• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Zen 5 rumours

"30 to 40% IPC" was never going to happen, it might be in a few very specific server type tasks, like all CPU's they are primarily designed for Data Centre, the best silicon gets skimmed off for that and the rest is packaged for retail.

There are a couple of interesting things about them, they have an ladder core to cache to core design, its the bit that Intel calls the "Ring Bus" this is completely original and should have a better core the core latency, tho AMD's version of the "Ring Bus" is already very good since they improved it from Zen 2 to Zen 3, there is a reason why despite only having about a 9% IPC gain from Zen 2 to Zen 3 but a jump of some 40% in gaming, suppasing Intel for the first time with Ryzen 5000. See slide below, so it should help gaming performance.
This is different to memory latency, tho it looks like AMD have also worked on brining that down by improving the way the CCD's connect to the IO die, this should also help gaming.

While the IPC might not be 'a lot' higher, this is more than just an iterative design, its a fundamental architectural improvement. Its more efficient and also much wider, you're Cinebench scores might not go up a lot, but it might be much better at code compile and gaming.

fAT33eY.png
 
Last edited:
I started to notice all the security updates slowing down my Intel 6700K so i jumped on a Ryzen 2700X, from there i went 3900X > 5700X > 5800X3D.

Only thing that stopped me from going to 7000 series was the whole AM5 upgrade and that i didnt need that upgrade after using this X3D.

Will be looking closely at the 9000 series when they arrive. Im more excited about CPU's than any GPU's right now :p
 
Thinking if the 9000 series (if that's what it end up getting called) is good, sort of around 70003d levels of gaming performance with a bump to productivity, im gonna upgrade from my 3900x
Same is my thinking. 5800x3d is great but already I can feel it holding back my 4090 in UW resolution in some titles (monitor can reach 175Hz, CPU maxed out lower often enough to notice). And since I don't plan to upgrade GPU, that would be plenty fast. I can always go for 7800x3D if it's all faster and cheaper. :)
 
Last edited:
Doubtful, they have experimented with that on 7k series and decided it's not worth it, hence it never hit retail - as per their own engineers.

yeah remember watching gamersnexus tour of AMD, said didnt make any difference in gaming and just increases cost, and the delay going across ccds if the game needs more cores
 
yeah remember watching gamersnexus tour of AMD, said didnt make any difference in gaming and just increases cost, and the delay going across ccds if the game needs more cores
The latency on that would make that a no go imo, I would like to see what would happen if they designed an X3D variant of the 8700g, in my head al larger die = more cache = more FPS.
 
Not having the 3D cache on both CCD's is clearly not without its problems, i think when they say "its not worth it" it doesn't mean they think its not without its problems. It just means they don't think its cost effective, or something like that.
 
Last edited:
Or increase the cores on each ccd , they are currently capped at 8 cores per ccd ?

Example going forward ccd with 12 or 16 cores with 3d cache which won't have issues with latency for gaming compared with 2 ccds with 3d cache
 
Last edited:
Or increase the cores on each ccd , they are currently capped at 8 cores per ccd ?

Example going forward ccd with 12 or 16 cores with 3d cache which won't have issues with latency for gaming compared with 2 ccds with 3d cache
Power density issues that's why the frequency is lower on c chiplets, wonder if there's a technical reason why they couldn't make a large unified L3 and stack two chiplets on top of it, would also help with cooling the chiplets.
 
Or increase the cores on each ccd , they are currently capped at 8 cores per ccd ?

Example going forward ccd with 12 or 16 cores with 3d cache which won't have issues with latency for gaming compared with 2 ccds with 3d cache

Might be a little pricey on the current node. That would be a pretty decent sized piece of silicon.
 
Power density issues that's why the frequency is lower on c chiplets, wonder if there's a technical reason why they couldn't make a large unified L3 and stack two chiplets on top of it, would also help with cooling the chiplets.
It would probably be difficult to deal with the chiplet’s wiring if the L3 was under the cores, think it’s possible the L3 could be moved onto the IO die but the latency will be higher so?
 
Might be a little pricey on the current node. That would be a pretty decent sized piece of silicon.

Indeed. With Zen5 being on 4nm for the desktop at least, we will have to wait for 3nm and Zen6 for AMD to develop a 12 or 16-core CCD for the normal core. That doesn't stop AMD from offering 8 Zen 5 and 16 Zen 5c chips along with their normal 16 Zen5 variants. It might be more expensive but it will probably offer better multicore performance than the 16-core Zen5 variant. Hopefully, AMD will also improve the memory speeds so that a supposedly 24-core Zen 5/c variant will have sufficient bandwidth to operate in a dual-channel system.
 
Indeed. With Zen5 being on 4nm for the desktop at least, we will have to wait for 3nm and Zen6 for AMD to develop a 12 or 16-core CCD for the normal core. That doesn't stop AMD from offering 8 Zen 5 and 16 Zen 5c chips along with their normal 16 Zen5 variants. It might be more expensive but it will probably offer better multicore performance than the 16-core Zen5 variant. Hopefully, AMD will also improve the memory speeds so that a supposedly 24-core Zen 5/c variant will have sufficient bandwidth to operate in a dual-channel system.

Good point on the memory bandwidth. 24/32 cores running at full tilt would be very bandwidth intensive.
 
Back
Top Bottom