Ryzen

[Mercutio]

The reviews are out for the big new Zen5s and they do seem to be fairly underwhelming with very modest, sub 10% performance gains to be had across most applications and, for the high-core models, a fresh Windows install and some goofing around with power modes and chipset drivers to get CPUs to choose the number of cores they engage for gaming workloads. There are gains in efficiency and those are meaningful, but for someone like LM or myself, there's little meaningful difference between work completion between the 7950X and the 9950X.

Probably still gonna buy a 9950X though.

 

[LunarMist]

I'm not reinstalling Windows; that's for sure.

 

[Handruin]

Definitely feels underwhelming given earlier claims from AMD and the dollar/value to the previous generation is not great. Overall it's nice to see the very large gains when compared to my 3950X but I'll still be waiting for X3D versions before considering an upgrade.

 

[Mercutio]

There's some indication that Linux sees a lot more performance gain on Zen 5 than Windows does just because kernel level improvements are already present there that haven't happened on Microsoft's side.

 

[Handruin]

This guy does a decent job covering Linux benchmarks... almost too much details

AMD Ryzen 9 9950X & Ryzen 9 9900X Deliver Excellent Linux Performance Review - Phoronix

www.phoronix.com

 

[LunarMist]

Definitely feels underwhelming given earlier claims from AMD and the dollar/value to the previous generation is not great. Overall it's nice to see the very large gains when compared to my 3950X but I'll still be waiting for X3D versions before considering an upgrade.

I had the strong feeling that Zen Part 5 was going to be a clunker, though I was not expecting it to be that bad.
I'm not convinced that the X3D version will be any good for the vast majority of programs, at least not in the asymmetrical kludge configuration with the same old IO die.

The options now are to buy a 9950X for myself and use the 7950X in the rebuild, or to keep the 7950X and get something else for the rebuild. I'm thinking this nonsense of CPU cores not playing well together solidifies the decision to move away from it. It makes the most sense to get the 9700X for mATX and crank up the PBOs in the BIOS. The only reason I was doing the 7950X -> 9950X upgrade was the last resort to improve myself under Windows 10.

We'll see what iNtel does with the Arrow Lake, but APPLE is looking better and better as iNtel and AMD are crapping out.

 

[LunarMist]

There's some indication that Linux sees a lot more performance gain on Zen 5 than Windows does just because kernel level improvements are already present there that haven't happened on Microsoft's side.

But if Windows does it will almost certainly be under 11.

 

[Mercutio]

This guy does a decent job covering Linux benchmarks... almost too much details

AMD Ryzen 9 9950X & Ryzen 9 9900X Deliver Excellent Linux Performance Review - Phoronix

www.phoronix.com

From the conclusion of the piece: "In total I ran nearly 400 benchmarks across all the CPUs. When taking the geometric mean of all the raw performance results, the Ryzen 9 9950X came out to being 17.8% faster than the Ryzen 9 7950X. The Ryzen 9 9900X meanwhile was 21.5% faster than the Ryzen 9 7900X across this wide mix of workloads. The Ryzen 9 9950X was 33% faster than the Intel Core i9 14900K performance overall and even the Ryzen 9 9900X was 18% faster than the Core i9 14900K. For those still on AM4, the Ryzen 9 9950X was delivering 1.87x the performance of the Ryzen 9 5950X processor. These are some great gains found with the Ryzen 9 9900 series."

All of that does make me think there's something fucky going on with Windows.

I'm thinking this nonsense of CPU cores not playing well together solidifies the decision to move away from it

The nonsense of cores not playing well together specifically involves gaming. One of the two CCXes in the bigger Zen5s is slightly more capable than the other one, so it is beneficial to make sure all your lightly threaded tasks run on that guy instead of trying to spread them across the other CCX as well. It's not even REALLY a big deal for gamers so long as they're happy with their performance and don't care to fix it, but it matters since 90% of the benchmarks that are being released are gamer-focused.

 

[LunarMist]

From what I understood even the regular (not X3D) 9000 series with two symmetrical CCXs underperforms because there is a drastically greater latency between the two and Windows drivers are suck. Is that the problem that is dragging down the non-gaming performance also or is it some other stuff?

 

[Mercutio]

The understanding I have is that there is more or less never perfect parity between the CCXes, so Windows does a bunch of tests and determines which of the two is better when it needs to favor fast, lightly threaded workloads. Given that the hardware is drastically better on Linux than on Windows, it's pretty clear that something is wrong in the software stack but since we don't have any kind of source access to anything, we'll be lucky if we ever find out anything more about it. Maybe the 25H1 Windows 11 Update will have some fixes that ameliorate the issue.

 

[LunarMist]

The slight difference between the two CCXs (without any lopsided gamer-cache) is probably less than the delays of switching around more often than necessary to distribute heat. It won't be 5 or 10% for sure. That's like some bad driver on the highway constantly changing lanes and getting nowhere fast. And why was it not an issue with the 5950X or 7950X? AMD had 2 years to plan the 9000 with the MS.

It's not clear to me how many of those Linux tests are equivalent to Windows or how many of the applications themselves rather than the Linux OS are just better. Obviously the AVX-512 speed is much better in the 9000 series, but sometimes that is skewing the results. I'm not in favor of the geomean value for the somewhat arbitrary tests.

 

[LunarMist]

This part about PPM is confusing from the Thompson hardware.

• The feature requires four components: a new chipset driver, updated BIOS, Windows Game Mode, and the Xbox Game Bar (make sure to update it through the Microsoft store). The Xbox Game Bar contains a KGL (known good list) of games that it detects when active, thus triggering Game Mode (you can also instruct the game bar to recognize unknown games and/or other applications as games). The driver communicates with Windows Game Mode to trigger the AMD PPM Provisioning File Driver (installed with the chipset drivers) to park the cores on a single CCD, thus constraining latency-sensitive workloads (like games) to the higher-performance chiplet.
• AMD didn’t tell reviewers this feature was active in the new chipset driver until late in the review process, which was problematic. As we’ve covered in the past, the core parking feature has a major problem: It can’t be uninstalled from the operating system. As such, if you later install another processor but use the same operating system, the feature will persist and can continue to park cores (potentially unbeknownst to the user), thus hamstringing performance with processors that aren’t designed to use the feature. We remarked back in April 2023 that it was ‘almost unbelievable’ that this known issue exists, and it is even more unbelievable that it still exists 16 months later, in 2024.
• If you swap from a dual-CCD chip to a regular processor, you must completely reinstall Windows. Additionally, we've heard reports that upgrading from a standard single-CCD model to a dual-CCD model could also require a complete reinstall, an unnecessary and quite irritating situation for end users who might not even be aware of this requirement.

If the four conditions are necessary for the PPM and you don't have them, then why would it be a problem if you swap around from 7000/7000X3D/9000 series in any order after the OS is installed? I planned to just drop in the new 9000 CPU and reboot, assuming that Windows will work. Why exactly does MS require reinstallation? If it is just that the PPM is active or inactive in the wrong situation, that should not be an issue for users that explicitly want to compute and not play games. It's not like a no boot, BSOD, or activation situation. Is the ambiguity just a poor language translation/writing issue?

 

[Mercutio]

I think the idea is that you'd leave some performance on the table by not parking a CCX when running applications that are lightly threaded, but associating all of that with "game mode" leaves me wondering whether or not only games can benefit from doing so, or if there's a broader implication for poorly optimized applications (e.g. Luminar AI, which does not seem to hit more than four cores and also won't do a damned thing with a GPU in spite of its name). Can we manually add applications to the KGL?

I'm ruminating over a 9950x and the 7950x. If I'm getting 95% of the performance at 80% of the price, the cheaper guy is the better deal, right?

 

[Santilli]

I think the idea is that you'd leave some performance on the table by not parking a CCX when running applications that are lightly threaded, but associating all of that with "game mode" leaves me wondering whether or not only games can benefit from doing so, or if there's a broader implication for poorly optimized applications (e.g. Luminar AI, which does not seem to hit more than four cores and also won't do a damned thing with a GPU in spite of its name). Can we manually add applications to the KGL?

I'm ruminating over a 9950x and the 7950x. If I'm getting 95% of the performance at 80% of the price, the cheaper guy is the better deal, right?

I resemble that remark. Your information has certainly made me happy with my choice. I do wonder if I had waited a bit, will the 7950X3D take a huge price drop?

 

[Handruin]

I was hoping to see a bundle deal at my microcenter but they list the 9950X at $649 MSRP. Maybe those deals are no longer a thing. That could make it more approachable when compared to the 7950X

 

[Handruin]

Also notably different is that when I tried buying my 3950X on release, it took weeks to get one because they were always sold out. Now the 9950X shows 25+ in stock

 

[Mercutio]

So far as I can tell, the Xbox KGL is not something end users can control, so it's probably games-only and worse, there's a record of games that should be on it but aren't because of software updates on one side or the other.

 

[LunarMist]

Are you saying that the XBOx Kegl is automatically installed, even in Win 10? I don't have the XBOX or other games humanly installed.

 

[Mercutio]

Xbox Game Bar is a standard part of any Windows 10 or 11 installation, even if you never bother to use it.

 

[LunarMist]

I was hoping to see a bundle deal at my microcenter but they list the 9950X at $649 MSRP. Maybe those deals are no longer a thing. That could make it more approachable when compared to the 7950X

Nope. There are some specific deals, but not with 99x0X. There is just the usual $20 off for adding a motherboard. I decided not to buy a 9950X for now. Maybe I'll decide what to do this weekend. 9950X or 9700X, that is the question.

 

[LunarMist]

Xbox Game Bar is a standard part of any Windows 10 or 11 installation, even if you never bother to use it.

Bastards. Is there is no way to disable it, e.g., in the Registry?

 

[Mercutio]

You can uninstall it through Programs and Features, just like anything else.

 

[LunarMist]

But it's not there. Is there some secret command to identify the status?

 

[Mercutio]

You may have already uninstalled it. I habitually remove most Windows Store apps from my installation media but I suspect most people will have it.

AMD has already indicated that it's going to ship a BIOS update for Zen 5 to allow them to run at the Zen 4 TDPs and pick up better performance. That's still not a massive uplift and it does nothing for the 9950x, but it'll make the smaller guys more appealing at least.

 

[LunarMist]

It would be good to get more out of the 9700X, but I'm not hanging any hats on "Leaker Chi11eddog on X tells us that both the 9600X and 9700X could get their TDP upped from 65W to 105W with an incoming BIOS update." Maybe he is buddies with the silly BrianSlug.
I'm thinking that 9700X CPU is my best bet for now. Surely with only one chip it will not be affected by the CCD and XBOX bug.

I will also buy the new flat-faced HSF for the 7950X to maximize its capabillity. As of now I can use PPO Enh 2 stably for continuous use, but the D15 is at the 95°C limit with all cores in SMT and 92-95°C without SMT.
Maybe I'll get a 9950X or something with the X3D later on for X670E or maybe I'll totally pass.

 

[Mercutio]

There are leakers with very good reputation. If you're looking for that information, you'll see the same names come back up for specific types of info.

Everyone is going to shit their pants over the X3D products next month because gaming is the only thing the majority of enthusiast media cares about. I'm sure it'll be an "OMG AMD redeemed!"narrative. They won't be wrong because X3D does its one trick very well but I sincerely hope AMD addresses Zen 5 performance more generally.

 

[LunarMist]

But isn't it going to have the same old lopsided cache? I suspect we will need Zen Part 6 for a significant improvement and it will require a new mainboard with faster RAM and fewer free PCIe lanes yet.

 

[LunarMist]

I'm not an expert in the "Leakers" theory, so maybe that guy has a PhD in leaking; I would not know. It sure would be easier if they just jacked up the power in a simple BIOS setting, but even without it the PBO should make the 9700X a good CPU for a few years. I'm hoping for about sqrt(2)x improvements over the 57000G.

 

[Mercutio]

But isn't it going to have the same old lopsided cache? I suspect we will need Zen Part 6 for a significant improvement and it will require a new mainboard with faster RAM and fewer free PCIe lanes yet.

Gamers want lopsided cache. As far as gamers are concerned, there's no other reason to have PC. Even the 5800X3D is an amazing chip for gamer needs and that thing is now two generations old. The PCIe limits aren't an issue for those gamers, either. All they want is as much GPU as they can afford anyway. It's hard enough to justify a full ATX motherboard to those people.

It doesn't sound like PBO is making a huge difference for anything Zen 5 right now but the difference between Linux and Windows performance gains IS quite puzzling. Some analysis says that AMD didn't make any changes from Zen 4 to Zen 5 in its memory controller and others suggest that Windows is seeing doubled latencies between the CCXes on Zen 5 vs Zen 4, so there are clearly places where some part of the architecture has fallen down.

I'm reading everything I can find because I want to justify the CPU switch from the 7600x to the 9950x for my presumptive new desktop, which means I AM following along with Twitter and Reddit nonsense right now but I have compatible hardware and it's not going to kill me to not spend $600whatever until there's some sign of improvement.

AMD does have a habit of releasing refined products after a while. I can't imagine that they were rushing to ship something new, but I do kind of wonder why they put something out knowing full well that it would be AMD's version of the fourth generation that Intel released a 14nm-based CPU.

 

[LunarMist]

Supposedly the max PBO is worth 10-15% for the 9700X and less for the others.

 

[LunarMist]

How do we boot with the old BIOS? CPU not in or out?

 

[LunarMist]

The BISO flashing is the bullshit, so I must swap the CPUs.
Meanwhile the new low convexitity Noctuals is HUGE, i.e., even taller than the NH-D15.

 

[LunarMist]

So here we go with Passmark 11 of the R7 9700X default settings with 32GB EXPO DDR5-6000.
CPU single-thread: 4726
CPU Mark: 37481
The single thread is quite good. We will now attempt to jack up the power somehow in BIOS...

 

[Mercutio]

Just install Ryzen Master and IIRC it's just a one click "enable PBO" button.

 

[LunarMist]

I enabled PBO and also a few other changes in the BIOS. The PBO only makes a slight difference. The main difference is separately selecting the TDP, which is not useful over 95W (142W peak) since it is hitting 95°C on the low profile cooler. With Enh 3 also on the PBO, it is a bit less than +8%. Still, over 40K pasamarks is pretty good from 8 cores. Multi-threaded speed is not the main value though. The single-threaded performance is 10% faster than the 7000 series at any power level. I could get more out of the 9700K with the D15 and further tuning, but am sick of working on computers today. My main system is in pieces since I had to take the 7950X out to update the BIOS on the other board.

 

[Mercutio]

The first rule of taking one's own computers apart: Never take all your best buddies down at the same time.

 

[LunarMist]

The first rule of taking one's own computers apart: Never take all your best buddies down at the same time.

I was really surprised that the CPU-less, RAM-less, push the button and wait for the update from USB did not work. I have used it before on other boards.
Fortunately I still have the 5950X system and of course I'm typing this from the awesome little 8840U, both of which have full access to all the NAS.

 

[Mercutio]

I've found that those systems can be really finicky, for example not working with a larger than 64GB drive even when only the first 1GB was even formatted.

 

[LunarMist]

I tried three thumb drives including two that were 32GB FAT32 in case that was an issue. There were no issues with the BIOS when I got in there, but there is no reset BIOS button on the back as on the better boards.

 

[LunarMist]

Noctua has some updated fans and coolers on the horizon with options to adapt better to the AMD and Intel heat spreader differences. They're also working on a vapor style radiator cooler with no pumps, so it should be more reliable than an AIO...but also they're updating their massive air cooler and fans to compete with AIOs.

I finally got to install the NH-D15 G2 with the flattened base on the old 7950X. As a replacement for the Nh-D15 it's just a little better, allowing the CPU to use about 8 watts more at 90+°C. That reduces the throttling so the PaasmaRK 11 results are better on the PBO. The overall value is minor as an upgrade but makes more sense in a new build or if you have something less than the D15. I'm not hearing any rattles yet.