AMD Ryzen 8000 Strix Point APUs Utilize 12-Core Zen 5 and Zen 5C Hybrid Config
wccftech.comI'm surprised that WCCF tech is getting on the HN front page. They are known for just running with any inkling of a rumor, or even just jumping to their own wild conclusions and running articles on that.
this is an older info via TomsHardware :
"Ryzen 8000 Strix Point APU Comes Forth With 12 Zen 5 Cores"
https://www.tomshardware.com/news/ryzen-8000-strix-point-apu...
~ "As always with leaks, take the news with a pinch of salt."
There's nothing in the Tom's Hardware article about a hybrid CPU config or Zen5c cores, so this Wccftech rumor is not just a retread of the Tom's Hardware article and the Tom's article provides very little support for the Wccftech rumor.
In a few days, TomsHardware will also publish the same news.
And Videocardz has already published it.
https://videocardz.com/newz/amd-ryzen-8000-strix-point-apu-s...
"According to Olrak29, the Strix Point APU is expected to feature four Zen5 cores and eight Zen5c cores. On the other hand, the larger APU variant, known as Strix Halo or Sarlak, is said to boast 16 Zen5 cores divided into two chiplets, similar to the desktop version. The primary difference between desktop and ‘ultimate APU’ variant lies in the IO die, which houses the GPU, claims another leaker. The desktop processor will have a less capable GPU, while the Halo variant is rumored to incorporate up to 40 Compute Units, as suggested by earlier reports."
Does the word 'Strix' have some sort of computing relevance that I am unaware of or is it really just a cool sounding name for a genus of owls? Because Asus uses this word in products too and I've never known if it was supposed to relate to something.
Asus specifically is a reference to the Owl: https://www.asus.com/us/content/strix/
it has X in it
Oh... That is disappointing.
I was hoping Strix Halo would be a big, power efficient, Apple M-esque SoC with a wide iGPU.
The 16x Zen 5 cores (with no efficiency cores) and the "sharing" of the desktop architecture suggest its actually a Dragon Ridge 7045HX replacement. So a power hungry CPU for 17" desktop replacement laptops you can basically only use while plugged in... That has a large IGP for some reason? That doesn't make much sense, as you mind as well use the dGPU.
Desktop Zen 5 (Granite Ridge) is 128-bit while other rumors said Strix Halo is 256-bit so if that's true they aren't the same. I suspect there will be three or four different laptop Zen 5s (baby, regular, halo, and desknote) so this article is probably just incomplete.
But the Zen 5 CCDs will presumably be the same.
If AMD used a single Zen 5C or 4C CCD, I would be more optimistic... Even though AMD didn't differentiate the cores, the 4000 series and 5000 seres were essentially Zen 2C and Zen 3C. But 2x Zen 5 CCDs suggest they are going for all out performance.
> That has a large IGP for some reason? That doesn't make much sense, as you mind as well use the dGPU.
But maybe the goal is to finally offer an H class SoC that doesn't need a dGPU?
RDNA 3.5 with up to 40 CU should be well into Apple SoC Pro/Max territory.
But low power/idle efficiency will be awful using the desktop Zen 5 CCDs, if they are anything like the 7000 series.
They are optimized for high clocks, and the interconnect burns tons of power, especially with 2 CCDs (which reach into each other's L3 over the link).
In terms of IPC @ GB6, with an expected ~20% gain from Zen 4, Zen 5 still need another 15% improvement before it catches up to A16.
But it will be interesting to see any write up on the benefits of using Zen5c ( Cache scaled down version of Zen 5 / P-Core ) instead of using a smaller but different uArch E Core as in Intel and ARM are doing.
Granted that it is too early to know, but I like the AMD approach. If I’ve understood correctly, the c-cores have absolutely identical functionality on all fronts. They can’t clock as high due to the layout being redone to optimize power & area, and they lack some cache vs the high-power cores.
So essentially, any core should be able to handle any workload but you’d bias single-threaded tasks onto a few 5 GHz (or whatever) capable cores. And if you’re running well-threaded code across many cores, they’ll have to slow down anyway to stay within the power envelope, at which point all are effectively identical, barring the extra cache.
Only question is whether Intel saves a lot more power & area with their (IMHO crippled) efficiency cores. If they do, it ought to give Intel a cost/price advantage.
Perf/area for the Intel E cores is very good. IIRC they are 1/4 the size of P cores with more than half the performance, and that's not counting any L3.
And Intel is uniting the ISA with future cores.
AMD's Zen cores are (IIRC) smaller than the Intel P cores, and they have a big process/packaging advantage for now, so their "full core" strategy is indeed working very well.
They have half the L3 cache as well. However a Zen 5 chip with half L3 cache will very likely outperform a Zen 4 chip with full cache.
I wouldn’t expect any hybrid designs out of AMD for desktop, only mobile.
What’s the point of putting AI tech in their cpus when they’re clearly not interested in making it work even in their gpus?