• 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

I think Derbauer showed his 9700x at 180w with PBO on and 80w with it off in the same test

So its a 20% performance gain in Cinebench for 125% increase in power draw. Up to you if its worth it

I think this is how all CPUs and GPUs should work. I want it to use low power out of the box and let me decide if its worth adding extra power for 20% more performance, don't make that choice for me
 
Last edited:
Old rumour claims certainly do look funny in hindsight

Take this one for example: Where he claims 25% IPC, higher clocks speeds and ok efficiency.

Instead the 9700x has about 10% IPC, lower clocks and very good efficiency

 
Last edited:
Personally, I think the click bait reviews are mis-placed and much ado about nothing beyond seeking names in lights. Its becoming a tired old trope.

If it were a car manufacturer that turned round and said: "here's our new car model redesign ... we put a new engine in it, it has around the same driving performance, but uses 16% less fuel to do that, oh, and we we took a little off the RRP compared to the old model's RRP" there would be general positivity around it. Yet here in the computer enthusiast world its a flippin disaster ??? They've acheived a similar performance using less power ... that is an improvement.

Yes, of course you can buy an equivalent older model for less, just like you could buy an older pre-registered car for les+ ... or an older, higher specced model for the same price, there is nothing new to this situation.

Maybe I'm just getting tired of whining by tech-tubers.

The old model is still cheaper hence the reason why all advice is to buy a 7700x or 7700 instead of the new stuff.

Also your car analogy doesn't work because you didn't mention what the car is used for. In terms of gamers, that would be a drag strip car - because we only care about performance, top speed is what we want, more frames. So you give me a car that's 7% more efficient but it's no faster than my current one and I'm like... uh ok, but that's a pointless product for me unless I had to buy my first car or my old one was broken and the extra efficiency is completely lost on the depreciation of my current car and the price of buying the new one, the fuel/electricity is still cheaper


The only people really interested in getting rid of their current car for one that's 7% more efficient but otherwise identical is taxi companies. And for us gamers, we don't care about taxis/businesses uses, that's boring, that's why the reviews are negative, it's a boring product
 
Last edited:
What a disaster

AMD has told reviewers they have to setup windows to act like the 9950x is a X3D part, so that game threads stick to a single CCD and can't move around - effectively making it a 8 core CPU in games

But why are they doing this? The 7950x ran fine with just windows doing the scheduling across 16 cores, why do we need to lock down the CPU to a single CCD in games now, the 9950x isn't X3D

Well, ask you shall receive. Behold the reason why. If you don't park one CCD while gaming you will see a significant performance loss because the core to core latency in Zen5 is almost 3 times higher than Zen4

 
Last edited:
Could make sense if they really were spun off from a design intended more for server type task as some claim and optimised for crunching data.


It doesn't make sense from what we know of the architecture so far but it's real; hardware unboxed touched out at the end of their video, though they did provide benchmarks for it, but they said if you don't use core parking with the 9950x you see a large gaming performance drop

And in any case, even with core parking working the 9950x averaged like 1% faster than the 7950x in games and often used more power as well
 
Last edited:
Looks like the higher core latency is on zen5 mobile too. The latency on the AI9 HX370 Zen5 CPU is more than double last gen Zen4 models
 
For me that just reinforces my theory about the Windows Scheduler.

Whatever the case is, it looks like AMD has changed something, its just whether its AMDs change thats the issue, or whether Windows is having a hard time dealing with the change and needs an update, or maybe a combination of drivers/bios and windows updates is needed

It would be great if AMD can quickly clarify what has happened as the media is just running around trying to guess whats wrong
 
Last edited:
Actually it looks like whatever AMD did its not just CCX to CCX latency affected, though that is the worst. A thread jumping between the CCXs has a latency penalty of 250% over Zen4, however the latency of a thread jumping from one core to another within the same CCX also has a latency increase of 40% compared to Zen4.

Meaning even single CCX Zen5 CPUs like the 9700x are suffering from higher latency between the cores. Can Windows fix that?

And both of these scenarios where the latency is increased is occurring on both the Ryzen 9000 desktop parts and also the Strix point AI HX mobile Zen5 parts.
 
Last edited:
MLIDs claims:

1) Zen5 was launched in a state similar to engineering samples and with broken software. Hopefully performance can be fixed with future code updates.

2) AMD's marketing team completely lied and made up fake benchmark numbers; much of this is due to poor internal communication between marketing and engineering.




 
Last edited:


Well there goes the windows theory out the door, Zen5 is showing massive core to core latency even in Linux

Wouldn't be surprised if this where all Zen5's missing gaming performance went to

Should be easy to test - benchmark some single threaded games, like Crysis - as long as the game only runs on a single core and doesn't switch cores then the latency issue doesn't matter and you should see a large performance gain over Zen4, that is if latency is the issue with gaming performance.
 
Last edited:
Sounds like a potential issue with microcode and new Agesa incoming. Just very weird if they missed it before, but then again they increased cache potentially because of this exact reason? But that would mean it wouldn't be visible in standard use aside specific latency tests. Many questions to be answered! :)

Edit: I just found a post showing AMD slides seemingly confirming they changed inter-core topology in Zen5 from ringbus to mesh. The latter has been used in servers for large number of cores CPUs and it indeed could explain higher inter-core latency than previous topology from Zen4. In other words, it would seem Zen5 based Ryzen CPUs are structured more like server CPUs for productivity and not for gaming. I wonder if this is what AMD meant saying x3D chips will have more changes than just cache added this time - different inter-core topology? But that would mean they would be segmenting Ryzen CPUs between productivity (and general use) Vs purely gaming ones even more than before. And possibly slap even higher prices on x3D ones.


After a quick Google search, it looks like mesh is supposed to produce lower latency than ring bus, but maybe it's a software problem somewhere
 

Update: Not a bug, the exact same behaviour occurs in Windows 10 - log in as super admin and you see the same performance, this is just how Windows works: you bypass security features and your framerate goes up.

You also don't need to use super admin, just right click on your games and run as admin - you get the exact same performance boost, and even better this boost applies to Intel too . Just run all your games as admin and you will get higher performance and this applies to Windows 11, 10 and 7 and applies to AMD and Intel

Again, not a bug, AMD trying to pass blame
 
Last edited:
That might review poorly as the non 3D cache CCD achieves higher clock speeds for productivity work. Also there's not much that would benefit from the second CCD having 3D cache is there? Games will all run on a single CCD.

The only thing having v cache on both ccx's do is being able to get rid of the driver, bios and windows workaround required to make sure the game uses the right ccx. You'd be paying more for convenience and not more game performance
 
Last edited:
I got all my games to run in the end, while in the menu of each i changed to window mode and 2560x1440p, loaded a save and could play, then quickly changed back to 4k and hey presto no more CTD

A few of my most played games, 4k, settings ultra unless stated

Cyberpunk 2077 i seem to be around 5-7fps more than before, sitting around 97fps on the 9950x. (mix of ray tracing settings, frame gen off)

Horizion Forbidden West i seem to be around the same fps in all honesty but the game feels much better, dunno maybe the 0.1% lows are much better.
(high settings to get it to work, had shader compile issues initially)

Starfield as before a decent uplift, explored Akila City and the outskirts, getting a few dops to around 65fps in heavy crowd areas, but out in the wilds a lovley 120 - 130fps, and again the game just feels much smoother like Horizon (mix of ultra and high settings)

Destiny 2 same performance but thats due to the old game engine it uses, i upped the frame cap to 180fps from 144 and it was very close, a steady 172fps average. (graphics settings set to max)

The boot times are absolutely savage too on the 9950x and bios v2204, apart from changed to expo II and setting 6200mhz speed, i'm seeing 16.1s total boot time which is stupidly quick, my 7950x would take around 48s to get into windows before, its a very welcomed change :D

Weird my 7950x is around 10-12 secs to get into windows. 48s sounds like your motherboard was training the memory on every boot
 
Last edited:
Back
Top Bottom