Speculation: Ryzen 4000 series/Zen 3

[Shivansps]

Do you expect a detailed circuitry analysis? Here is a excerpt from an article by Tom's. This is not a straight port to 7nm.

'AMD chose to stick with the Vega graphics engine found in its previous-gen Picasso APUs because its hasn't optimized Navi for mobile yet, but the company chose to make several architectural enhancements when it ported Vega from the 12nm to the 7nm node. That includes reducing the number of compute units (CU) from 10 to eight, but a range of improvements led to an overall performance increase of 59% per CU (based on TimeSpy test results).'

As i said, that could mean a lot of things, for example they did not say anything about reducing ROP number, and they did. I dont need a deep analysis, but that 59% could easily be the result of freq increase and the use of DDR4-3200.

Maybe cutting rops by half and reducing CU number along with the die shrink helped to archive those high freqs. But thats my guess, and nothing they said leads me to belive the Renoir CU performs better per clock than Picasso CUs. And thats was my point when we were discussing Renoir IGP a few months ago.

Wharever they did allows for very high freqs, thats is what makes up for the CU loss.

 

[maddie]

As i said, that could mean a lot of things, for example they did not say anything about reducing ROP number, and they did. I dont need a deep analysis, but that 59% could easily be the result of freq increase and the use of DDR4-3200.

Maybe cutting rops by half and reducing CU number along with the die shrink helped to archive those high freqs. But thats my guess, and nothing they said leads me to belive the Renoir CU performs better per clock than Picasso CUs. And thats was my point when we were discussing Renoir IGP a few months ago.

Wharever they did allows for very high freqs, thats is what makes up for the CU loss.

I'm talking circuitry layout low level changes. What I'm saying is that they said architectural enhancements. This is not rearranging the number of CU, rops, etc.

You of course can believe whatever you want, but expect to continue being surprised when reality is different from your reasoning.

 

[TESKATLIPOKA]

VEGA IGP can clock pretty high, much higher than Vega GPU could even on 7nm. The question is how high can RDNA2 go, turbo in PS5 is 2.23Ghz. If the next gen APU will have 12CU and high clocks, then It would mean a serious increase in performance!

 

[DrMrLordX]

Vega20 can already clock to 2000 MHz pretty easily in Radeon VII. That was the 7nm pipecleaner for AMD, so it stands to reason that they learned a lot from that experience.

@uzzi38

I will believe that when I see it. Pity AMD hasn't shown us anything yet.

 

[TESKATLIPOKA]

Manual OC yes, but Radeon VII was not factory clocked at 2GHz, max turbo ended at 1.8GHz. And now we have 2.1GHz for APU.

 

[Ajay]

The guy in question is possibly the premier guy in leaking details... about Samsung's phones?

Uh... odd source, I know, but a guy like this has credibility and a track record, and guys like that don't really 'leak' stuff they aren't sure of.

Definitely worth trusting if you ask me.

Is he from Korea? If so, hard to imagine he has insight into Zen3. "Actual performance of the sample" must be a translation error, since it would need to be plural to be relevant.

 

[coercitiv]

"Actual performance of the sample" must be a translation error, since it would need to be plural to be relevant.

It's obviously auto-translation, "performance has risen to an alarming degree" is definitely AI humor.

 

[DrMrLordX]

Manual OC yes, but Radeon VII was not factory clocked at 2GHz, max turbo ended at 1.8GHz. And now we have 2.1GHz for APU.

The headroom was there @ stock volts on most cards. If AMD had continued iterating on Radeon VII, they probably could have gotten there.

edit: it looks like AMD is continuing to iterate on Radeon VII, since they're releaseing a Radeon Pro card based on Vega20 with full DP performance. For $2k.

 

[inf64]

Charlie "Bunny suit" D has been saying this about Zen3 for over a year now . It is a major performance jump, I'm surprised no leaks so far have come out.

 

[moinmoin]

Is he from Korea?

The post is from Weibo, a Mainland Chinese microblogging platform.

 

[CHADBOGA]

Charlie "Bunny suit" D has been saying this about Zen3 for over a year now . It is a major performance jump, I'm surprised no leaks so far have come out.

How much of the improvement is due to implementing a setup like the 3300 has?

 

[inf64]

How much of the improvement is due to implementing a setup like the 3300 has?

We can only guess how much of an impact the new shared L3 cache will have, but this will not be the only major thing in Zen3. I think that AMD redesigned the whole pipeline and especially FP co-processor, expanded further the OP cache and register files. I think that the overall int IPC jump will be bigger than Zen2's because the changes from Zen1/2 to Zen3 will be bigger than the ones from Zen1 to Zen2.

After all, Zen1, Zen3 and Zen5 are called new uarchitectures by AMD

 

[amrnuke]

We can only guess how much of an impact the new shared L3 cache will have, but this will not be the only major thing in Zen3. I think that AMD redesigned the whole pipeline and especially FP co-processor, expanded further the OP cache and register files. I think that the overall int IPC jump will be bigger than Zen2's because the changes from Zen1/2 to Zen3 will be bigger than the ones from Zen1 to Zen2.

After all, Zen1, Zen3 and Zen5 are called new uarchitectures by AMD

Keep going, I'm almost there!

 

[Panino Manino]

Of course Zen 3 will be an improvement. For what I know just fixing part of the memory and cache latency in Zen 2 would already do wonders but still, I'm conflicted if I allow myself to "dream".
I still remember the Pre Zen 1 days, the rumors about the performance. I see the target moving from Sandy Bridge level, to Ivy Bridge to maybe Haswell... I was on the pessimistic side, with Sandy Bridge. In the end exceeded expectations, and than Zen 2 that I thought would be just a tweak to buy time for Zen 3 was already enough to come above Intel. Things are a bit too good for too long for AMD, it's time for tragedy to strike again. I just want Zen 3 deliver the expected, nothing more.

 

[Thunder 57]

...Things are a bit too good for too long for AMD, it's time for tragedy to strike again. I just want Zen 3 deliver the expected, nothing more.

Why would you say that? Intel is still the Goliath in the room. If anything we should want AMD to continue to gain market share. Competition is good. Did you ever say things were "a bit too good for too long for" Intel? We saw what happened there, and it sucked. To say it's time for tragedy to strike again at the company that brought you 6/8/12/16 cores with excellent performance is absurd. What you should be wishing for is that Intel catches up, not for "tragedy" to strike AMD again. That is the stupidest thing I've read in a long time.

 

[CHADBOGA]

Why would you say that? Intel is still the Goliath in the room. If anything we should want AMD to continue to gain market share. Competition is good. Did you ever say things were "a bit too good for too long for" Intel? We saw what happened there, and it sucked. To say it's time for tragedy to strike again at the company that brought you 6/8/12/16 cores with excellent performance is absurd. What you should be wishing for is that Intel catches up, not for "tragedy" to strike AMD again. That is the stupidest thing I've read in a long time.

I don't think he is wishing for bad things to happen to AMD, he just fears it will and doesn't want to get his hopes up of AMD having an uninterrupted good run, because then it will sadden him more if AMD trips over some kind of hurdle.

 

[inf64]

Anything is possible guys, I agree that it is better to under promise and over deliver. We had no hard numbers from AMD so they had essentially not promised any specific jump in IPC or features. Same was true for Zen+ and Zen2 (Zen1 was exception as we had some goals such as the low-ball 40% figure).

 

[Thibsie]

I don't think he is wishing for bad things to happen to AMD, he just fears it will and doesn't want to get his hopes up of AMD having an uninterrupted good run, because then it will sadden him more if AMD trips over some kind of hurdle.

And that AMD has very very slight tendency to shoot itself in the foot.

 

[Ajay]

And that AMD has very very slight tendency to shoot itself in the foot.

List of CEOs

1. Jerry Sanders, 1969–2002 (founder, electrical engineer) - freaking crazy dude that somehow got AMD off the ground delivered some great products.
2. Hector Ruiz, 2002–2008 (electrical engineer) - incompetent self serving fool.
3. Dirk Meyer, 2008–2011 (computer engineer) - highly intelligent engineer, just not ready for a leadership role.
4. Rory Read, 2011–2014 (information systems) - very competent manager, reorganized AMD put good people in the right place (saved AMD, IMHO).
5. Lisa Su, 2014–present (electrical engineer) - focused, intelligent, creative, respected by rank and file, top notch leadership team

AMD is now a world class semiconductor manufacturer again (albeit, a bit on the small side compared to rivals). I think their odds of shooting themselves in the foot have greatly diminished compared to the past.

I think their biggest challenges are Intel's and Nvidia's huge brand loyalty, momentum, deeply entrenched industry ties and respect. Intel's 10nm fiasco truly gave AMD a great opening that would have been almost impossible otherwise. Nvidia is just a tough competitor - simple as that. Intel needs to get it's process development on track and really shake up it's engineering teams to stay. If they do both, they will remain a juggernaut - but that's pretty questionable right now.

That said, no company is immune to human error.

 

[DrMrLordX]

I think their odds of shooting themselves in the foot have greatly diminished compared to the past.

Hector is still available.

 

[TESKATLIPOKA]

The headroom was there @ stock volts on most cards. If AMD had continued iterating on Radeon VII, they probably could have gotten there.

Then 2.1GHz for Renoir APU might not be the limit and there will be additional headroom to OC It without adding more voltage.

 

[DrMrLordX]

Then 2.1GHz for Renoir APU might not be the limit and there will be additional headroom to OC It without adding more voltage.

I am thinking the same thing. Probably in the 2.3-2.4 GHz ballpark. Anyone saying it'll be bandwidth-limited will probably be right though.

 

[LightningZ71]

I just don't see a lot of room for improvement for AMD's APUs on the desktop. I think it was D8auer that produced the youtube video that demonstrated a whole slew of overclocking results for a 2400G with different memory sticks at different timings. When pushed from 1500 to 1600 Mhz, he saw very little performance increase. When pushing the RAM past 3200 to 3400 and trying to tighten the timings, he was also seeing very little improvement. For the APU in Renoir to exceed the TFlops in Vega 11 at 1.6Ghz, it needs to be North of 2.0Ghz. So, even if we manage to see the desktop APUs pushing DDR4 at 4266 on two channels, at just 2.1Ghz, its only really addressing the VRAM bandwidth starvation problem by roughly 30% over VEGA 11 with 3200 DDR4 and not making any significant gains on actual iGPU core video computation throughput.

I'm going to be quite shocked if desktop Renoir exceeds Picasso performance by more than 20% when the performance limit is entirely iGPU related. On titles that are CPU starved, or that were hitting thermal/power walls on Picasso, Renoir should show significant improvements.

 

[Shivansps]

I'm talking circuitry layout low level changes. What I'm saying is that they said architectural enhancements. This is not rearranging the number of CU, rops, etc.

You of course can believe whatever you want, but expect to continue being surprised when reality is different from your reasoning.

Well, i dont think anybody considers Vega at 2.1Ghz stock on an APU as not suprising. When we discussed this a few months ago we were talking about 1.7Ghz with 2.0Ghz being possible with OC. Now we are talking about 2.5Ghz as possible with OC, how this is not a suprise to you?

And im sorry, but wharever AMD did to Vega they did not announce it, the only thing we know is that they clock higher, any sustancial change to the arch would have caused at minimum a name change to "Vega 2" or something to draw attention to it, not sweep it under the rub.

 

[maddie]

List of CEOs

1. Jerry Sanders, 1969–2002 (founder, electrical engineer) - freaking crazy dude that somehow got AMD off the ground delivered some great products.
2. Hector Ruiz, 2002–2008 (electrical engineer) - incompetent self serving fool.
3. Dirk Meyer, 2008–2011 (computer engineer) - highly intelligent engineer, just not ready for a leadership role.
4. Rory Read, 2011–2014 (information systems) - very competent manager, reorganized AMD put good people in the right place (saved AMD, IMHO).
5. Lisa Su, 2014–present (electrical engineer) - focused, intelligent, creative, respected by rank and file, top notch leadership team

AMD is now a world class semiconductor manufacturer again (albeit, a bit on the small side compared to rivals). I think their odds of shooting themselves in the foot have greatly diminished compared to the past.

I think their biggest challenges are Intel's and Nvidia's huge brand loyalty, momentum, deeply entrenched industry ties and respect. Intel's 10nm fiasco truly gave AMD a great opening that would have been almost impossible otherwise. Nvidia is just a tough competitor - simple as that. Intel needs to get it's process development on track and really shake up it's engineering teams to stay. If they do both, they will remain a juggernaut - but that's pretty questionable right now.

That said, no company is immune to human error.

What this really shows is that it only takes one idiot to ruin decades of work. Easy to destroy, difficult to build is the lesson. Strangely enough, Intel is facing the same one fool nightmare.