AnarchoPrimitivZen3 will double the cores peer CCX, and I feel like some of the benefits seen her will evident with zen3 (yes, I know Zen3 won't be monolithic) . Just look at the 3100 v 3300x, with the single CCX design of the 3300x, it had an average of 12% better performance clock for clock than the 3100...thats what I'm talking about. Reports are already saying Zen3 will have a 17% IPC uplift, you add in the doubled cores per CCX, 200-300Mhz increase to frequency and I don't see anything crazy about predicting a 25% overall performance increase core for core with Zen3.

TomorrowNot sure where you get the double cores. Number of cores will remain the same within each CCX. However each 4 core group will now have access to all the cache within a CCX.
As far as i understand there will still be 4+4 cores with in a CCX instead of unified 8 cores.

ncrsThe number of cores in each CCD will remain the same, but each CCD will contain one CCX instead of up to two as it is today. This is based on semi-official AMD material.

TomorrowThats correct. However it is not year clear if all 8 cores within that CCX (now entire CCD) can communicate with each other without latency penalty or are there still two groups of 4 cores that now can share the cache. The diagram on the slide seems to suggest that (the 4 green Z3 blocks on either side).

ncrsI'm suspecting that they will be fully uniform 8-core CCD/X.
The diagram itself is just a visualization, when you look at the Zen 2 CCD die shot the CCX are top and bottom instead of left and right ;)

medi01I'm lost in AMD lineups...

So what series will the Zen3 lineup be? 5xxx???

medi01I'm lost in AMD lineups...

So what series will the Zen3 lineup be? 5xxx???

Tsukiyomi91these are still 7nm refresh that the mobile Ryzen 4000 APUs are using, just being "adapted" into the desktop socket. I'm interested in how it performs in games, on top of memory OCing. Zen3 is powering the "real" Ryzen 4000 CPU (without iGPUs, obviously.) If these APUs are better than Intel's iGPU, or close to a GTX1050Ti level, both parties are gonna be in a rough spot.

AssimilatorAMD's APUs are faster than Intel's iGPUs by a factor of 2 - 3 times, but that's still nowhere near a midrange card like the 1050 Ti, and it's never going to be. The physical, thermal and power constraints of a CPU package will always dictate the maximum performance of any graphics integrated with said CPU, and the end result is that said maximum performance can never exceed the low end of the discrete GPU market.

OctopussMy head hurts, what's the difference between a CCD and a CCX?

OctopussMy head hurts, what's the difference between a CCD and a CCX?

laszlogoogle is your friend no?
www.hardwaretimes.com/what-is-a-ccx-vs-ccd-in-an-amd-ryzen-processor/

windwhirlCCX is a Core Complex. Basically 4 cores plus cache and some other stuff. A CCD is a Core Chiplet Die. Two CCXs in the same die.

OctopussYea I wanted a quick and dirty explaination like the person below gave me, no thank you.


So CCD=the actual CPU, roughly speaking.

windwhirlPretty much, yeah. Mem and IO controller are on a separate die, though, different to past approaches of unifying everything in a single die.

AssimilatorAMD always releases their APUs as the last parts in an architectural cycle, while naming them the same as their next architecture.

So Zen CPUs were 1000-series but Zen APUs were 2000-series, 2000 CPUs = 3000 APUs, and now 3000 CPUs = 4000 APUs. Zen 3 CPUs will thus be 4000-series and Zen 3 APUs will be 5000-series.

It's weird and confusing but it is what it is.

AssimilatorAMD always releases their APUs as the last parts in an architectural cycle, while naming them the same as their next architecture.

So Zen CPUs were 1000-series but Zen APUs were 2000-series, 2000 CPUs = 3000 APUs, and now 3000 CPUs = 4000 APUs. Zen 3 CPUs will thus be 4000-series and Zen 3 APUs will be 5000-series.

It's weird and confusing but it is what it is.

AssimilatorAMD's APUs are faster than Intel's iGPUs by a factor of 2 - 3 times, but that's still nowhere near a midrange card like the 1050 Ti, and it's never going to be. The physical, thermal and power constraints of a CPU package will always dictate the maximum performance of any graphics integrated with said CPU, and the end result is that said maximum performance can never exceed the low end of the discrete GPU market.

TomorrowLets hope so.