Question AMD Phoenix/Zen 4 APU Speculation and Discussion

[uzzi38]

I can finally make this thread.

https://twitter.com/x/status/1519669375283957760

Phoenix is indeed RDNA3. My advice to everyone: treat the old APU rumours as being out of date.

 

[Abwx]

It is losing to the RTX 4080. And no, AMD memory controller sucks with JEDEC timings - it is a proven fact.

Edit - Downvotes won't change the reality:

View attachment 78195

What are those desktop CPUs tests have to do with DTR dedicated SKUs...?...
And of course a comparison using a 7700X wich clock lower than a 7950X...

FTR the tested laptop, both Intel and AMD, use DDR5 4800 and are not clocked the same as DT parts, so you came with yet another fantasy as argument.

 

[tamz_msc]

What are those desktop CPUs tests have to do with DTR dedicated SKUs...?...
And of course a comparison using a 7700X wich clock lower than a 7950X...

FTR the tested laptop, both Intel and AMD, use DDR5 4800 and are not clocked the same as DT parts, so you came with yet another fantasy as argument.

It's about the relative drop in performance on AMD compared to Intel. Dragon Range is the exact same configuration as desktop Zen 4.

I see you've learned one or two new words, but are still incapable of interpreting data.

 

[moinmoin]

Dragon Range is the exact same configuration as desktop Zen 4.

That's a claim you'll need to prove with hard data. Binning and different microcode/firmware are still a thing, even if technically exactly the same dies are used.

 

[tamz_msc]

That's a claim you'll need to prove with hard data. Binning and different microcode/firmware are still a thing, even if technically exactly the same dies are used.

Ah yes, binning and microcode will overcome the same technical limitations of the chiplet architecture and sub-optimal memory controller.

 

[Tigerick]

AMD's 6nm Rembrandt, 4nm Phoenix CPUs May Arrive On AM5

The Ryzen APU invasion is on the horizon.

www.tomshardware.com

Wow, both Rembrandt and Phoenix are coming to AM5 desktop platform. Too bad cause that's mean AMD won't price Phoenix APU aggressively...

 

[BorisTheBlade82]

AMD's 6nm Rembrandt, 4nm Phoenix CPUs May Arrive On AM5

The Ryzen APU invasion is on the horizon.

www.tomshardware.com

Wow, both Rembrandt and Phoenix are coming to AM5 desktop platform. Too bad cause that's mean AMD won't price Phoenix APU aggressively...

Why should they? Raphael is also an APU which quite likely is cheaper to produce and can be priced very competitively. Rembrandt and Phoenix might be for premium SFF and stuff.

 

[leoneazzurro]

It is losing to the RTX 4080. And no, AMD memory controller sucks with JEDEC timings - it is a proven fact.

Edit - Downvotes won't change the reality:

View attachment 78195

No, it's not "losing to the 4080". Apart the fact that both mobile 4080 and 4090 have nothing to do with the desktop versions, because the 4090 mobile is the 4080 desktop limited to 175W (150W in the notebook tested) and the 4080 mobile is the 4070Ti still limited to 175W max, the that the test suite used by notebookcheck for calculations its "score" is quite bad (no new games and the ones that are there are badly optimized for the AMD side), there is a table below where one can check the individual games scores also in newer titles and there one can see that the scores are between the 175W 4080 mobile and the higher power 4090 mobile.

In any case, the problem in Zen4 dependence on memory timing is not the "suboptimal" memory controller because it even gains more with higher memory speeds than its Intel counterparts, even with the Zen4 3D versions, the issue is the Infinity Fabric. The memory controller itself is, on the opposite, quite good, because its output scales well with the IF clocks. If it was suboptimal, it would have reached a plateau before the IF reached its optimal clock. Instead, it is quite clear that the memory controller is limited by the IF bandwidth.

Anyway, the problem there is also the OEMs using crap memory timings for their gaming laptop lines. Using a DDR5-4800 nowadays on a $3500+ gaming notebook is pure evil, considering the amount of DDR5-5600 SO-DIMM modules already available, with DDR5-6000 on the horizon.

At the moment, however, Intel 13700/13900/13950/13980HX have still in my opinion the edge in strictly CPU limited gaming situations respect to the Dragon Range, the fact is however that with the Dragon Range line the difference will be negligible in every practical situation (whereas in the past the mobile APUs from AMD suffered more in gaming because of the smaller cache) so when deciding about what to buy it should be more based on the price and configuration than pure CPU performance, especially considering that Dragon Range will have lower power consumption/heat output under load, and better application performance even at a lower power consumption.

 

[moinmoin]

AMD's 6nm Rembrandt, 4nm Phoenix CPUs May Arrive On AM5

The Ryzen APU invasion is on the horizon.

www.tomshardware.com

Wow, both Rembrandt and Phoenix are coming to AM5 desktop platform. Too bad cause that's mean AMD won't price Phoenix APU aggressively...

Wait a moment, do I read this article right? The author says himself he's referring to the Gigabyte leak from last year, even recommending "throwing a bit of salt over" but in the whole article there's no other, newer source mentioned for Rembrandt and Phoenix coming to AM5? Are we already in the slow news days?

 

[uzzi38]

Dragon Range may be worse performing in CPU bound games but it has got a very distinct edge in one area:

It's blisteringly fast at multi-core loads, beating a 130W 13900HX at just 65W in all of the workloads tested here.

 

[uzzi38]

Wait a moment, do I read this article right? The author says himself he's referring to the Gigabyte leak from last year, even recommending "throwing a bit of salt over" but in the whole article there's no other, newer source mentioned for Rembrandt and Phoenix coming to AM5? Are we already in the slow news days?

Yeah the Gigabyte leak just found it's way onto Twitter again and some bozo decided to make an article saying it's confirmed to come to desktop.

I don't think we'll see Rembrandt come to desktop. The chip was used to validate AM5 motherboards before Raphael was ready, and we still haven't seen it release to this day. There's practically no chance it releases at this point.

 

[leoneazzurro]

Dragon Range may be worse performing in CPU bound games but it has got a very distinct edge in one area:

It's blisteringly fast at multi-core loads, beating a 130W 13900HX at just 65W in all of the workloads tested here.

Yeah as said it will be slightly lower performing in CPU bound games but the difference will be indistinguishable except some rare corner cases, while on the other side app performance is seemingly very good even at that power level.

 

[tamz_msc]

No, it's not "losing to the 4080".

>The chart literally shows the Asus Zephyrus Duo with 4090 losing to both Razer Blade 16 and Lenovo Legion 7 equipped with 4080.
>"It's not losing to the 4080".

Want more evidence? Just check the graphs for X-Plane 11 - a CPU limited game. It's there in the notebookcheck article.

In any case, the problem in Zen4 dependence on memory timing is not the "suboptimal" memory controller because it even gains more with higher memory speeds than its Intel counterparts, even with the Zen4 3D versions, the issue is the Infinity Fabric. The memory controller itself is, on the opposite, quite good, because its output scales well with the IF clocks. If it was suboptimal, it would have reached a plateau before the IF reached its optimal clock. Instead, it is quite clear that the memory controller is limited by the IF bandwidth.

Whatever be the reason, AMD's memory-controller IS suboptimal. Heck, on the desktop it can't even boot 24GB modules.

 

[leoneazzurro]

>The chart literally shows the Asus Zephyrus Duo with 4090 losing to both Razer Blade 16 and Lenovo Legion 7 equipped with 4080.
>"It's not losing to the 4080".

Want more evidence? Just check the graphs for X-Plane 11 - a CPU limited game. It's there in the notebookcheck article.


Whatever be the reason, AMD's memory-controller IS suboptimal. Heck, on the desktop it can't even boot 24GB modules.

Lol, cherrypicking at its best. I explained why that chart is laughable at best, and there are of course more games running better on Raptor Lake than vanilla Zen4, and I also said that Raptor lake is better in games than Dragon Range, but that does not mean that the CPU is responsible for "making the 4090 slower than a 4080" as there is another chart a bit lower with more actual game results at different resolutions which - by only comparing to the reviews of the other notebooks involved, paint a quite different picture than the one you want to show.

Examples: Shadow of the Tomb Rider Ultra runs faster on the Zephirus than the Blade, Cyberpunk is only marginally faster on the Titan than the Zephirus, and on the latter it runs much faster than on the Blade. F1 2022 Runs much faster on the Zephirus even at med details than the Lenovo Legion Pro with the 4080.

Xplane is a redactedoptimized game, especially on Zen4 vanilla, but there are even games that run better on Zen4 , I could cherrypick too (i,.e. Horizon Zero Dawn) but I am not a redactedlike you.

Conclusion (which was the same as before): Raptor Lake mobile is better than Dragon Range at games and worse at applications, in real life no one will find differences in the experience with either.

Memory controller:
No, don't do the strawhat trick, the original sense was about the timings and performance, now you shift the discussion to memory support which has nothing to do with optimization.

It is so bad that a Zen4 without V-cache gains up to 20% going from DDR5-4800 to DDR5-6000:

AMD Ryzen 9 7950X3D Memory Scaling Benchmark | TechSpot

This while RL gains are quite lower with the same jump. Now, fanboys could open a discussion about which controller is "more optimized" when one is gaining more and the other less by using better memory, as a tech enthusiast it's easy to understand that this behaviour is simply the result of technical design choices and the way these CPUs are supposed to work and have nothing to do with a supposed "optimization" which has no meaning.



No profanity in tech or calling members fanboys.


esquared
Anandtech Forum Director

 

[trivik12]

Efficiency of the Asus is terrible. I hope we see more Dragon Range laptop reviews from other OEMs.

 

[tamz_msc]

Lol, cherrypicking at its best. I explained why that chart is laughable at best, and there are of course more games running better on Raptor Lake than vanilla Zen4, and I also said that Raptor lake is better in games than Dragon Range, but that does not mean that the CPU is responsible for "making the 4090 slower than a 4080" as there is another chart a bit lower with more actual game results at different resolutions which - by only comparing to the reviews of the other notebooks involved, paint a quite different picture than the one you want to show.

Nearly everyone who responds to me is blind it seems - they cannot see the data for what it is, and need to be spoon-fed.

Consider the models with 4090 alone in the charts. Those will be the Zephyrus Duo, XMG Neo 16 and MSI Titan GT77:

	Witcher 3	GTA V	FFXV	Strange Brigade	Dota 2	Xplane	Cyberpunk 2077	Witcher 3 V4.00	SotTR	F1 22
Zephyrus	178.6	134	169.6	333	179.7	117.8	130.6	133.9	196	114.3
XMG Neo 16	216.4	137.5	180.7	350	177.3	137	122.8	141.8	207	137.4
MSI Titan GT77	198.9	140.9	187.4	373	185.9	134.2	135	150	214	138.2

The Zephyrus comes dead last in all but two tests, those being DOTA 2 and Cyberpunk. If you want to believe that a meager 10 W difference between the 4090s (140 W vs 150 W excluding dynamic boost) is responsible for the differences, that's on you.

(Actually to think about it, Dynamic Boost should work better with the AMD CPU given its already lower power consumption while gaming, so any difference in TGP is going to end up as negligible)

This while RL gains are quite lower with the same jump. Now, fanboys could open a discussion about which controller is "more optimized" when one is gaining more and the other less by using better memory, as a tech enthusiast it's easy to understand that this behaviour is simply the result of technical design choices and the way these CPUs are supposed to work and have nothing to do with a supposed "optimization" which has no meaning.

Techspot/HWUB never did any detailed tweaking to optimize subtimings on Intel - they basically do some minor tweaking around XMP and call it a day. Intel gains are also pretty substantial, certainly much more than what HWUB shows, when properly tweaked.

 

[Abwx]

8.2

The Zephyrus comes dead last in all but two tests, those being DOTA 2 and Cyberpunk. If you want to believe that a meager 10 W difference between the 4090s (140 W vs 150 W excluding dynamic boost)

Why exclude dynamic boost, because one config has significantly higher TDP when this boost is included,?..
Not counting that this extra power help in the max FPS.

FTR the GPU TDP difference is 13% and if we take account of the GPU RAM not consuming more for the higher TDP set GPU then the actual power difference is close to 20% at the GPU level since all this extra power will be consumed by the GPU compute units, wich explain largely the 9% difference in graphic perf in favour of the 13xxxHX.

Next time get your numbers right before spilling your usual bad and sad faith in whatever AMD related thread.


Edit : All difference is due to the GPU higher TDP, there s no CPU bottleneck here since the 7945HX use something like 36-38W in games according to NBC test in Witcher 3.

 

[leoneazzurro]

Nearly everyone who responds to me is blind it seems - they cannot see the data for what it is, and need to be spoon-fed.

Consider the models with 4090 alone in the charts. Those will be the Zephyrus Duo, XMG Neo 16 and MSI Titan GT77:

	Witcher 3	GTA V	FFXV	Strange Brigade	Dota 2	Xplane	Cyberpunk 2077	Witcher 3 V4.00	SotTR	F1 22
Zephyrus	178.6	134	169.6	333	179.7	117.8	130.6	133.9	196	114.3
XMG Neo 16	216.4	137.5	180.7	350	177.3	137	122.8	141.8	207	137.4
MSI Titan GT77	198.9	140.9	187.4	373	185.9	134.2	135	150	214	138.2

The Zephyrus comes dead last in all but two tests, those being DOTA 2 and Cyberpunk. If you want to believe that a meager 10 W difference between the 4090s (140 W vs 150 W excluding dynamic boost) is responsible for the differences, that's on you.

(Actually to think about it, Dynamic Boost should work better with the AMD CPU given its already lower power consumption while gaming, so any difference in TGP is going to end up as negligible)


Techspot/HWUB never did any detailed tweaking to optimize subtimings on Intel - they basically do some minor tweaking around XMP and call it a day. Intel gains are also pretty substantial, certainly much more than what HWUB shows, when properly tweaked.

Thanks for confirming your bias and not being able to read the data (i.e. dynamic boost is limited to 15W, so it's 140+15 vs 150+25), tech reviews (Techspot actually used quite tight timings for RL on their scaling tests, and even used an higher rate RAM for RL in that review), nor reading my comments. BTW, I linked another review where there is a comparison between the Zephirus and the Strix Scar with the Raptor Lake in actual real usage scenarios which actually confirms my conclusion. Maybe it's not that "everyone is blind" but it's your problem being a fanboy who thinks to be fair and instead shits on AMD in all cases.

 

[leoneazzurro]

Efficiency of the Asus is terrible. I hope we see more Dragon Range laptop reviews from other OEMs.

View attachment 78255

Gaming "laptops" are not efficient by design, Zephirus Duo worse than all also because it has two screens which adds a lot on power consumption (not only because of the screen, but because the second screen is driven by the dGPU as well, and this increases the idle consumption). You will see quite probably other designs with better consumption, but in any cases these are beasts with 330W power adapters, and these are there for a reason.

 

[Abwx]

DP

 

[Shmee]

A reminder and warning to everyone here, that personal attacks and trolling are not allowed. For now, everyone should try to chill out a bit. If you cannot be civil, you will be infracted.

AT Moderator Shmee

 

[uzzi38]

Nearly everyone who responds to me is blind it seems - they cannot see the data for what it is, and need to be spoon-fed.

Consider the models with 4090 alone in the charts. Those will be the Zephyrus Duo, XMG Neo 16 and MSI Titan GT77:

	Witcher 3	GTA V	FFXV	Strange Brigade	Dota 2	Xplane	Cyberpunk 2077	Witcher 3 V4.00	SotTR	F1 22
Zephyrus	178.6	134	169.6	333	179.7	117.8	130.6	133.9	196	114.3
XMG Neo 16	216.4	137.5	180.7	350	177.3	137	122.8	141.8	207	137.4
MSI Titan GT77	198.9	140.9	187.4	373	185.9	134.2	135	150	214	138.2

The Zephyrus comes dead last in all but two tests, those being DOTA 2 and Cyberpunk. If you want to believe that a meager 10 W difference between the 4090s (140 W vs 150 W excluding dynamic boost) is responsible for the differences, that's on you.

(Actually to think about it, Dynamic Boost should work better with the AMD CPU given its already lower power consumption while gaming, so any difference in TGP is going to end up as negligible)

4090 mobile runs very close to minimum voltage even at 150W, even if it only seems like 10W it's actually quite a big deal. Gains with extra power should be pretty close to linear frankly.

Not saying that Dragon Range is stronger at gaming, just pointing out that 10W is legitimately pretty significant. Until someone measures how much power both CPU and GPU are pulling in these titles, I would hold off judgement.

 

[Abwx]

4090 mobile runs very close to minimum voltage even at 150W, even if it only seems like 10W it's actually quite a big deal. Gains with extra power should be pretty close to linear frankly.

Not saying that Dragon Range is stronger at gaming, just pointing out that 10W is legitimately pretty significant. Until someone measures how much power both CPU and GPU are pulling in these titles, I would hold off judgement.

It s 175W vs 155W, so that s 20W more totaly dedicated to the GPU compute units.

 

[TESKATLIPOKA]

This is only 3dmark, but It still shows that even RTX 4080 Laptop scales with higher W and RTX 4090 laptop should do better.

 

[uzzi38]

https://twitter.com/x/status/1636448352454729728

https://twitter.com/x/status/1636612998994296832

PHX2 is finally starting to show up

 

[moinmoin]

PHX2 is finally starting to show up

Imo bound to be the most interesting chip this gen considering how much out from the left field this early hybrid design by AMD comes. Hope we get more insight into it.