- Feb 21, 2020
- 2,559
- 3,864
- 17,570
To be clear: Not letting you uninstall the thread scheduler driver is utter incompetence. It's practically inconceivable, especially since it has been known for 18 months that if you have it enabled with a single CCD Zen 4 (and now Zen 5) chip it will hurt performance. AMD says it's a Microsoft problem, but that's just bollocks. AMD can install the driver when it's needed, and it should know on a fundamental level everything that it changes, and yet it says it can't roll back the installation? That's nuts. This should have been fixed 18 months ago.
Now, did AMD or anyone else screw up some of their numbers due to the core parking and thread scheduling getting enabled on single CCD tests? I'm absolutely sure some places did, and it's entirely possible that in the time crunch even AMD's testing made a mistake. Because testing takes a lot of time — days if you're retesting everything! Paul started testing for these launches a few weeks ago. If you asked him to verify all the data, it would take weeks to do so on every CPU. And he's working through that, incidentally.
That AMD supposedly reran everything within two days of the launch... well, it's possible, because partly it's just a manpower problem, but even then there are scaling issues with trying to do lots of testing in a short period of time. TLDR: No one other than AMD itself can say with certainty that all the testing actually happened and that the thread scheduler was/wasn't used as appropriate.
Normally, you just install a new CPU and rerun the numbers and it takes maybe a half a day to gather the data, for each CPU you want to test (depending on the size of the test suite, obviously). If you have two or three different systems that are all identical, you can speed that up some, but there are quite a few games that require manual testing and so there's a practical limit to what you can do. (And let's also note that even two "identical" PCs can have slightly different performance.)
With this core parking / thread scheduler, you have to interrupt the normal workflow and do a clean OS installation, then update everything and make sure all the previous settings are in place, and potentially even download a couple of terabytes of games. That takes hours, at a minimum. If you have an image of the OS from before installing the thread scheduler and chipset driver, you could reimage instead (which is what Paul has done), and then maybe it's just a 30 minute delay or less (depending on how your images are created, like if the games are on a separate drive or not). I know in my case I don't have an image of my full test PC that I could roll back to if needed, because I just have a 4TB SSD with my entire 2.5TB test suite installed.
So, I can say with reasonable certainty that Paul's data is correct for how he tested. (I'm not Paul and I didn't run all those tests, so I can't be absolutely certain.) But for other sites and AMD itself? Unless they issue a statement saying, "We did not have the thread scheduling dual-CCD chipset driver installed on our single CCD testing," it's simply a guess as to whether or not they did that. And since it's unusual in the first place to have such a driver that's required for some chips and causes problems for others, that makes it entirely possible that it got overlooked. Is that incompetent? Perhaps, but I'd almost certainly have missed it if I were the one doing the testing of all these CPUs! And then I'd be a couple of weeks behind schedule...
-----------------
The second part about the lack of stellar sales is easily enough explained. Even if Zen 5 is faster than Zen 4, launching without X3D parts means the previous generation Zen 4 X3D chips are often better. However you want to explain the reasons behind this (there are plenty of valid reasons), the fact is that a lot of gamers don't want non-X3D from AMD now. Everyone who cares knows that 9000 X3D chips are coming, within a few months, and thus we have a "wait and see" mentality — potentially also coupled with a "wait and see what Arrow Lake does" mentality for people who aren't set on using AMD.
And frankly, new CPU / platform launches don't typically cause tons of people to queue up — even when everything goes perfectly, which it certainly hasn't in this case. A new GPU is a simple enough upgrade, but a new CPU can mean a lot more than just a chip swap. Probably 12th Gen Core was the last major upgrade cycle where people got truly excited. It was a big change relative to 11th Gen (and everything before it). And it was Intel, which still leads in market share on the CPU space. AMD's Ryzen lines have been more of a slow but steady chipping away at market share than a massive surge in upgrades. At least, that's my take from the sidelines.
If I were a typical gamer / enthusiast with a Socket AM5 system, there's almost nothing here to get me to upgrade right now. I can only assume a lot of people feel the same. Even 10% higher performance at 1080p doesn't actually matter much, because I'll be gaming at 1440p or 4K. And if I have an X3D Zen 4 chip, it would actually be lower performance for gaming. If I have an older PC that's more in need of an upgrade, then I have to swap mobo, RAM, and CPU to get there, which is a much bigger deal and I wouldn't want to jump the gun if potentially better stuff is less than three months away.
So, for most people, Zen 5 isn't a good upgrade option right now. Over time, new PC builds will naturally shift to Zen 5 instead of Zen 4, and Zen 5 X3D as well. There are plenty of good aspects for the new chips, but they're just not enough to get me, personally, to think I'd need to upgrade from anything that's less than three years old.
Facebook
Twitter
Instagram