Gamers who find Ryzen 9000s disappointingly slow are testing it wrong, says AMD AMD has responded to community concerns about underwhelming gaming performance of its Ryzen 9000 Series desktop processors by attributing disappointing results recorded by independent testers to the vagaries of their system and software setups. One factor in the complex interplay of technical variables is that AMD conducted …
Doesn't encourage upgrading enough I have to say. Is it unreasonable these days to expect a newer processor to run faster, regardless of operating system? It used to be the optimisation was generally application specific, rather than OS specific
Amd have been pretty transparent and accurate in their internal benchmarking, since before ryzen.
Any benchmarker knows that deviations are quite often due to differences in the benchmark (specific games in this case) as well as various hw os and game settings.
To claim that one set of benchmarks is "reality" but another set isn't, is just trying to get attention for a normal and predictable finding. at a minimum they need to look for other factors before declaring "amd is lying".
And "who doesn't run games in admin mode" ? just the part timers that are ok going slow, i guess. if that's a setting needed for speed, then use it !
What do you mean it doesn't matter that it runs faster in the Admin account. If there was a 50% performance hit between using the admin account and not, are you saying it wouldn't matter to you? If not, where do you stop caring with this performance difference? It does matter, why does it run faster, what has Microsoft done to impact performance for users when not running with the admin account. Is UAC and all the redirections that they do for writes to system locations impacting performance so much as to slow the system down this much?
Why does Admin mode make a notable difference?
Apparently there's some branch prediction code that's only available in admin mode. I don't know the details of why but presumably because there's something a bit Spectre-y in there that's a potential security risk and you'd only let trusted code use it. More aggressive branch prediction could allow some CPU-bound processes to run quicker.
The upcoming release of W11 appaently makes that improved branch prediction available in user-mode (presumably with protections).
"> (presumably with protections).
I admire your optimism."
"... with protections" does not mean that those protections *work*, just that they are there !!!
There are *many* protections in all versions of 'Windoze', just that over time more and more are found to not work anymore ... or perhaps never did to the people in the know (Hackers/crackers and other ne'er-do-wells).
:)
Is it unreasonable these days to expect a newer processor to run faster
Actually, kind of yes. Most of the development is increasing the number of cores and better optimization of the cores but unless software is written to take advantage of more cores it'll not run any faster.
Newer CPUs can efficiently multitask more processes than an older CPU but it might not run those individual processes any faster. A new i7 may have 8 cores, a i7 from around 6 years ago had 2 cores running at the same speed as 6 of the cores on the new one.
Dude, can you please not selectively quote parts of my post to make the opposite point?
What I said was
'Is it unreasonable these days to expect a newer processor to run faster, regardless of operating system? It used to be the optimisation was generally application specific'
What I'm questioning is if it is unreasonable to expect, with application changes only, to receive a performance advantage when the linked data seems to be requiring an OS update.
I suppose that even before modern P and E cores going way back the introduction of hyperthreading slowed some processes down if the application/OS wasn't hyperthreading aware, but it was at least possible to disable it in the BIOS so this could be worked around.
Is it unreasonable these days to expect a newer processor to run faster, regardless of operating system?
If I understood Wendell (https://www.youtube.com/watch?v=0eY34dwpioQ) correctly, not all cores perform the same. When running a game, you want to prioritize which cores do what.
It feels iffy... If a game employs a dozen threads, how can you predict that two of the threads do most of the grunt work and should run on your two best performing cores..?
The chaotic results (some games run faster on X, others do the opposite) suggests that there is no generic way of doing this. Maybe each game ends up with a manifest that declares what optimizations are expected. (sigh)
I suspect the casual gamer will not be affected (they'll want the UHD eye candy). At least nobody claims these particular CPUs are slowly grilling themselves.
Bottom line, for me, I think I'd be totally fine with one of the 9000-series Ryzens. I'd like some more cores than what I have in my 2700X, but it will all be good.
...Is the future. Someday all workloads will be optimized on the fly for the platform on which they are running. We are not there yet. True benchmarks shouldn't be optimized at all in this day and age. Not for a GPU, CPU, OS, or anything else. This is the only way to be fair. However brands want their names in lights so we will see things like this from time to time.
This is just an iteration in the evolution of both software and hardware. In the future the most successful software will be that which can self optimize to the largest number of hardware stacks.
While bloated inefficient code should be nuked from space...
... shouldn't the processor attempt to optimise for itself, like out of order execution or something? Because if programmers/compilers are going to have to optimise for specific types of processor, then this means that either performance will start to suck for other processors that require different optimisations, or it'll become a nightmare of tangled linguine trying to have a program flow that attempts to work "best" for several different types. For example if this Windows update provides better results with the AMD chip, how does that affect Intel devices? Or older AMD, come to think of it?
All modern processors optimised for themselves, with out of order execution, executing multiple instructions in parallel, elaborate branch prediction, all manner of caching, etc. I think the problem is one of diminishing returns - we've got so clever at all of this optimisation that further improvements require increasingly complex hardware and only give marginal benefits.
Testing with what's actually available is not exactly testing it wrong.
If there are CPU specific drivers needed to take advantage, not in circulation, then of course downstream benchmarks will be different.
I'll be looking forward to GamersNexus and others redoing their benchmarks once the software releases to confirm/deny AMDs claims.
My money is on the reviewer, not on the marketing exec.
"...are waiting for the X3D models. Or 'maybe' just giving up and buying a 7000 series X3D now."
Same, but as there is nothing technbically wrong with my current system, I am quite happy to wait untill January, worst case the 7800x3d will be even cheaper by then..
But man, AMD, do they never fcuking learn like. They keep making these easily avoidable mistakes..
I'm sat on a 5950X, and barring things falling out of firmware updates there is absolutely NO rush to upgrade... The recent security findings and patches being extended back to the 3000 series does put a timeline on roughly when one might expect the firmware to dry up for current release.
I cannot complain, AMD have produced good quality and relatively long-lived products in the main. If the 9000 is a small, incremental improvement only that's fine, it's an easy upgrade skip.
Also sitting this out for a little bit longer.
Built an AM4 system back in 2019 with a 3800X.
Looked at AM5 when it released, but switching at the time, which would have needed new CPU, motherboard and RAM, and this just wasn't financially viable in my mind with cost vs the performance bump (no AM5 X3D out yet at that time, and the new RAM was very expensive back then).
So I dropped a 5800X3D into the existing AM4 system in Sept 22 instead. Which is what I'm still using now and so far in no rush to switch to AM5.
I might have a look again when the 9800X3D (or whatever they call it) drops. But I think my next purchase will be replacing the GFX card anyway (RX 6900XT atm, which is decent enough for now).
At least Techpowerup.com initially reviewed the new AMD CPU's and a few days later followed up with a "found the missing speed" article where they disabled SMT (Simultaneous Multi-Threading) and the performance was better in applications which don't scale with the core count.
The AMD/Intel CPU drivers tell optimization hints to the Windows kernel internal scheduler. They can among other things inform the preferred cores, preferred core combinations for better heat distribution and to inform which cores are for performance (P) and which are the low-power efficiency (E) cores so that games and such are not handicapped by the slow cores which don't have SMT. Maybe the functionality also extends to instruct the scheduler to NOT prefer SMT.
Hardware unboxed retested the CPU's with the administrator mode... and found that all of the gains they found with the 9xxx series.... also improved the 7xxx series by about the same amount.
It's almost as if AMD looked at the strife intel are kicking up with their failed handling of the 13/14th gen oxidizing vias... and went... here HOLD MY BEER
I'm laughing with my 5800X3D and 6900XT... I reckon I'll get a decent GPU upgrade in before that CPU becomes a bottleneck for anything... certainly not planning a rebuild before 2026 now.
Well, I just got myself a new machine about a week ago, with a Ryzen 7900. I was betting that there would be problems with the new generation of processors, and the speed gain wouldn't be that much anyway. Maybe I was right on the second count. But perhaps I could have got myself a cheaper processor by waiting for the older generation to fall in price. Nah, not worth the bother!
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
