Z370 new BIOS - DE/D6 Microcode - discussion

t0yz

Member
SECOND LIEUTENANT
Joined
Feb 1, 2012
Messages
113
Some fast testing:

- there is a performance impact still, observable in Cinebench R15 and R20 (the old score can be seen in the same picture), but also in SSD scores AnvilPro (970 Evo+ NVMe). The Anvil score is down circa 400-800 points (16200-16800 scores for previous BIOS with B4 ucode)
- the default Uncore/Ring multi is still 43, down from 44 (same as with CA). Setting back the uncore to 44 won't fix the performance gap
- AIDA64 scores seem similar
- CPUZ score seems similar
- AC Origins is running well and identical to past ucode

Album with screenshots since the forum hates attachments.
https://imgur.com/a/MEgpSby

I'm unsure what to say about the BIOS. It's still pretty damn bad. Performance is still meh, Uncore still wrong, and yeah, the Lite Load thing has been yet again modified (I think from 11 to 22 to 17 now), resulting in weird power consumption patterns which should be repairable with proper settings, but yeah, that's just more work.

You will also get new Intel ME firmware with this and other EFI modules:
11.8.78.3689 Intel ME version
17.7.0.4404 Intel RST EFI version

It's good that the AIDA64 and CPUZ benches are similar-ish and gaming seems fine. It's likely you won't notice much of a difference in day to day usage other than being annoyed by CB scores.
 

citay

Active member
SERGEANT
Joined
Oct 12, 2016
Messages
803
Thanks for testing. This also applies to Z390, my MEG Z390 ACE also has a new BIOS update available that lists the new microcode in the changelog (and nothing else).

But i'm also interested in the "something else", for example, have they fixed the laggy navigation and "freezes" every couple seconds within the BIOS? Is Command Rate 1T working again?

It would take a lot to get my off my BIOS from two versions ago, before they messed everything up with the first new microcode BIOS...
 

t0yz

Member
SECOND LIEUTENANT
Joined
Feb 1, 2012
Messages
113
Thanks for testing. This also applies to Z390, my MEG Z390 ACE also has a new BIOS update available that lists the new microcode in the changelog (and nothing else).

But i'm also interested in the "something else", for example, have they fixed the laggy navigation and "freezes" every couple seconds within the BIOS? Is Command Rate 1T working again?

It would take a lot to get my off my BIOS from two versions ago, before they messed everything up with the first new microcode BIOS...
Late reply since new forums.
My RAM never accepted 1T so can't comment on that. The BIOS nav was always fine for me, regardless of version. I've been on this latest BIOS since and things look normal. Without CB telling me something's wrong and HWinfo64 maxing Uncore at 43x, there would be no way to tell this from older BIOSes. However, i only have a 1070ti, so maybe somebody with something like a 2080 Super or above and playing at lower 1080p rez would be able to tell there's a lack of performance.
 

citay

Active member
SERGEANT
Joined
Oct 12, 2016
Messages
803
Late reply since new forums.
My RAM never accepted 1T so can't comment on that. The BIOS nav was always fine for me, regardless of version. I've been on this latest BIOS since and things look normal. Without CB telling me something's wrong and HWinfo64 maxing Uncore at 43x, there would be no way to tell this from older BIOSes. However, i only have a 1070ti, so maybe somebody with something like a 2080 Super or above and playing at lower 1080p rez would be able to tell there's a lack of performance.

For my Z390 ACE, a new BIOS version appeared today:

Version 7B12v19
Release Date 2020-08-19
Description
- Improved memory compatibility.

I'm guessing a lot of boards will get their own version of this update.

About the performance drop in the dreaded BIOS with the changelog "- Update Microcode. - Improved TPM function. ": I can notice it within one minute, just by running the WinRAR integrated benchmark, it shows a considerable drop in performance. I say considerable, because when i optimized my RAM, such a drop would come from using way looser timings for example. Like, you have to fight really hard to get such a performance increase by normal means, that you get from going back to the BIOS before that.
 

t0yz

Member
SECOND LIEUTENANT
Joined
Feb 1, 2012
Messages
113
For my Z390 ACE, a new BIOS version appeared today:

Version 7B12v19
Release Date 2020-08-19
Description
- Improved memory compatibility.
I have a feeling this will only get released for the boards that needed that RAM improvements/fixes.

Anyway, I have more (disappointing) data on the new DE ucode which got some enthusiastic/positive "reviews" online as being "faster" and the sort. Modded the latest BIOS for my board with it, and sadly it's basically the same as D6 and CA before or w/e MSI gave us. Default Uncore/Ring is still 43x instead of 44x, which is really annoying, you don't covertly underclock your CPU two years after release, vulnerabilities or not.

More importantly, performance is largely unchanged across the board, with a small bump in NVMe performance (my 970 Evo Plus seems quite sensitive to CPU speed in benchmarks).

R15 - near identical scores of around 1400-1410 multicore scores and 193 CBs for single core. With B4 microcode, the scores would be consistently and reliably +50 (1455-ish) on multi and around 198-200 on single.
R20 - near identical scores of around 3430 multi and 468 single. With B4 ucode, scores would be consistently 3480-ish and 475-ish.
CPUZ is one of the benchmarks that is not affected as much, if at all, by the new round of mitigations, so I'd get a 3820-ish multi and 515-520 single. With B4 the single core score was a bit higher, 525-ish.
AIDA64 is also somewhat unaffected. For the truly curious, I'll attach the benchmarks with D6 and now DE ucode.
The Anvil score for my 970 Evo+ is a bit better than with D6, one would consider it "back to normal" for the B4 era (16.2K vs 15.9K-sih). Anvil seems to fluctuate significantly with CPU speed, and OC on the 8700K to 5GHz would boost the Anvil score to 18K+, on par with some Gen 4 SSDs, so that's why I like to see how the NVMe does after big updates.
 

Attachments

citay

Active member
SERGEANT
Joined
Oct 12, 2016
Messages
803
Almost all the Z390 boards have gotten this new BIOS release now, see https://forum-en.msi.com/index.php?...lable-for-almost-every-msi-z390-board.345101/

I had to think for a bit that with R15 and R20 you meant Cinebench R15 and R20 😅 I thought it was something related to the microcode for a second.

So you're saying, the newest microcode is still not as fast as the last good one. Meh.

I did extensive testing with all sorts of benchmarks (primarily for the effects of RAM timings), and believe it or not, the simple WinRAR benchmark (pressing ALT-B in WinRAR) is by far the most useful one for me. It gives extremely fast results (like i said, under a minute) and picks up everything well, as it's a good combination of CPU and RAM performance. Sometimes you have to cancel the benchmark and start it again to get a realistic number, because Windows is often running something in the background right after booting. But with AIDA64 it's just the same. And with WinRAR you would immediately notice it, because the result is there so quickly and the number would be off drastically. All in all, you should give that a try, it should work in the trial version. I've tried AIDA64, Cinebench R15/R20, Geekbench 3/5, custom Handbrake batch file, just about everything. But i like WinRAR bench the best.
 

t0yz

Member
SECOND LIEUTENANT
Joined
Feb 1, 2012
Messages
113
So you're saying, the newest microcode is still not as fast as the last good one. Meh.
I didn't test in a while with WinRAR so I don't have the old ucode scores, with the DE version it's 18.7K or so (Ultimate Perf. power plan, Realtime priority for the benchmark process). No idea if it's good or bad.
You can put prio on Realtime for benchmarks to minimize the effects of Win10 doing background stuff and to get more consistent scores, but yes, overall the DE ucode seems similar to D6 which is similar to CA. Differences are so small it's hard to say if it's just normal variance or actual performance changes, but the difference from all these new 43x uncore ucodes to the B4 ucode from 2019 it's large enough that's no longer within statistical error.
I guess it's no longer relevant by now, the Z370 platform is 3 years old, and the architecture is even older, and even the newest Comet Lake chips are just Skylakes on steroids basically, so it's all a dead end, Intel would be foolish losing too much time to improve it. The good part is that some of these chips were decent enough at the time to still perform very well to this day so they have enough life in them and overclocking headroom to not be worth upgrading, at least for my case with design work and some gaming it's surely not worth changing the 8700K.
I guess I'll just mod the B4 ucode to a new BIOS eventually and stick with it.
 

Attachments

citay

Active member
SERGEANT
Joined
Oct 12, 2016
Messages
803
That score is quite good! My i5-9600KF with DDR4-3600 CL16 is getting 12,000 KB/s on there. Of course, it's a hexacore without HT, and yours is with HT. WinRAR likes cores, and also likes HT. It likes memory speed as well, i only had a score of 10,500 with DDR4-2666, and around 11,000 with DDR4-3200. It will also detect every impact of RAM timings, good or bad. Just a short & sweet little benchmark.

You really put this performance slump mostly down to the slower uncore multi of 43x? I thought it were the mitigations that caused the drop.
 

t0yz

Member
SECOND LIEUTENANT
Joined
Feb 1, 2012
Messages
113
The uncore multi fail doesn't have much of an effect on performance, because the 43x (or 44x in the past) is the top end figure you'll never see in a real world scenario or benchmark. You'l basically only see 44x on the Uncore if you have a rather lightweight OS without many background apps and leave it at idle. Preferable at High or Ultimate Performance power plans, as those will force the higher clocks. The moment you start a game or do some browsing, the clocks will drop as threads get used. In benchmarks or stresstesting, the 8700K will be at 43x for the cores and 40x for the uncore, at least for MSI, other vendors had the uncore as low as 37x since Intel never made crystal clear what the specs really are. It used to drop even further in the first 4-5 BIOSes, because MSI used to cap the package power at 95W as per Intel's spec, with a short 20s boost to 118W, which was not sufficient for even a default 8700k when it had to run prime95 with AVX at the horrendously high default voltages. I remember core clocks as low as 38x in those times, under stresstesting. MSI eventually had a change of heart and just slapped a 255W Auto setting for power limits even on my 4 phase cheap motherboard and left it at that.
Even if you manually set the uncore to 44x, which you can do in the BIOS, it won't fix performance. It will however be perceived as an OC and will require re-tuning the vcore as that'll get boosted, and from what I remember it also didn't clock up and down as smoothly as it would do on Auto, for example it would just stay at 44x for light workloads, and then drop directly to 8x when you were mostly idling, instead of the small, incremental changes it does on Auto.
So yeah, the uncore multi is basically a psychological annoyance. To be fair even the performance drop is mostly psychological, as you won't see it in games or most tasks.
 

citay

Active member
SERGEANT
Joined
Oct 12, 2016
Messages
803
Yes, you only tend to really notice performance drops when it's about 1/4 slower or so. But even if performance is "just" several percent worse, i don't like it. With the RAM timing optimization, i had to work really hard just to get a few percent higher performance. And now i'm gonna let it all disappear with a simple microcode update? I don't think so.

I'm still undecided if discovering all these theoretical flaws in the way (mostly Intel) CPUs handle things were good or bad. Yes, it's good to have flaws fixed, but for the most part, they were theoretical, which wouldn't have been found, had it not been for some people trying absolutely everything. Even now, with many flaws having been found, they're still rather exotic, and AFAIK are not used in the wild as attack vectors on home PCs. Simply because there are so many simpler and more effective ways out there than attacking the CPU.

Many CPUs, especially old ones, have gotten slower and slower with each fix, cause they disabled a lot of things that were useful for better performance. For mostly theoretical threats. That whole thing doesn't sit so well with me.
 

t0yz

Member
SECOND LIEUTENANT
Joined
Feb 1, 2012
Messages
113
There are no attacks in the wild AFAIK for any of these. Why bother when you can do it so much easier with other methods. Plus some require physical access, admin and so on. But what can Intel do, they cannot leave them wide open either, imagine the criticism. They have more than enough issues already. RAM tweaking is truly annoying, I had to tweak voltages on mine to make it stable at its XMP of 3000MHz CL16, and boy did it take a year of on/off over the night stresstesting runs, blindly basically, as I assumed that I needed more voltage, as it's usually the case, yet my dumb Spektek ICs on the RAM hate anything above 1.33V or so and will eventually BSOD, or worse, just give your games and apps crashes you won't know what's the cause for.
 

citay

Active member
SERGEANT
Joined
Oct 12, 2016
Messages
803
Hehe yeah, RAM testing is the worst, takes so much time for every little change. My Samsung B-Die like voltage, i have them at 1.38V, barely over XMP voltage (XMP being disabled of course, raises SA and IO voltages too much). I keep SA at 1.060V and IO at 0.060V. That's enough for 3600 CL16-16-16-34 CR2T.
 
You must log in or register to reply.
Top