hope Biostar just made a sucky mobo.

It may not matter if IF clock can't reliably go over 3000 MHz.

 

[Kaluan]

(snip)

Except the 5,85GHz clock is stock behavior, so I'm not quite sure what high end cooling has to do with 7950X peaking at 5,85GHz on 1-2 cores in some workloads, out of the box.
I think Uzzi38's right on the money with their reply:

https://twitter.com/x/status/1563852395163095044

BTW, HXL's source also talks about this "not being the top Zen4 bin", take that as you wish.
XT3D refresh coming later next year?

 

[eek2121]

since that cpu-z screen shows 4,85 GHz clock, how does that confirm 5,8/5,85 ST clocks?
is that 4,85 all core turbo?

There are other leaks that confirm the boost clock will be 5.7 GHz (fmax is 5.85 GHz). We do not know what was going on with the system in that screenshot. Typical all core boost will likely be greater than 5 GHz, with extremely heavy loads causing the chip to dip down to the base clock, which is rumored to be 4.5 ghz for dual CCD models. As an example, CPU-Z says my clock speed is 3.575 ghz right now.

 

[Timmah!]

There are other leaks that confirm the boost clock will be 5.7 GHz (fmax is 5.85 GHz). We do not know what was going on with the system in that screenshot. Typical all core boost will likely be greater than 5 GHz, with extremely heavy loads causing the chip to dip down to the base clock, which is rumored to be 4.5 ghz for dual CCD models. As an example, CPU-Z says my clock speed is 3.575 ghz right now.

i am aware of the other leaks and i am inclined to believe them, and i dont really mean to question this particular twitter post, i just dont see connection between the screenshot and the other text, since the screen shows different clocks. One would expect the screen to be there to support what is being said there, but that does not seem to be the case - unless i am missing something.

 

[exquisitechar]

5.85 GHz fmax is exactly as I thought, Angstronomics nailed it again. I'm more curious about all core clocks at stock and all core overclocks. Cooling will matter a lot now.

 

[nicalandia]

Yeah, it ain't gonna double deleted there...

GB5 does not care about Memory bandwidth. Thats why an OC Alder Lake Celeron with 2C/2T can match a 8C/16T Alder Lake.





I fully expect That 7950X double the crypto score of the 13900K

Quoting profanity is the same as posting it. Not allowed.
admin allisolm

 

[nicalandia]

7950X has reach Speed parity with the 13900k, the 13900k has thread parity with 7950X. Both have reworked Memory subsystems. This is a first for both companies since a decade ago. Now its a matter of how apps and games behave and respond to their very different uArch.

 

[Markfw]

7950X has reach Speed parity with the 13900k, the 13900k has thread parity with 7950X. Both have reworked Memory subsystems. This is a first for both companies since a decade ago. Now its a matter of how apps and games behave and respond to their very different uArch.

except one runs at about twice the power level of the other and requires extreme cooling.

 

[Exist50]

except one runs at about twice the power level of the other and requires extreme cooling.

Your source for power consumption numbers between the two?

 

[Markfw]

Your source for power consumption numbers between the two?

170 watt for the 7950x is widely advertized, even by AMD. The same for the 350 watt for Raptor lake. If you don't believe that, lets wait for official reviews, I will not bother to spend the time to look it up for you.

 

[nicalandia]

except one runs at about twice the power level of the other and requires extreme cooling.

Yes, both have reach that Ultra High End Performance in by taking difference path. They are different Beasts at stock. Enthusiasts will push them further. I suspect that the 7950X will have more room.

 

[Exist50]

170 watt for the 7950x is widely advertized, even by AMD. The same for the 350 watt for Raptor lake.

170W is advertised as the base TDP for the 7950x, while that 350W is a rumored high power mode for some motherboards we have no reason to believe is reflected in any of these tests. Boost power to boost power, the rumors are 230W vs ~250W.

If you don't believe that, lets wait for official reviews, I will not bother to spend the time to look it up for you.

Yes, I look forward to you completely ignoring you ever made this claim, as this pattern usually goes.

 

[Karnak]

170 watt for the 7950x is widely advertized, even by AMD. The same for the 350 watt for Raptor lake. If you don't believe that, lets wait for official reviews, I will not bother to spend the time to look it up for you.

170W is only the TDP (like 105W for AM4) and not the acutal PPT (like 142W for AM4, for AM5 it's 230W now).

350W for RTL is only some sort of extreme mode and not the actual PL2 (that's 253W if rumors are true).

So it's more like 230W vs. 253W.

 

[nicalandia]

170W is only the TDP (like 105W for AM4) and not the acutal PPT (like 142W for AM4, for AM5 it's 230W now).

350W for RTL is only some sort of extreme mode and not the actual PL2 (that's 253W if rumors are true).

So it's more like 230W vs. 253W.

Lets make this simple. A 4.8 Ghz All Core 5995X was pulling 700 Watts and breaking World Records in CBR23 on Water Cooling. A quarter of that is 175 Watts. There is absolutely no way a stock 7950X can pull more than that.

 

[Abwx]

All depend where did AMD put the bar for their claimed >35% better perf at >25% better perf/Watt (in Cinebench R23).

The improvement is much easier to get at the lowest possible power as well as using 142W PPT as basis for the 5950X, this way they can somewhat inflate the perf/watt numbers compared to using real measurements on the stated load.

In this logic the 7950X should score a little more than 35400 pts at 153W, and thats the most likely target.

If AMD used instead a rating of 170W PPT for the 7950X then it should score something like 39200 pts at said 170W for the >25% better perf/watt claim to be accurate, that s quite big perfs numbers and are way less probable.

 

[PJVol]

GB5 does not care about Memory bandwidth

Really?

 

[exquisitechar]

I wouldn't expect Raphael to be much more efficient than Raptor Lake at stock, but we'll see. They're pushing it pretty hard with the presumed 230W PPT.

 

[igor_kavinski]

They're pushing it pretty hard with the presumed 230W PPT.

I think Raphael's best power efficiency will be at 170W. Beyond that, people will just be burning loads of power for a little bit of improvement.

 

[Abwx]

I wouldn't expect Raphael to be much more efficient than Raptor Lake at stock, but we'll see. They're pushing it pretty hard with the presumed 230W PPT.

7950X@153W will score in Cinebench as well as a 13900K@250W, at 230W it should be at the same level than a 13900K@350W.

 

[igor_kavinski]

at 230W it should be at the same level than a 13900K@350W.

I'm skeptical that pumping more power into Zen 4 will make it perform much better. But let's see. The real issue will be taming the extreme heat.

 

[Abwx]

I'm skeptical that pumping more power into Zen 4 will make it perform much better. But let's see. The real issue will be taming the extreme heat.

7950/7900X have two chiplets to distribute heat, so they can be pushed easily at 170W PPT, not so much for the 7700X wich will surely be at substancialy lower PPT.

 

[Markfw]

170W is advertised as the base TDP for the 7950x, while that 350W is a rumored high power mode for some motherboards we have no reason to believe is reflected in any of these tests. Boost power to boost power, the rumors are 230W vs ~250W.

Yes, I look forward to you completely ignoring you ever made this claim, as this pattern usually goes.

If you look at all the responses since yours, its obvious that everyone thinks it will be way different than 230 to 250. But you always have to try and dispute anything I say.

 

[FangBLade]

Tomorrow is presentation? At what time in europe?

 

[Kaluan]

I wouldn't expect Raphael to be much more efficient than Raptor Lake at stock, but we'll see. They're pushing it pretty hard with the presumed 230W PPT.

Bruh, 8 Alder Lake P cores alone pull more power than all 16 Zen3 cores/5950X PPT + 88W ("230W PPT"). And I think we can expect Raptor Lake P cores to be even thirstier.

We also don't really know that much about how well AMD's custom TSMC N5 node scales with more power. It is a new node and it is likey tuned to hell for high performance.

Only one of the chipmakers is pushing their SKUs silicon hard, has been for half a decade now, and it ain't AMD.