AMD boost (beta) fix - tested by Toms

This thread is closed to replies: https://hardforum.com/threads/amd-a...000-boost-clocks-hits-september-10th.1986162/ So we might as well use this one since Rangers is also closed (I don't see any other threads on this either)

Tom's hardware said:
In either case, we do see some improvements that fall in line with our expectations for the Ryzen 7 3700X, but we also spotted an odd performance regression with the Ryzen 9 3900X, indicating this firmware is a work in progress.

Beta, but work is being done. Good sign I'd say.
 
I installed the beta fixed bios on my X570 Aorus Ultra. Sadly, this did not completely correct my boost clocks. I say completely because I am seeing slightly higher boost clocks. Before I'd see 4250 max, now I'm seeing 4333 max, but on a core other than the designated fastest core.

It did, however, lower my idle voltage which dropped idle temps a good bit.
 
if you are using an offset voltage then your boost clocks would be affected.
 
I installed the beta fixed bios on my X570 Aorus Ultra. Sadly, this did not completely correct my boost clocks. I say completely because I am seeing slightly higher boost clocks. Before I'd see 4250 max, now I'm seeing 4333 max, but on a core other than the designated fastest core.

It did, however, lower my idle voltage which dropped idle temps a good bit.

Did you clear your cmos after installing and load defaults/optimized settings, make sure all the boost features are turned on, as well as make sure your memory settings are correct in bios after flashing? I know from reading overlockers.net in the crosshair vii hero section, that there are some settings in the memory that if not set correctly will cause boost clocks to be lower, even when set to auto. (keep in mind just setting to use the XMP profile in the bios does not correctly set all proper settings, as that is an intel specification and is missing some of the AMD specific settings, and is why it is recommended to manually set all memory settings). Just trying to give you suggestions to hit the correct speed. I don't know if it helps.
 
Cool people will be happy. Of course as Tom points out actual performance is really unaffected.

As expected a 1-2% (at most) bump in a single core boost clock means a big whooping 0.5% bump in performance.

I'm not saying this isn't something AMD should fix up... if for nothing else to avoid stupid law suits. Its just such a non story, makes me almost hope AMD does just bring back PR numbers. It would be so much easier to just say this is our 12 core 5000+ and this is our 8 core 4600+ and avoid sillyness. Of course having this stuff end up in Intel marketing PP decks is pretty damn funny.
 
Cool people will be happy. Of course as Tom points out actual performance is really unaffected.

As expected a 1-2% (at most) bump in a single core boost clock means a big whooping 0.5% bump in performance.

I'm not saying this isn't something AMD should fix up... if for nothing else to avoid stupid law suits. Its just such a non story, makes me almost hope AMD does just bring back PR numbers. It would be so much easier to just say this is our 12 core 5000+ and this is our 8 core 4600+ and avoid sillyness. Of course having this stuff end up in Intel marketing PP decks is pretty damn funny.

The point was not to fix the chips 25-50mhz below advertised boost. The point was to fix the chips 300 to 500mhz below advertised boost.
 
To those stating these threads already exists they dont, at least not any more. I made this thread with the sole intent of it being new and fresh.

This is new news. Its news because the fix is actually being tested. Discuss that. Dont fight over who's definition is right. Use AMDs definition and discuss Toms findings.

It's that simple. The other threads were locked because people want to argue over semantics.

Now...
Lots of people not paying attention to their base clocks. Some boards run 99mhz others 101+
This will affect your final highest clock. Keep that in mind.

Tom's did not keep that in mind. That's a flaw worth noting I think.
 
The point was not to fix the chips 25-50mhz below advertised boost. The point was to fix the chips 300 to 500mhz below advertised boost.

That is fair guess when the end user versions drop we'll see if that has happened. My guess is still allot of people that are really seeing 500mhz below the single core boost have stupid setups or something. I can see AMD selling chips missing by 200... I suspect people reporting more then that are really just doing something stupid, or got a chip that got dropped or left in a shipping crate that got stupid cold or hot moist or something. I don't know imo if your really missing the boost by 500mhz perhaps make sure your outtake fan ins't backwards or something. lol (ya I'm half joking) AMD seems to be way to good at binning their own stuff for me to believe 500mhz misses are wide spreed or even legit outside of the odd damaged part.
 
Did you clear your cmos after installing and load defaults/optimized settings, make sure all the boost features are turned on, as well as make sure your memory settings are correct in bios after flashing? I know from reading overlockers.net in the crosshair vii hero section, that there are some settings in the memory that if not set correctly will cause boost clocks to be lower, even when set to auto. (keep in mind just setting to use the XMP profile in the bios does not correctly set all proper settings, as that is an intel specification and is missing some of the AMD specific settings, and is why it is recommended to manually set all memory settings). Just trying to give you suggestions to hit the correct speed. I don't know if it helps.

Yes, cmos cleared, flashed, all settings returned to where they should be, RAM timings manually set.
 
it is recommended to manually set all memory settings

I've seen this advice recently. Does it mean just the big 4 timings that you usually see specified (e.g., 14-14-14-34 3200MHz) or all the secondary and tertiary timings as well? If the latter, how are you supposed to find out what those are?
 
That is fair guess when the end user versions drop we'll see if that has happened. My guess is still allot of people that are really seeing 500mhz below the single core boost have stupid setups or something. I can see AMD selling chips missing by 200... I suspect people reporting more then that are really just doing something stupid, or got a chip that got dropped or left in a shipping crate that got stupid cold or hot moist or something. I don't know imo if your really missing the boost by 500mhz perhaps make sure your outtake fan ins't backwards or something. lol (ya I'm half joking) AMD seems to be way to good at binning their own stuff for me to believe 500mhz misses are wide spreed or even legit outside of the odd damaged part.

I was missing around 400 for a month until ABB. Some still are.
 
I've seen this advice recently. Does it mean just the big 4 timings that you usually see specified (e.g., 14-14-14-34 3200MHz) or all the secondary and tertiary timings as well? If the latter, how are you supposed to find out what those are?

So the easiest way to do this is to use a program called Ryzen Dram Calculator. https://www.guru3d.com/files-details/download-ryzen-dram-calculator.html
Step 1 is to find out the specifics of the chips in your specific dram using a program like Thaiphoon burner. Once you have done this you plug your ram into the calculator and its spits out some timings. the calculator has "Safe" and "fast" settings for you to try. start with the safe and then work from their. Worked great for me.
 
Thanks !! Clear and complete !
Guess tinfoil hats will need to go onto W10 and stop blocking windows update ?

Note: Not saying there's nothing else... simple guesstimates based on how many are still running W7 or blocks updates on this forum. High probability that this doesn't even change a thing. Move along lol.
You'll never take my win98! XP looks like Fischer price! (yes the same idiots whining today about Win10 whined about XP too).
 
To those stating these threads already exists they dont, at least not any more. I made this thread with the sole intent of it being new and fresh.

This is new news. Its news because the fix is actually being tested. Discuss that. Dont fight over who's definition is right. Use AMDs definition and discuss Toms findings.

It's that simple. The other threads were locked because people want to argue over semantics.

Now...
Lots of people not paying attention to their base clocks. Some boards run 99mhz others 101+
This will affect your final highest clock. Keep that in mind.

Tom's did not keep that in mind. That's a flaw worth noting I think.

The mind boggles.
 
Looks like AMD is going to come correct and give the max clock speeds they advertised but, as expected, overall performance will probably not change much.
I have a slight, sneaking suspicion AMD has mostly changed the granularity (allowed by that W10 patch), to report higher clocks to software in most scenarios, while only slightly increasing the actual average clock rate, which can't really be measured terribly accurately with software (yet?), due to the insane granularity now available, hence the displayed performance disparity.
 
The mind boggles.

I think youre thread got confused with another etc... it happens. Ownership of a thread carries with it no reward. Either way your information was shared one way or another.

Your timestamp was 409am and mine was 10something am.... but it was the timing I guess, either way we should be excited that AMD is owning up and fixing this issue.

I've had my threads closed before by admins. After all, these threads are information-ally the property of HardOCP. They do what they please with them, thats ok.
 
So the easiest way to do this is to use a program called Ryzen Dram Calculator. https://www.guru3d.com/files-details/download-ryzen-dram-calculator.html
Step 1 is to find out the specifics of the chips in your specific dram using a program like Thaiphoon burner. Once you have done this you plug your ram into the calculator and its spits out some timings. the calculator has "Safe" and "fast" settings for you to try. start with the safe and then work from their. Worked great for me.

Thanks for answering him, as I hadn't had a chance to respond.
 
I think youre thread got confused with another etc... it happens. Ownership of a thread carries with it no reward. Either way your information was shared one way or another.

Your timestamp was 409am and mine was 10something am.... but it was the timing I guess, either way we should be excited that AMD is owning up and fixing this issue.

I've had my threads closed before by admins. After all, these threads are information-ally the property of HardOCP. They do what they please with them, thats ok.

Thanks. I have no issue with the thread being closed...it was duplicated in the AMD processor section as was kindly pointed out by "hititnquitit" above. It's just that when the same link is reposted by another member and it stays open, the inconsistency is problematic.
 
Looks like AMD is going to come correct and give the max clock speeds they advertised but, as expected, overall performance will probably not change much.

That depends on how much boost a person is missing. I doubt someone will notice 25-50 Mhz. But, those that are missing 150-300 Mhz should see a slight bump.

With these beta bios my single core max boost used to be 4250 on one core, now it boosts to 4374 bouncing between two cores. And my all core boost is about 100 Mhz higher. My single and multi Cinebench scores went up. What does that mean in the real world? I don't know. But at least I'm closer to not having a deficit in performance.
 
Cool people will be happy. Of course as Tom points out actual performance is really unaffected.

As expected a 1-2% (at most) bump in a single core boost clock means a big whooping 0.5% bump in performance.

I'm not saying this isn't something AMD should fix up... if for nothing else to avoid stupid law suits. Its just such a non story, makes me almost hope AMD does just bring back PR numbers. It would be so much easier to just say this is our 12 core 5000+ and this is our 8 core 4600+ and avoid sillyness. Of course having this stuff end up in Intel marketing PP decks is pretty damn funny.
This says it all.
 
My 3900x went from 4.3 max single core to 4.55 max with ABB and now with ABBA I'm seeing 4.65 max single core. Benchmark scores are up between 1 and 3% versus ABB and over 5% from the first one.
Any before/after voltage numbers? Curious if it went up to get these gains.
 
from what i see on the non-fix bios, even without pbo turned on, the settings for what precision boost look at on my motherboard are maxed out for all but the max current setting. That's why turning precision boost overdrive on and off doesn't really matter. The current limit doesn't vary, so i'm always pegged at it when under full load.

If this bios update fixes that, then frequency can increase with potentially minimal voltage increase ...since there will be more current available. Current limits will directly impact peak frequency available.
 
I have a slight, sneaking suspicion AMD has mostly changed the granularity (allowed by that W10 patch), to report higher clocks to software in most scenarios, while only slightly increasing the actual average clock rate, which can't really be measured terribly accurately with software (yet?), due to the insane granularity now available, hence the displayed performance disparity.
I don't think AMD would cheat and just change the reporting, but I do think they will adjust the ratio between the peaks and valleys of the boost.

For example, boosting for less time but with a higher clock, and then "resting" for longer. This may help with quick single-threaded tasks but probably not with sustained loads, which I think is what AMD was saying in the Reddit post.
 
https://arstechnica.com/gadgets/2019/09/amd-improves-real-world-boost-and-idle-on-ryzen-3000/
"This really isn't the end of the world—a Ryzen 9 3900X that peaks at 4.5GHz instead of 4.6GHz is only missing out on 2% of its total possible boost clock rate, and even that 2% clock rate does not generally translate to 2% slower application performance. In other words, you're going to need artificial tests to discover the problem—you absolutely would not just suddenly realize, "hey, this isn't as fast as it ought to be!" in the middle of a gaming or content creation session."
 
https://arstechnica.com/gadgets/2019/09/amd-improves-real-world-boost-and-idle-on-ryzen-3000/
"This really isn't the end of the world—a Ryzen 9 3900X that peaks at 4.5GHz instead of 4.6GHz is only missing out on 2% of its total possible boost clock rate, and even that 2% clock rate does not generally translate to 2% slower application performance. In other words, you're going to need artificial tests to discover the problem—you absolutely would not just suddenly realize, "hey, this isn't as fast as it ought to be!" in the middle of a gaming or content creation session."

I don't disagree with them if we were living in a vacuum. However, we unfortunately live in a world where people idolize AMD or Intel and if Intel had something like this happen then AMD fanatics would be crying bloody murder. As it is the AMD fanatics (shills) say it's no big deal. I love the smell of hypocrisy in the morning.
 
I don't disagree with them if we were living in a vacuum. However, we unfortunately live in a world where people idolize AMD or Intel and if Intel had something like this happen then AMD fanatics would be crying bloody murder. As it is the AMD fanatics (shills) say it's no big deal. I love the smell of hypocrisy in the morning.
otoh, the intel fans (shills) would also say it's nothing. Maybe it actually IS nothing, regardless of whether it happens on an AMD or Intel processor? Feel free to quote me when it happens and I don't throw shit at the fan.

Edit: honestly, I don't think it is exactly nothing. Clearly, there was an issue, and AMD is working to resolve it. 100MHz is not huge, especially when one thread is only hitting it 1% of the time, but depending on how it is handled (in the os and microcode) it can make a big difference in certain workloads. Not in time to completion, necessarily, but in other less predictable ways. The problem is that there isn't really any way to show how this actually affects people in real workloads, and benchmarks + monitoring programs only show part of the problem... sometimes.
 
Last edited:
I don't disagree with them if we were living in a vacuum. However, we unfortunately live in a world where people idolize AMD or Intel and if Intel had something like this happen then AMD fanatics would be crying bloody murder. As it is the AMD fanatics (shills) say it's no big deal. I love the smell of hypocrisy in the morning.

It's ludicrous that you have the gall to type that in the face of all the BS going with Intel and their mountains of lying and deceit in regards to just how secure their chips are. AMD's boost issue isn't really an issue as it's adjustable with a simple bios update. Try again with your fishing...
 
Just ranges. Voltage has dropped from around 1.35 to 1.39v to 1.26 to 1.28v under load since the release bios. If anything, they are doing some clock stretching.

the other side of it could be that the voltage was never the limitation and instead they were using it to overcome some sort of protection limit(maybe something that was put in place for fear that they'd kill b350/x370 boards?) in the microcode which could be why PBO has been useless up to this point. i'd be interested to see though once the final release of the bios comes out whether or not PBO will actually do anything or at least the auto overclock option(not part of PBO).
 
Back
Top