Info 64MB V-Cache on 5XXX Zen3 Average +15% in Games

[Kedas]

Well we know now how they will bridge the long wait to Zen4 on AM5 Q4 2022.
Production start for V-cache is end this year so too early for Zen4 so this is certainly coming to AM4.
+15% Lisa said is "like an entire architectural generation"

 

[Hitman928]

Where are most of those gains though? Zen3's biggest gains were in avx2, especially aes operations, and single theaded benchmarks that didn't live well in zen2's smaller CCX L3. Not everything lives in AVX2 world, and in highly threaded, low crosstalk situations, the 8 core ccx isn't really a factor. If you check Servethehome's teview, they note the trade off of higher power draw for higher clocks in like for like parts as well.

Yes, in the hpc loads, you saw bigger uplifts. In everything else, where a lot of the rest of the market lives, the gains s were more modest, came from improvements in memory throughput due to clock sync improvements, and other tweaks outside of the core.

I'm not taking a dump on zen3. It's certainly a nice improvement over zen2, but, some of its biggest gains are quite situational, and in desktop and mobile situations, it's MT gai s are far more modest.

This is just not true. If you look at the SPECint rate results (integer heavy results, obviously), the 5950x was 33% faster on average over the 3950x. Both are 16 core CPUs with 105W TDP. Equalizing for clock speed, you get around 15% average uplift. Of course some work loads don't gain much and some gain a lot, but we're talking averages and on average, Zen3 is a good step up over Zen2 in integer and fp. Where you usually don't see much improvement is in memory bound and IO situations, which is to be expected, as well as a few particular applications here and there for whatever reason (rendering seems to be a particular 'weak' point in terms of generational uplift).

My company just went through testing Rome vs Milan for a new server installment. On average, we saw about the same as Phoronix and Anandtech. In our particular application that is most critical for us, it was actually close to 25% uplift over Rome core for core at the same TDPs. If Milan really only achieved 5% uplift at the same TDP, it would be a real stinker and you would be able to source Milan servers instead of having to wait through the crazy lead times they currently have because they are in high demand. The added power from Zen3 is typically more than made up for by the increased performance so you still get a decent bump in perf/w for even better perf.

 

[LightningZ71]

For our work, for like parts with the same tdp, we saw almost 20% improvement for a noracle increase in actual power draw. Both parts were reasonably compliant with tdp, but zen3 pushed it right to the edge. When you made an effort to hold them both to the same actual power usage, zen3 only gave us just under 13% more performance. A gain to be sure, but not magic.

 

[Gideon]

For our work, for like parts with the same tdp, we saw almost 20% improvement for a noracle increase in actual power draw. Both parts were reasonably compliant with tdp, but zen3 pushed it right to the edge. When you made an effort to hold them both to the same actual power usage, zen3 only gave us just under 13% more performance. A gain to be sure, but not magic.

That's still very impressive considering the same process node and packaging (and higher clocks in the I/O die).

 

[Hitman928]

For our work, for like parts with the same tdp, we saw almost 20% improvement for a noracle increase in actual power draw. Both parts were reasonably compliant with tdp, but zen3 pushed it right to the edge. When you made an effort to hold them both to the same actual power usage, zen3 only gave us just under 13% more performance. A gain to be sure, but not magic.

Looks like your 13% isn't far from the 15% I gave and is still a significant step up from the 0-5% you initially proposed. As always, particular work loads will vary so let's call it 15% +- 2% on average when limited to the same power consumption.

 

[LightningZ71]

In fairness, my original focus was on mobile (which is more what I deal with on the daily thee days). In mobile, cezanne doesn't justify te price premium that it often has over an equivalent lucienne system that's feature comparable. If it was about games, then sure, but, for productivity in thin and lights and 2 in 1s, it's hard to beat 5700u and 5500u systems for the money.

 

[Zepp]

when you compare mobile chips, the MT uplift from Zen3 over Zen2 is barely there. Witness the 5700u vs. the 5800u in benchmarks. They are restricted to similar power envelopes and are in similar environments, and the 5800u has a unified CCX with double the L3 cache, yet, the gains in MT benchmarks are often under 5%, and even regress in one or two cases. Zen3 was a single thread improvement at the cost of increased power draw. That power draw increase comes back to bite it in high count MT tests as it hits power and thermal limits. There's also likely a wall with respect to Ram bandwidth as both have the same capabilities there.

I also recall reading that the zen3 core design traded some SMT efficiency for increased single thread performance.

 

[gdansk]

I also recall reading that the zen3 core design traded some SMT efficiency for increased single thread performance.

"Trading"? It's more an inevitable consequence of having better utilization of existing resources without increasing execution resources. If one thread can take advantage of more instruction level parallelism then those resources are not available for the other thread to use.

It's more related to the decreased sustained clock rates Zen 3 achieves at the same power envelopes.

 

[Joe NYC]

AMD had this staged "interview" talking about Zen, while not announcing any new Zen products for the year.

On top of it, AMD sliced this interview into segments and is running the segments of this interview as YouTube ads (spending money).

I don't see this as a winning strategy:

 

[Ranulf]

Eh, I thought that marketing interview had one of them saying that the 3D chips were coming out soon?

 

[Joe NYC]

Eh, I thought that marketing interview had one of them saying that the 3D chips were coming out soon?

It needs to come out this week. After that, it will not matter very much.

 

[gdansk]

It needs to come out this week. After that, it will not matter very much.

Why do the chips need to come out now? I don't follow.

 

[itsmydamnation]

It needs to come out this week. After that, it will not matter very much.

righto ...... its funny how people like you crop up everytime something new from someone else is announced.

 

[maddie]

Why do the chips need to come out now? I don't follow.

Didn't you get the memo yet? It's because the world is stopping this week. No more advances will be allowed.

 

[HutchinsonJC]

AMD had this staged "interview" talking about Zen, while not announcing any new Zen products for the year.

We watched this on page 28.

 

[Joe NYC]

Why do the chips need to come out now? I don't follow.

Because by being late by a month or 2 with the chip with which AMD could claim performance leadership, AMD is going to lose this performance leadership for a year... So the damage is 6 to 12x the multiple of the time of delay.

And AMD will have to dig itself out of the deficit with expensive 5nm chip as opposed to coasting, printing money with 7nm chip.

 

[positivedoppler]

It would be foolish to release Z3D before all the win11 bugs are sorted out and performance is as optimal as possible. They will get one round of bench upon release and not all reviewers re-bench after a bug fix. You rather delay launch by a month for a product that will be out for over a year.

 

[Joe NYC]

We watched this on page 28.

What's new is that AMD marketing thinks it was so good that they are running segments of the interview as ads on YouTube.

It would have been just a little bit cringe if it was in the vacuum, but much worse at the time Intel is taking performance crown from AMD, the 2 gentlemen are announcing that AMD is missing the boat.

 

[Joe NYC]

It would be foolish to release Z3D before all the win11 bugs are sorted out and performance is as optimal as possible. They will get one round of bench upon release and not all reviewers re-bench after a bug fix. You rather delay launch by a month for a product that will be out for over a year.

Yes, that's the only saving grace - that releasing Zen 3D in middle of Windows 11 fiasco could overshadow the merit of the processor.

But it's not clear if AMD was ready to go and this is a strategic delay or if AMD dropped the ball and was not ready - and was just incidentally saved by the Windows 11 fiasco.

Edit: There is an AMD Datacenter event next week, during which some people expect that Milan X may be announced along with CDNA 2. (Milan X that would unlikely be used by anyone to run Windows 11...)

 

[gdansk]

Because by being late by a month or 2 with the chip with which AMD could claim performance leadership, AMD is going to lose this performance leadership for a year... So the damage is 6 to 12x the multiple of the time of delay.

And AMD will have to dig itself out of the deficit with expensive 5nm chip as opposed to coasting, printing money with 7nm chip.

If they're losing performance leadership for a year I don't see how launching it two months earlier is any better. Given the novel nature of Zen3D it seems reasonable to launch it when it is ready. The AM4 install base is pretty big at this point and some people would settle for "better MT", "nearly as good", or "more efficient" if they could reuse their existing motherboard.

 

[Joe NYC]

If they're losing performance leadership for a year I don't see how launching it two months earlier is any better. Given the novel nature of Zen3D it seems reasonable to launch it when it is ready. The AM4 install base is pretty big at this point and some people would settle for "better MT", "nearly as good", or "more efficient" if they could reuse their existing motherboard.

With Zen3D, it would be a tie. Intel would not have regained performance leadership.

So now, AMD is going to be in a hole, performance wise, and if Zen 3D just barely ties it Alder Lake a month or 2 from now, it will not matter, Zen 3 completely lost its relevance.

AMD will have to take a hit on consumer ASPs for a full year as a result, and hope that 5nm will turn the tide.

The goldmine of being able to sell performance leading product on a plentiful TSMC 7nm node, for a full year, was ruined.

 

[UsandThem]

With Zen3D, it would be a tie. Intel would not have regained performance leadership.

So now, AMD is going to be in a hole, performance wise, and if Zen 3D just barely ties it Alder Lake a month or 2 from now, it will not matter, Zen 3 completely lost its relevance.

AMD will have to take a hit on consumer ASPs for a full year as a result, and hope that 5nm will turn the tide.

The goldmine of being able to sell performance leading product on a plentiful TSMC 7nm node, for a full year, was ruined.

That whole proclamation is a little heavy the dramatics, no?

Even if Zen 3 loses the performance crown, it's not like all Zen 3 CPUs will suddenly disappear from the face of the earth.

 

[Markfw]

That whole proclamation is a little heavy the dramatics, no?

Even if Zen 3 loses the performance crown, it's not like all Zen 3 CPUs will suddenly disappear from the face of the earth.

The reviews are not out yet, BUT, the efficiency crown will still most likely be AMD and Zen3. So what again is ruined ?

For years, Zen has been at the forefront. When Zen 3 came up, all performance and efficiency went to AMD. When Rome came out, the server world went to AMD/EPYC (as far as performance and efficiency) I think they even had the mobile world, but I am not as much up on that.

So all this MIGHT change is a slight game win for ADL. We don't even know squat about applications.

 

[UsandThem]

The reviews are not out yet, BUT, the efficiency crown will still most likely be AMD and Zen3. So what again is ruined ?

Even if it turns out it isn't, it's not the end of the world for AMD.

The whole posts just seemed to be way too dramatic / doom and gloom. It's not like all the Zen 3 CPUs will all self destruct or something if they were suddenly in 2nd place.

 

[Rigg]

With Zen3D, it would be a tie. Intel would not have regained performance leadership.

So now, AMD is going to be in a hole, performance wise, and if Zen 3D just barely ties it Alder Lake a month or 2 from now, it will not matter, Zen 3 completely lost its relevance.

We'll see if Intel regains performance leadership (and in what specific workloads it may or may not have an advantage) in a few days. We'll also see what caveats there are in terms of platform cost (DDR 4 vs DDR 5). We don't know how good intel's launch availibility will be. We don't know how AMD will respond with price cuts.

We do know DDR5 is expensive. We do know Z690 is expensive. We do know that running a brand new version of Windows can be problematic. Those could all be potential dealbreakers for people looking to upgrade.

AMD will have to take a hit on consumer ASPs for a full year as a result, and hope that 5nm will turn the tide.

The goldmine of being able to sell performance leading product on a plentiful TSMC 7nm node, for a full year, was ruined.

AMD rebuilt it's reputation with the first 3 generations of Ryzen despite being at disadvantage in single thread and gaming. AM4 has a huge a userbase that could be tempted into upgrading to Zen3 at a discount or Zen3D with similar performance to ADL. Lets wait for benchmarks before hitting the panic button. A return to zen 2 pricing would be perfectly fine with me. Competition is good for consumers. AMD's consumer desktop margins are of no concern to me.

 

[Joe NYC]

The reviews are not out yet, BUT, the efficiency crown will still most likely be AMD and Zen3. So what again is ruined ?

For years, Zen has been at the forefront. When Zen 3 came up, all performance and efficiency went to AMD. When Rome came out, the server world went to AMD/EPYC (as far as performance and efficiency) I think they even had the mobile world, but I am not as much up on that.

So all this MIGHT change is a slight game win for ADL. We don't even know squat about applications.

I look at it differently: Intel is throwing 241 Watts at the problem (in other words, going to ridiculous lengths) to regain performance crown (because Intel has found out what it is like to be an also run).

What is AMD doing in the meantime?

Worse than nothing. Making a video, announcing that AMD is not going to fight for performance leadership. And then paying money to Google/YouTube making sure everyone knows that AMD missed the boat as far as having a competing processor against Alder Lake.

There was a perfect window to launch Zen3D, and AMD is missing it.

By the time CES comes around, watching AMD announce Zen3D will be a snooze fest. For most viewers like Joe Biden at the climate change speech - hard to keep the eyes open... And for a good reason.