PC Processors

[waltc]

I own an MSI Gaming Pro Carbon AC motherboard, with an R5 1600 (perfectly stable @ 3.8GHz, with 2x8GB DDR4 @ 3200Mhz) and I've noticed that when running the latest MSI bios (2.A0) that every single cpu benchmark contained in AID64 Extreme (or Engineer) slows down by 5%-15% when compared to the performance of the same version of AIDA64 cpu benchmarks when running the bios that preceded it, 2.8.  I have verified this several times, btw, in subsequent reflashes back and forth between bios versions. All UEFI bios settings are the same--everything in the software environment is the same between bios versions. Currently, I am back on 2.8 because of this.  CPU Benchmarks like Cinebench 15.x, which I gather use no OOSE, do not seem to be affected by the Spectre2 microcode in terms of cpu performance, however. (Obviously, I am not concerned only with how the AIDA64 cpu benchmarks perform, but rather with how any similarly written cpu application may perform.)  Example: the AIDA64, 5.97.4600 version, running the Memory Read bench repeatedly shows 2.8 (pre-Spectre2 microcode) 49,xxx mb/s, whereas Memory Read under the 2.A0 bios shows 40,xxx mb/s.  Big drop there.

Under 2.A0, InSpectre tells me that I have full coverage against Spectre, including cpu microcode.  Under 2.8, same version of InSpectre (#8) tells me that while the OS portion of the Spectre defense is present, the cpu microcode is not, and so it equates to vulnerability to Spectre2.  That is how I know that the 2.A0 bios is installing the Spectre2 microcode.

Any comments?  Do the 12nm Ryzen + cpus display this behavior, or is the performance factor mitigated with the x470 chipset and Ryzen +?  Thanks for any insight!

Note: Although the normal MFlash method of updating MSI x370 bios versions will not let the user flash back to an earlier version, MSI provided me with a bios utility that allows me to flash to whatever bios version I want.  Here's a link:

Flash tool[1.05] for MSI UEFI BIOS [from UEFI Shell]

I actually prefer this method to MFlash.

[_blackbishop_]

Turn off "DRAM power down" or whatever it is called in your BIOS.

There is no slowdown on ASUS mobos with AGESA PinnaclePI-1002a so it is not spectre fix. It change mentioned above does not help than ping MSI forums for help.

[waltc]

I always turn off the DRAM power down feature, although I certainly wish that was the case--it would be so easy to fix...;)  As I mentioned, with the exact same bios settings I get a definite slowdown with the AIDA64 cpu benchmarks--every single one of them.  Other cpu benchmarks like Cinebench 15.x run without negative performance from the Spectre2 microcode, and It is perfectly repeatable between reflashings.  The Spectre2 fix for my MSI mboard is of course going to be different from your ASUS board--not the Spectre2 fix itself, but the implementation of it in the MSI bios code, of course.  MSI may well be doing something non-optimal here.

Your post does give me some optimism, however; the latest MSI bios (2.A0) includes the 1.0.0.1a AGESA, the one with the Spectre2 microcode fix, not the 1.0.0.2a AGESA version you are using in your ASUS board, so perhaps this will be fixed next bios release from MSI. (That would be nice!) Bios 2.8 runs perfectly--just not with the Spectre2 microcode.  As there apparently are *no* pieces of malware loose in the wild that exploit Spectre2 at this time, I'm not worried about it...;)  Still, I would like an implementation of the Spectre2 patch that wouldn't slow down these particular cpu benches...at least not to the extent I'm seeing with the latest MSI bios 2.A0.

Who knows?  Was really hoping to get some feedback from AMD here.  But thanks very much for your input--it's a head-scratcher, indeed.  I've posted in the MSI forums and no one knows anything there, either.  Maybe I'll send AIDA a line, too.  Thanks again!

[daybreak]

Windows10 ver1709's hotfix KB4073757 includes Spectre Variant2 fix for AMD too.

Did you receive KB4073757 through Windows Update ?

[waltc]

No, but I have the Windows fix installed already--I'm running the latest version of Windows, version 1803, build 17134.5, atm (officially released by Microsoft tomorrow, 4/30, to the public.)  When I disable the software fix for Spectre2, via Inspectre:

GRC | InSpectre 

It does absolutely nothing to improve the performance of the above-mentioned cpu benchmarks (which also indicates that the Windows AMD Spectre2 fix does nothing to hurt performance, etc.)

From what I've read so far, it appears the Intel cpus are receiving their cpu microcode fixes through Windows Update, but for AMD the only vulnerability is Spectre 2, and it requires *both* a microcode update through a bios delivery and the Windows software fix. I think the bios-level microcode delivery is far better since the Windows cpu microcode fixes must be reinstalled if you install a different OS, like a newer version of Win10x64, or Linux, etc.  With the bios microcode fix, it doesn't matter which OS you install, you have the CPU portion already present and installed.

Here's what Inspectre says when I have the 2.A0 bios installed, quoted:

"*This system's present situation:

*This 64-bit version of Windows has been updated for full awareness of both the Spectre and the Meltdown vulnerabilities. If the system's hardware (see below) has also been updated, this system will not be vulnerable to these attacks.

*This system's AMD processor is not affected by the Meltdown vulnerability and it has been updated with new features required to allow its operating system to eliminate the Spectre vulnerabilities and/or to minimize their impact upon the system's performance."

(The actual InSpectre text is dark green, so I got it as close as I could with this editor's text-color choices...;))

When I have the 2.8 bios installed, Inspectre says that I am *not* protected from Spectre 2, because while I have the Windows10 portion installed, the cpu microcode is not.

[elstaci]

Do you have the "GAME BOOST SWITCH" on in BIOS. That might help improve the performance of the CPU if your CPU can be OCed. :

[waltc]

I think you maybe misunderstood my OP...;)  (But I appreciate your reply, anyway.)  In my first post, you'll notice I am already running at 3.8GHz without a problem.  At the moment, the only thing showing a concrete and repeatedly demonstrable slowdown when the cpu microcode is applied via the 2.A0 bios, are the CPU benchmarks in AIDA64 Extreme and Engineer, version 5.97.4600--every single one of them.  Cinebench 15.x does not show a slowdown, nor does the cpu benchmark included with CPU-Z, version 1.84 x64, nor does the Valley benchmark, or the Heaven benchmark slowdown, and I can detect no games slowdowns thus far.  With the 2.8 bios--which installs without the Spectre 2 CPU microcode--performance is anywhere from 5% to 15% better, depending on which of those AIDA64 benchmarks I run.  The problem is not that the cpu itself slows down in terms of MHz because that isn't happening--what is happening is that on these CPU benchmarks the slowdowns are sizable in processing speed (not MHz. IP x cpu MHz = processing speed, more or less.)  As mentioned in the OP, all bios settings and clocks are the same between bios versions, including the the cpu MHz speed of 3.8GHz. 

The Spectre2 microcode seems to be be negatively affecting the Out of Order Speculative Execution performance of Ryzen--but I cannot of say with absolute certainty that this is a problem. For instance, the AIDA64 Memory Read benchmark above shows 49,000+ MB/s under the 2.8 bios (no Spectre2 cpu microcode) and shows 40,000+ MB/s under the 2.A0 bios, which installs the Spectre2 microcode cpu patch.  Again, all clocks, bios settings and OS settings are the same when running both bios versions.  I think it will take someone from AMD to definitively answer my query here, unfortunately. 

[elstaci]

Sorry about that, I thought maybe the CPU would perform better with that BIOS feature enabled.

Anyway, It has been mentioned that the fix for Spectre might slow down a computer. as per this link: AMD processors will get a firmware update for the Spectre CPU exploit, but it's optional | PCWorld :

When Meltdown and Spectre were uncovered, AMD said that its CPUs aren’t vulnerable to Meltdown, the first Spectre flaw would be resolved via operating system and software updates, and “differences in AMD architecture mean there is a near zero risk of exploitation” of the Spectre second variant. Nevertheless, while AMD CTO Mark Papermaster says “we believe that AMD’s processor architectures make it difficult to exploit Variant 2,” AMD now plans to patch its CPUs against it.

AMD will start releasing “optional microcode updates” (emphasis ours) for Ryzen, Threadripper, and Epyc processors to its partners this week, with firmware updates for older processors coming in the future. Like the Intel firmware updates, you won’t get AMD’s directly from AMD itself. Instead, you’ll need to download them from the maker of your PC, laptop, or motherboard through their support pages.

But you might not want to be so quick to do that. When Microsoft said the Meltdown and Spectre fixes could slow down PCs, it specifically said Spectre’s second variant is the only one that requires a CPU microcode update—and that the fixes for the second Spectre variant are the ones that can affect your PC’s performance. Rushed firmware updates can be finicky too. Intel’s CPU fixes are making some Haswell- and Broadwell-based computers reboot more often, for example.

If AMD says its processors have near-zero risk to the second Spectre variant and that these CPU updates are optional, you might want to hold off installing them until their potential performance impact has been tested and benchmarked in the wild, especially if you’re on an older PC or running Windows 7 or 8. Those systems suffer bigger performance hits with microcode patches, Microsoft says.

[waltc]

Thanks again for the comments...;)  Yes, the base clock for the 1600 is 3.2GHz; setting the "game boost switch" merely increases the clock to 3.4GHz, which would make the cpu run faster by 200MHz, if I was using default clocks.  The switch only functions at the base clock of the cpu--but thanks for the effort, anyway. At 3.8GHz it has no effect (this cpu won't do 4GHz, anyway, no matter what.) I am perfectly satisfied with the MHz clocks of the cpu. I am perfectly satisfied with the whole Ryzen shebang, actually...;) (Just in case someone might view this thread as a complaint about the hardware--it's not that at all.)  The slowdown only occurs because the Spectre2 microcode literally slows down the processing power of the cpu doing certain tasks--while it has no effect at all on other tasks, it appears.

What bothers me is the total lack of commentary by MSI about this.  There is nothing in the notes MSI supplies with the 2.A0 bios as available on their site(s) (US and Global)  that even mentions that the Spectre2 microcode is a part of the 2.A0 bios--I had to discover that myself, and I just stumbled upon it, really.  The InSpectre software is what validates the microcode condition of the cpu between the bios versions--the OOSE slowdowns (if that's what is happening) directly correlate only to the 2.A0 bios.

The problem is that if I want to purchase a newer Ryzen+ cpu (which I may do, actually) I will need the newer AGESA code for the Ryzen + cpu, for this MSI motherboard, should I decide to keep it, and so atm I have no choice but to install the bios version from MSI that includes the Spectre2 microcode--and I do not know what the effect is, if any, on the Ryzen+ cpus (2600, etc.), etc.  I would prefer MSI release two bios versions each bios release, one with the Spectre2 microcode, and one without, so that installing the Spectre2 microcode would truly be optional--as I need the cpu performance but am not convinced the Spectre2 slowdown is something critically necessary to my system and work environment. I agree with AMD that installation of the Spectre2 microcode should be optional, but it isn't, and that's why I would like more info from someone about this.  MSI would be the proper respondent, I think, but I have not been able to get an answer from them as yet, even though I think the situation is rather easy to understand.

Another thing that is of concern, is that while AMD has released the 1.0.0.2a AGESA code commercially to the motherboard OEMs, so far MSI has seen fit only to release the 1.0.0.1a AGESA version for all of its x370 motherboards, while its brand-new x470 motherboards have all been updated to the 1.0.0.2a AGESA.  Be nice to have an explanation of that, I think--if there is any difference between the AGESA versions, and if so, what? Does MSI plan to release an x370 bios with the 1.0.0.2a AGESA?  That sort of thing...;)  Lack of information creates a poor buying scenario, imo.

[elstaci]

Yes, I noticed that about MSI and one or two other Motherboard manufacturers. They don't give any type of  Specs concerning any new BIOS update. If they do it is extremely general in nature. I wish they would give a list of improvements (Release Notes)  that each BIOS update does. That way a Customer can make a educated decision whether to update or not.

[elstaci]

This User on another Thread here in AMD Forum mentioned that certain Asrock Motherboards just uploaded new BIOS:

Good news if you're a user of certain ASRock motherboards. New beta UEFI/BIOS updates were released that include AGESA version PinnaclePI-AM4_1.0.0.3b, and it seems to resolve this issue. That's the case for my ASRock AB350 Pro4 with UEFI/BIOS version 4.73A/PinnaclePI-AM4_1.0.0.3b, atleast.

I realized that you have a MSI Motherboard, but this suggests that mostly likely there might be a new BIOS update for your motherboard very soon since Asrock updated their BIOS with the new AGESA 1.0.0.3b.

[waltc]

Thanks! Good to know...;)

[suhr]

You're being oversensitive for such a minor drop in something that doesn't matter as much as you think it does. Let me clear up some things. First off, these bugs don't affect systems too often. However, the fixes DO decrease the performance on EVERY setup. Let it be Intel or AMD. You're looking at some MINOR losses from memory bandwidth. Of course Unigine Valley won't show any difference, it pretty much purely benches the GPU, and if you don't have ANY issues with any games or other benches, you shouldn't be going back and forth between BIOS'es. It makes no sense.

I have a secondary rig here that has a 2600 which I use two sticks of Trident-Z 3600 pulled out from my X299 primary, along with an MSI X470 Gaming Pro board. At 3.8 MSI Gaming Boost, it gets 125 pts in UserBenchmark floating point single core, while my 4.9ghz 7820X gets about 152pts. It's a pretty capable CPU and a %10 knock down in RAM bandwidth certainly ain't gonna kill it (if it even happens).

You don't have anything to be worried about. Have the patches installed and keep rocking.

[waltc]

Yep, I have a tendency to be "over sensitive" when information is not forthcoming from my hardware vendor...;)  The more info they provide their customers, the better, imo.  Also, I don't think a ~19% reduction in memory bandwidth as reflected in the AIDA64 cpu bench Memory Read is "slight," or "minor," etc.  I think it deserves at least a comment or two.  (The Spectre2 cpu microcode patch drops the mem bandwidth performance on my system from 49,000+ mb/s  to 40,000 mb/s.)  Especially in the case of Ryzen, because at least up to 3400MHz (if not beyond) Ryzen will take advantage of the faster bandwidth to process faster--unlike some cpus, for instance, that don't see much of a processing speed increase as the memory bandwidth increases beyond their default.  The thing that bothers me the most is that AMD officially states that Ryzen isn't vulnerable to either Meltdown or Spectre1 at all--thus requiring no patches for them--and that even with Spectre2 the company sees the vulnerability of Ryzen to be slim to none and has officially stated that installing the Spectre2 patch is optional as far as they are concerned.  So, it would *nice* if MSI would offer bios updates for Ryzen which make the Spectre2 patching optional.

I will agree with you, however, that how much this all actually means to someone will certainly depend on the individual and his work environment...;)

[suhr]

I'm running my Trident-Z 3600 at 3333 with the Ryzen 2600. Windows crashes at 3600, so no, Ryzens don't reach to those kind of memory speeds. Besides: AMD Ryzen Memory Analysis: 20 Apps & 17 Games, up to 4K Review | TechPowerUp

So much for this "faster" with faster memory theory... 4 FPS going from 2133 to 3200.

Like I said, you don't need to be worried about it.

[waltc]

Heh...;)  Sorry, but no...;)  Windows crashes because your ram can't do it--not Ryzen.

These people almost hit 3800MHz:

DDR4 Memory Scaling Performance with Ryzen 7 2700X on the AMD X470 Platform - Legit Reviews

Besides, you missed my point entirely, which was that the faster the memory bandwidth, the faster the Ryzen architecture runs--which is why you want to install the fastest compatible ram you can afford, and why you do not want to see memory bandwidth slow down.  Check out the above article for your edification...;)

[suhr]

You clearly don't get it. It has absolutely NOTHING to do with my RAM. The RAM I had there was Trident-Z 3600, Samsung B-die. Fastest RAM you can buy and I've been running them on my X299 at 3600 for ages. My 2600 simply can't do it, I don't care what they reached.

No, you don't want to "install the *fastest* compatible RAM you can afford". That's flat out pointless and would burn a hole in your pocket, and that "memory bandwidth SLOWS DOWN!" logic is just as nonsensical as you can clearly see that the setup is NOT bogged down by RAM bandwidth in games.

[waltc · Accepted solution]

UPDATE:     Got a new beta bios from MSI today 5/8 which seems to have solved the slowdown problem while retaining the Spectre2 vulnerability cpu microcode patch.  OK, this beta bios is looking very nice!  The Spectre2 patches are installed but the good news is the big slowdowns I had seen with 2.A0 were no longer evident!  Running just as fast with the new bios as the AIDA64 benches were running with the 2.8 bios!  Very nice, and very pleased with this.

If interested, you can download the bios version from this MSI link:  https://forum-en.msi.com/index.php?topic=303203.msg1759105#msg1759105

Further wrapping this entire issue up is the fact that MSI is even now releasing official bios versions (post betas) which correct the performance slowdowns the first Spectre2 bios fixes introduced. Thankfully, looks as if MSI has put this entire issue to bed--Spectre2 vulnerability vanquished *and* the initial performance slowdowns with it...;)  Again, good job MSI!...;)