AMD: Zen 4, Speculation, Rumours and Discussion

Why is this a given? It's not confirmed at least.

While the vcache die is obviously more power limited and thus more limited on all-core clocks, that should not affect the max boost clock, and the cores should be pretty decoupled from the L3 anyway.
An AMD representative confirmed it in a video for pcworld. ccd with vcache => lower speed, without vcache => high speed. He explained that they worked with MS so Windows scheduler would know what apps need top speed or more cache (I don't believe it but that's what they said). I can't find the video anymore :/
 
Really interesting. I have a 7700x the 8/16 core. I think my next chip will be zen 5 with the ai cores and 3d cache and 16 cores lol. Gotta wait I think for that. Hopefully its socket compatible .
It will be, AMD already committed to AM5 'til 2025 at least (so at least 2 or 3 more gens)
 
Hope so, went with AM5 because of it but for budget reasons didn't get anything top notch but mid-range for a MB with an asus x670e tuf, crossing fingers Asus doesn't screw us with lack of bios updates, if so, definitely won't repeat it ever again.
 
Hope so, went with AM5 because of it but for budget reasons didn't get anything top notch but mid-range for a MB with an asus x670e tuf, crossing fingers Asus doesn't screw us with lack of bios updates, if so, definitely won't repeat it ever again.
Err, anything X670E is definitely in the high end range of motherboards
 
Err, anything X670E is definitely in the high end range of motherboards
well, granted, entry lvl in the high-end options.

But it has its letdowns, and for the price of MB i expected a bit better.
For ex. on the sound features at the software side its DTS control panel features are at idiocracy level with 3 options, music, movies and games.
There's a 4th personalize option but most "improvements" are awful to meh.

The stereo improvements just ruin the sound with excessive levels for vast majority of sounds and only sound better on sound sources with very little to almost no stereo width at all.
The bass improvements do the same, overpowering all the other mid to upper frequency ranges, you would need to be playing a sound with severely weak bass levels to actually feel i could say it improves anything same for the voice option, often too aggressive and ruins the audio most often than not, you would have to be hearing some really old poor recording to improve anything i imagine.
Highs improvement actually redundant since all these options (appart the stereo ones) are just equalizer profiles and there's no clever adaptive mechanism behind them to avoid excessive adjustments, definitely no Auto-EQ, and only a 5 band EQ is a joke but at least has something, and pathetically that's something that microsoft could'v added to its sound options ages ago, but i digress a bit much.

I rather use StereoTool with VB-audio despite some added latency from insignificant/unnoticeable to acceptable depending on quality settings instead of any of these offered by the Asus DTS panel.
 
Last edited:
An AMD representative confirmed it in a video for pcworld. ccd with vcache => lower speed, without vcache => high speed. He explained that they worked with MS so Windows scheduler would know what apps need top speed or more cache (I don't believe it but that's what they said). I can't find the video anymore :/
they could also expose some manual trigger.
 
An AMD representative confirmed it in a video for pcworld. ccd with vcache => lower speed, without vcache => high speed. He explained that they worked with MS so Windows scheduler would know what apps need top speed or more cache (I don't believe it but that's what they said). I can't find the video anymore :/
How does it know if a program runs faster with or without vcache?
 
How does it know if a program runs faster with or without vcache?

He didn't explain. IMO it won't work well. They haven't figured out ccd yet (in some game you have to disable a ccd to solve weird framerates, or use process lasso), so adding more complexity is, well, not the smartest thing to do. Time will tell.
 
He didn't explain. IMO it won't work well. They haven't figured out ccd yet (in some game you have to disable a ccd to solve weird framerates, or use process lasso), so adding more complexity is, well, not the smartest thing to do. Time will tell.
Yea it sounds like a recipe for trouble. Interested to see 7900X3D vs 7800X3D in games and apps that aren't typically used for benchmarks.
 
It's always a bad news for schedulers when all your cores are not created equal (the same goes for the big-little arrangement in mobile phones and later Intel CPUs). For big-little arrangement it's a little easier because you can mark a thread as requiring high performance (or low latency) to be prioritized to big cores and other threads as "worker" threads which can be prioritized to small cores.
However, 7900X3D and 7950X3D are a bit different, as they are not just fast vs slow, but with different characteristics (i.e. one with big cache and another with higher frequency). It'll be harder to flag a thread as "love bigger cache" and another thread as "don't care about cache but give me higher clock". Not to mention that it could change for the same program with different workloads. For example, some workloads may be small so the built-in cache is big enough (therefore the higher clock rate could help), and another workload could be too big to fit into the built-in cache so it can benefit from the V3D cache.
It might be possible to use performance monitoring to try guessing which thread is better fit for which, but that could increase the cost of context swap, which is not good.
 
So the 7900X3D has only 6 cores with direct access to vcache? Even with perfect scheduling this could be a problem vs the 7800X3D.
 
So the 7900X3D has only 6 cores with direct access to vcache? Even with perfect scheduling this could be a problem vs the 7800X3D.
apparently PBO etc will be available for Zen 4 3D , so yeah its going to be really interesting to see how far a 7800X3D can be pushed , i dont like the 7900X3D at all.......

hopefully with a bit of negative curve offset , we can get to 5.2-5.3 at same voltage /power as stock.
 

Mostly what I suspected from launch, with AMD not just suggesting, but mandating that all testing for reviews be done with their specific high speed DDR5 kit that was supplied.

Along with making the default 'use all the power possible' instead of keeping that as an overclocking option, they were really trying to juice the benchmarks and make up for their earlier claims about performance, which were perceived as fairly disappointing for a new generation on 5nm two years later.

Hoping Zen 5 is a bigger fundamental leap.
 

Mostly what I suspected from launch, with AMD not just suggesting, but mandating that all testing for reviews be done with their specific high speed DDR5 kit that was supplied.

Along with making the default 'use all the power possible' instead of keeping that as an overclocking option, they were really trying to juice the benchmarks and make up for their earlier claims about performance, which were perceived as fairly disappointing for a new generation on 5nm two years later.

Hoping Zen 5 is a bigger fundamental leap.


What that review only touch upon, is that even with DDR5 5200 you can get excellent gaming performance, as long as you lower the timings. Just check LTT's recent review which shows how little Zen4 cares about bandwidth and how sensitive it is to timings.

Compare DDR5 5600 CL28 to DDR5 6400 CL32 in AMD tests ...
For people with too much free time on their hands there is also quite a bit of performance to be gained by manually tweaking all the secondary subtimings and not only relying on MB manufacturer and their implementation of XMP / EXPO (up to 10% (sic!) difference between different models of AM5 boards with the same CPU + Memory - see another HUB video).
 
AMD really need to figure out a way to have the CCD with the V cache, at the same speed as the other CCD / non 3D cache variant of the same CPU.
Too many compramises and really muddies the water, in terms of presenting a single best CPU.
 
Back
Top