AMD Hybrid Phoenix APU Comes With Performance and Efficiency Cores

[GFreeman]

According to the latest leak, AMD's upcoming Phoenix accelerated processing units (APUs) could feature a hybrid design, featuring Performance and Efficiency cores. While there are no precise details, the latest AMD processor programming guide, leaked online, clearly marks these as two types of cores, most likely standard Zen 4 and energy-efficient Zen 4c cores.

These two set of cores will features a different feature set, and the latest document gives software designers guidelines. Such hybrid CPU design, similar to ARM's BIG.little architecture, will allow AMD to be more competitive with Intel's similar P- and E-core design, allowing it to achieve certain performance levels while also maintaining power efficiency.



Last week, AMD's alleged Phoenix 2 APU has been spotted with such hybrid design, but the recent leak suggest that such similar, or same design, could be coming with AMD Phoenix APUs that are coming this year. AMD's Zen 4c cores were only mentioned with AMD's EPYC "Bergamo" CPUs, so it is quite interesting to see them implemented in other market segments as well.

View at TechPowerUp Main Site | Source

 

[Dristun]

I think frankly the biggest effect of this going to be not on efficiency (they're already more efficient lol) but in the ability to claim same number of cores as Intel has in marketing materials.

 

[RamiHaidafy]

It's awesome that both the performance and efficiency cores will be based on Zen 4.

Would love to see a 2P+4E super low power SoC for tablet/hybrid devices. Maybe even one for a future Surface Neo - but that's double the wishful thinking.

 

[AnarchoPrimitiv]

I think frankly the biggest effect of this going to be not on efficiency (they're already more efficient lol) but in the ability to claim same number of cores as Intel has in marketing materials.

You're probably right....man, doesn't stink that the vast majority of consumers are willfully ignorant and convinced by the thinnest marketing? Makes you realize that the success of certain companies is not due to them necessarily having a superior product.

 

[phanbuey]

It's awesome that both the performance and efficiency cores will be based on Zen 4.

Would love to see a 2P+4E super low power SoC for tablet/hybrid devices. Maybe even one for a future Surface Neo - but that's double the wishful thinking.

Sort of... it's pretty obvious from the 7950x3d that there's not a hardware scheduler, so there might be some issues with these chips and scheduling to the weaker cores. The fact that the cores are the same probably has to do with that (no scheduler) more than anything.

 

[Squared]

It'd be nice to learn if these 4c cores are more about power efficiency or die-area efficiency. Intel's e-cores (and I suspect most e-cores) are at least as much about die-area efficiency as power efficiency.

Which makes sense, since Intel designed their p-cores to be power efficient as well, but a lot of their die area goes to resources that enable high clock speed, and once a few cores reach a high clock speed there's not enough power or cooling for the others to reach it. Smartphones have the same problem, only they reach their power and thermal limits much more easily, which probably explains why many smartphone processors have had e-cores for years.

 

[Deleted member 185088]

I heard somewhere that those zen 4c cores are just the same cores with less cache.

 

[Camm]

I heard somewhere that those zen 4c cores are just the same cores with less cache.

Correct. This probably makes them mostly a drop in replacement, Windows Zen scheduler already has a concept of preferred cores. Going off the package difference between Epyc Zen 4 / Zen 4c (and ignoring if theres other downstream issues by increasing core counts), you can expect to roughly fit 25% more cores in the same package space which would give you a 2P+10E part at best.

I doubt we'd get this, and instead the cutting of cache being just a power improvement where for most workloads most users wouldn't notice the difference, especially on lower end parts.

 

[GreiverBlade]

to me the point i pick up : AMD do not need P/E core hybrids to compete, but use them for where it matter...

APU or SOC is where hybrids matters, tablets, AIO, HTPC ... et caetera

 

[ThrashZone]

I think frankly the biggest effect of this going to be not on efficiency (they're already more efficient lol) but in the ability to claim same number of cores as Intel has in marketing materials.

Hi,
Thread count really
I wouldn't call single thread a core especially one that turbos at 4.3 lol
AMD issue is they have the thread count issue is frequencies are lower on real cores because all 16 are real cores/ performance cores not e-threads.

 

[kapone32]

The only reason AMD is doing this is to get back to the top of Cinebench and other synthetic benchmarks. The heat will be an issue unless these are seriously underclocked.

 

[ThrashZone]

The only reason AMD is doing this is to get back to the top of Cinebench and other synthetic benchmarks. The heat will be an issue unless these are seriously underclocked.

Hi,
At 16 cores amd is not far behind and I'm not sure why they would bother for this big/little stuff on desktop.

 

[Wirko]

Anyone remember this old rumour about pairs of cores?

AMD Readying 16-core "Zen 4" CCDs Exclusively for the Client Segment with an Answer to Intel E-cores?

AMD already declared the CPU core counts of its EPYC "Genoa" and "Bergamo" processors to top out at 96 and 128, respectively, a core-count believed to have been facilitated by the larger fiberglass substrate of the next-gen SP5 CPU socket, letting AMD add more 8-core "Zen 4" chiplets, dubbed CPU...

www.techpowerup.com

Maybe AMD continued to work on that concept, which we'd call "clusters" of cores today. One Zen 4 core plus one or two Zen 4c cores in a cluster with shared L2 cache may be a good idea - that depends on many implementation details of course. For one, the scheduler can quickly move execution of a software thread between the two types of cores without wasting the contents of L2.

 

[Psychoholic]

Nice,

I Hope they never go 100% Hybrid, and leave the option to go with a relatively high number of full size cores in the future.

 

[ThrashZone]

Hi,
Intel's desperate
AMD not so much
Why depend on win-11 to optimize big little nonsense
Most normal people reduce background crap anyway so little is not needed.

 

[Chrispy_]

If these two different core types support different feature sets, it means that the Zen4C cores aren't just cache-less Zen4 cores, but something new and purpose-made.

For a 15W ultraportable, I'd rather have 2P+8E over regular 4C/8T stuff we're accustomed to seeing.

 

[Minus Infinity]

A Phoenix hybrid 8P+8E would be an instant buy for me if available for desktop. Zen 4C cores will smoke Gracemont and AMD should have less trouble with schedulers as the 4c cores will look just like regular cores.

The only reason AMD is doing this is to get back to the top of Cinebench and other synthetic benchmarks. The heat will be an issue unless these are seriously underclocked.

Synthetic benchmarks! Raptor Lake usually easily beats Zen 4 in scientific workloads, Matlab, Mathematica, COMSOL, Chemistry etc. 13700 annihilates my 5800X in COMSOL, far from synthetic. The e-cores are helping productivity a lot for Intel and v-cache does sweet FA for productivity.

I would bet an AMD 8 Zen 4 + 8 Zen 4c hybrid would easily beat 13700K where the 7700X gets beaten currently. It would probably make the 7900X look a waste of effort too.

 

[Crackong]

Regular Zen4 cores at low TDP levels already smoke Intel's P+E hybrid by a pretty significant margin.
Hard to imagine what would happen when Zen4c is introduced to the game.

 

[trsttte]

It's awesome that both the performance and efficiency cores will be based on Zen 4.

Would love to see a 2P+4E super low power SoC for tablet/hybrid devices. Maybe even one for a future Surface Neo - but that's double the wishful thinking.

The surface line up (the tablets I mean, the actual "surface", not the laptops that share the brand) have been craving for a solution of this type, the current (and previous) Intel iterations really do the name and form factor a disservice.

For a 15W ultraportable, I'd rather have 2P+8E over regular 4C/8T stuff we're accustomed to seeing.

Since you mentioned the 2P+8E what do you think about them and the rest of the Intel line up with 4p/6p+8E (28w p-series) and 6p+8E ("45"w h-series)?

 

[Athlonite]

I tend to think AMD would be better served improving their IMC / Fabric over doing anything else right now as it's still a shitshow as to whether or not the ram you buy will or won't work at it's advertised speed without a shitton of faffing around with timings and voltages

 

[Minus Infinity]

I tend to think AMD would be better served improving their IMC / Fabric over doing anything else right now as it's still a shitshow as to whether or not the ram you buy will or won't work at it's advertised speed without a shitton of faffing around with timings and voltages

Alas no change for Zen 5 even on IF.

 

[Athlonite]

Alas no change for Zen 5 even on IF.

See this is something they know is a problem but have not really done much to improve perhaps they need to licence so IMC IP from Intel or poach some engineers

 

[Jism]

I tend to think AMD would be better served improving their IMC / Fabric over doing anything else right now as it's still a shitshow as to whether or not the ram you buy will or won't work at it's advertised speed without a shitton of faffing around with timings and voltages

All you need to do is to watch the motherboards QVL list. If you stick to that it's guaranteed pleasure.

IMC's where not a issue; even the 2700X managed to run at 3466Mhz out of the 2900Mhz stock.

 

[Crackong]

I tend to think AMD would be better served improving their IMC / Fabric over doing anything else right now as it's still a shitshow as to whether or not the ram you buy will or won't work at it's advertised speed without a shitton of faffing around with timings and voltages

I ran my 7600x in EXPO 6000 CL30 2x16GB RAM without any problems.
I could further improve performance by manually lowering the sub timings too.

Maybe you should check your MB 's QVL list

 

[R0H1T]

they need to licence so IMC IP from Intel or poach some engineers

License what exactly? Intel's gonna run into the same issues when they go the chiplet route! AMD has no issues with RAM on monolithic APU's like 5700G, in fact they're among the best (wrt support for high speed mem) even counting the ones from Intel.

so there might be some issues with these chips and scheduling to the weaker cores.

Pretty obvious that hardware schedulers aren't that big of a deal as we saw with ADL or RPL, we've had similar programs on windows for eons like Process Lasso. Just get the OS, or dedicated software, to handle the scheduling & all's fine.