AMD Announces BIOS Fix for Ryzen 3000 Boost Clocks, Update Comes September 10

I am going to take a guess that this Sept 10 date is not the actual you can go to mobo site and grab it date. So ~a week after that?
 
ir0nw0lf said:
I am going to take a guess that this Sept 10 date is not the actual you can go to mobo site and grab it date. So ~a week after that?
Click to expand...


3 weeks after AMD releases. Gigabyte will pump it out faster than the rest. Asus will be last as usual.
 
"We closely monitor community feedback on our products and understand that some 3rd gen AMD Ryzen users are reporting boost clock speeds below the expected processor boost frequency."

It's not "some" users. It's the majority of the users reporting below expected boost frequency.
 
Already have fix, a modified CrossHair 6 Hero Beta Bios using an older SMU module from AGESA 1.0.0.2 with the 1.0.0.3ABB. It will use a multiplier of 46.25 on up to three cores or 4.625 ghz except most folks have something less than 100mhz Bus Speed. The difference is that it uses voltages up to 1.5v while the 1.0.0.3ABB SMU goes up to 1.48v. Will be interesting to see AMD fix when they already had it previously pushing the CPU to rated speeds.

CPUz showing the boosted clock:


2019-09-03 (3).png

​
The differences for the user looks to be immaterial and not even worth any extra thought or belief one is really missing out. Here is single core Cinebench 20, the higher number one of 523 result is from the modified bios using an older SMU. The 505 result is from ASUS Official 7403 bios:

CB20Single.jpg
​
Somewhat insignificant difference there. Here is the multiple thread test and really no real difference, top orange is with the older SMU, below that is from the Official bios SMU:

CB20Mult.jpg

​
The last result is a comparison between the boosted 4.625ghz SMU and the current 4.525ghz SMU using 3DMark - once again very much insignificant unless a few points or tenths of a percent is important here:

3dMarkComp2.jpg

3dMarkComp.jpg
​
I just don't see any significant benefit overall, for anyone with much less cooling it may actually be detrimental, due to the increase voltages used on the higher boosting SMU unless AMD keeps the voltage lower. Why yes you can run a benchmark and show a somewhat improvement for a single thread but in reality how many programs are really single threaded? And the ones that are, well are the very old programs that were effective with very old processors to begin with. Today games are multi threaded, today programs are multi threaded - in the end not much will change for the end user is my prediction.
 
Last edited:
noko said:
Already have fix, a modified CrossHair 6 Hero Beta Bios using an older SMU module from AGESA 1.0.0.2 with the 1.0.0.3ABB. It will use a multiplier of 46.25 on up to three cores or 4.625 ghz except most folks have something less than 100mhz Bus Speed. The difference is that it uses voltages up to 1.5v while the 1.0.0.3ABB SMU goes up to 1.48v. Will be interesting to see AMD fix when they already had it previously pushing the CPU to rated speeds.

CPUz showing the boosted clock:


The differences for the user looks to be immaterial and not even worth any extra thought or belief one is really missing out. Here is single core Cinebench 20, the higher number one of 523 result is from the modified bios using an older SMU. The 505 result is from ASUS Official 7403 bios:

Somewhat insignificant difference there. Here is the multiple thread test and really no real difference, top orange is with the older SMU, below that is from the Official bios SMU:

The last result is a comparison between the boosted 4.625ghz SMU and the current 4.525ghz SMU using 3DMark - once again very much insignificant unless a few points or tenths of a percent is important here:

I just don't see any significant benefit overall, for anyone with much less cooling it may actually be detrimental, due to the increase voltages used on the higher boosting SMU unless AMD keeps the voltage lower. Why yes you can run a benchmark and show a somewhat improvement for a single thread but in reality how many programs are really single threaded? And the ones that are, well are the very old programs that were effective with very old processors to begin with. Today games are multi threaded, today programs are multi threaded - in the end not much will change for the end user is my prediction.
Click to expand...

Where did you get that bios at?
 
Gideon said:
Where did you get that bios at?
Click to expand...
Gupsterg with help from The Stilt modified the bios for the C7H as well as for the C6H which can be read at ASUS.com Rog forums: Basically taking an older SMU which boosted clocks to spec and changing it out with bios with AGESA 1.0.0.3ABB. So all the benefits of the later AGESA such as memory compatibility and speed support to other fixes are the same as the Official bios but clocks like the older bios.

https://rog.asus.com/forum/showthre...-Hero-Essential-Info-Thread/page12#post782315
 
You show minor Cinebench score changes. I can reboot and re run Cinebench and my multicore score will vary by as much as 50 points. It will be very hard for me to tell any real difference.
 
After hours of memory testing, watching videos, editing images running memory test, looking at HWiNFFO64 Core Clock, the Maximum frequency obtain; 4 cores were clocked at some point at 4.616ghz. I had 3 cores that did not go above 4.2xxghz until I ran AIDA64 Cache & Memory Benchmark then those stubborn cores went up 4.3ghz+. Note the max core voltage of 1.5v - many were complaining about the high voltage which may have prompted AMD to reduce that and thus Maximum Boost as well with the newer AGESA - I don't know. AIDA Benchmark shows the BIOS Version 0002+, Gupsterg added the + to designate the bios was modified (in this case with an older SMU package). The 400% MemTest, it is good standard, had zero errors, I normally like to go above 1000% but I will be tweaking the memory more over time.

2019-09-04 (1).jpg​
 
kamikazi said:
You show minor Cinebench score changes. I can reboot and re run Cinebench and my multicore score will vary by as much as 50 points. It will be very hard for me to tell any real difference.
Click to expand...
I don't see as much variations as that but yes +- 25 or 50 is basically within benchmark error showing no real difference when using multiple threads hence in real world use, having multiple threads doing stuff, zero difference.
 
noko said:
I don't see as much variations as that but yes +- 25 or 50 is basically within benchmark error showing no real difference when using multiple threads hence in real world use, having multiple threads doing stuff, zero difference.
Click to expand...

Thanks. I also notice that the VID for the cpu cores in HWINFO shows as high as 1.5v for me on all cores, but the actual vote never goes that high.
 
hitting 4600 is only part of what the fix implies it will do. the 3800x and 3900x both appear to not have working PBO. I would expect that the bios fix is going to also impact that. So we should see these chips with adequate cooling able to be pushed up to 4800mhz on the good cores and perhaps up to 4500mhz all core.

Though, they might have a legal obligation to fix the issue with normal pb not hitting 4600mhz ...even if the difference between 4500 and 4600 can't be seen in any meaningful way performance wise, since it's written on the box as being capable of it.
 
  • Like
Reactions: noko
like this
noko said:
Already have fix, a modified CrossHair 6 Hero Beta Bios using an older SMU module from AGESA 1.0.0.2 with the 1.0.0.3ABB. It will use a multiplier of 46.25 on up to three cores or 4.625 ghz except most folks have something less than 100mhz Bus Speed. The difference is that it uses voltages up to 1.5v while the 1.0.0.3ABB SMU goes up to 1.48v. Will be interesting to see AMD fix when they already had it previously pushing the CPU to rated speeds.

CPUz showing the boosted clock:


The differences for the user looks to be immaterial and not even worth any extra thought or belief one is really missing out. Here is single core Cinebench 20, the higher number one of 523 result is from the modified bios using an older SMU. The 505 result is from ASUS Official 7403 bios:

Somewhat insignificant difference there. Here is the multiple thread test and really no real difference, top orange is with the older SMU, below that is from the Official bios SMU:

The last result is a comparison between the boosted 4.625ghz SMU and the current 4.525ghz SMU using 3DMark - once again very much insignificant unless a few points or tenths of a percent is important here:

I just don't see any significant benefit overall, for anyone with much less cooling it may actually be detrimental, due to the increase voltages used on the higher boosting SMU unless AMD keeps the voltage lower. Why yes you can run a benchmark and show a somewhat improvement for a single thread but in reality how many programs are really single threaded? And the ones that are, well are the very old programs that were effective with very old processors to begin with. Today games are multi threaded, today programs are multi threaded - in the end not much will change for the end user is my prediction.
Click to expand...
I wouldn't expect much difference here in cb unless you artificially limit the program to two or three threads anyway. cb uses each thread to the best of its ability, so there's not much room to boost any one thread. Same is true for most mt benchmarks, unfortunately.
 
  • Like
Reactions: noko
like this
noko said:
After hours of memory testing, watching videos, editing images running memory test, looking at HWiNFFO64 Core Clock, the Maximum frequency obtain; 4 cores were clocked at some point at 4.616ghz. I had 3 cores that did not go above 4.2xxghz until I ran AIDA64 Cache & Memory Benchmark then those stubborn cores went up 4.3ghz+. Note the max core voltage of 1.5v - many were complaining about the high voltage which may have prompted AMD to reduce that and thus Maximum Boost as well with the newer AGESA - I don't know. AIDA Benchmark shows the BIOS Version 0002+, Gupsterg added the + to designate the bios was modified (in this case with an older SMU package). The 400% MemTest, it is good standard, had zero errors, I normally like to go above 1000% but I will be tweaking the memory more over time.

Click to expand...

I still see C8H hero hit 1.5v on some cores max for short bursts. But mostly most cores around 1.48 max. So I am not buying that they lowered voltages. I doubt .02 is going to make world of a difference under single core loads.
 
  • Like
Reactions: noko
like this
Darth Ender said:
hitting 4600 is only part of what the fix implies it will do. the 3800x and 3900x both appear to not have working PBO. I would expect that the bios fix is going to also impact that. So we should see these chips with adequate cooling able to be pushed up to 4800mhz on the good cores and perhaps up to 4500mhz all core.

Though, they might have a legal obligation to fix the issue with normal pb not hitting 4600mhz ...even if the difference between 4500 and 4600 can't be seen in any meaningful way performance wise, since it's written on the box as being capable of it.
Click to expand...
Well I hope that is the case with more meaningful fixes as in higher boost for more threads. Still for the 3900x, the base clock is 3800mhz and anything above that is boost.

So others are seeing 1.5v routinely, I do not recall seeing that at all, 1.47v-1.48v was the top end. Seeing 1.5v caught my eye, under full threaded loads it is 1.35v-1.37v same as before.

In the end, hitting a max boost of 4.6ghz, at fractions of a second, only when a single thread is loaded, opportunistic condition, is really nothing much. If there was a sustain 4.6ghz on one core (which would be two threads), while other cores are lower, than that I think would make a noticeable difference. I like Intel's boost behavior much better for the Max Frequency. Regardless what really is important is overall performance and that is outstanding from the Ryzen 3's for the $.
 
noko said:
Well I hope that is the case with more meaningful fixes as in higher boost for more threads. Still for the 3900x, the base clock is 3800mhz and anything above that is boost.

So others are seeing 1.5v routinely, I do not recall seeing that at all, 1.47v-1.48v was the top end. Seeing 1.5v caught my eye, under full threaded loads it is 1.35v-1.37v same as before.

In the end, hitting a max boost of 4.6ghz, at fractions of a second, only when a single thread is loaded, opportunistic condition, is really nothing much. If there was a sustain 4.6ghz on one core (which would be two threads), while other cores are lower, than that I think would make a noticeable difference. I like Intel's boost behavior much better for the Max Frequency. Regardless what really is important is overall performance and that is outstanding from the Ryzen 3's for the $.
Click to expand...


I think it has to do with temps vs cores. As an example I have 3600 and base clock is 3.6ghz but it boosts to 4075, almost 4.1ghz all core when I running 3dmark or games and 4ghz when its stress tested. So pretty much 4-500mhz above base clock. It might be that AMD is reducing boost according to temps and they need to loosen that a bit.

I have another system with 3900x and it boosts to 4225mhz-4400 ish all core boost in games. I have seen it go up to 4400mhz but it fluctuates quiet a bit but doesn't go below 4200. Honestly wouldn't mind sustaining 4400 ish in games since my temps are in the 50s in games. It seems they are too aggressive with temps to clock ratio.
 
  • Like
Reactions: noko
like this
If you look at the sensors, it's tdp that they limit to, not temps, in the case where the temps aren't hitting the 90+C limit.
 
Darth Ender said:
If you look at the sensors, it's tdp that they limit to, not temps, in the case where the temps aren't hitting the 90+C limit.
Click to expand...

I don't think its just TDP. A lot goes in to it. Because I am not even hitting that much in games. I think it should be able to sustain 4400 MHz on 6 cores. It may be that current bios is being too stingy with temps lol.
 
I've seen no evidence of the tdp not hitting the rated amount and the frequency being kept below the all core 4.3Ghz and the single core 4.5xGhz frequencies. I always hit the tdp and the frequency falls where it needs to , to maintain that. Lots of cache hit processing will further reduce frequncies but tdp will remain because the cache power usage factors in (and there's lots of it) ...as well as traffic over the infinity fabric - so users who use faster ram, are restricted more than users who have somewhat slower ram for the same workloads in terms of peak frequency in order to maintain the same tdp.
 
Darth Ender said:
I've seen no evidence of the tdp not hitting the rated amount and the frequency being kept below the all core 4.3Ghz and the single core 4.5xGhz frequencies. I always hit the tdp and the frequency falls where it needs to , to maintain that. Lots of cache hit processing will further reduce frequncies but tdp will remain because the cache power usage factors in (and there's lots of it) ...as well as traffic over the infinity fabric - so users who use faster ram, are restricted more than users who have somewhat slower ram for the same workloads in terms of peak frequency in order to maintain the same tdp.
Click to expand...

well cpuz is an example if you select lets say 6 threads. I am averaging 80w tdp but my boost is around 4250 so it does have room for higher boost but it doesn't.
 
tdp is an avg. it's not like you could be expected to hit 105 watts on a single core if you ran just 1 thread at full load. While the max tdp for a single core is not going to be 105/12, it's going to be some amount greater than that but not so much that it would cause localized current/voltage/temperature limits to be exceeded. so 8.75watts per core compared to 80/6 == 13.3 . That would be roughly 160 watts if you had all 12 cores running at the same. That looks to be plenty in-line with what you would expect with precision boost exceeding the TDP of the cpu.
 
Darth Ender said:
tdp is an avg. it's not like you could be expected to hit 105 watts on a single core if you ran just 1 thread at full load. While the max tdp for a single core is not going to be 105/12, it's going to be some amount greater than that but not so much that it would cause localized current/voltage/temperature limits to be exceeded. so 8.75watts per core compared to 80/6 == 13.3 . That would be roughly 160 watts if you had all 12 cores running at the same. That looks to be plenty in-line with what you would expect with precision boost exceeding the TDP of the cpu.
Click to expand...

well if its that then that sucks lol. I didnt know they would average it. So beats the purpose of it. I am happy with my all core OC to 4350 at 1.31v on my 3900x. It actually consumes same amount of power my chip does on auto under load. 140 ish watts.
 
You also have to consider this, at 160 watts or all core 4.25Ghz as in your example, at 1.5v cpu voltage which is probably the max the cpu would auto-pb and pbo to, you're pulling almost 107 amps. If your voltage is lower than 1.5 volts at that speed/tdp then your current is even higher given the same frequency.

Technically, we would expect to be able to push as much current as possible until heat stops us. But I also think as die shrinks, max limits for currents decrease, just as as components are squished closer together, max voltages decrease.


Obviously though, something is incorrect as amd is making changes. Some part of the algo is being a bit too conservative or ignoring the fact that we want to exceed some specific limit. But i think we're also being impacted by a change in how these parts are being marketed. Where when the number says "this can go up to" etc...they mean ...that's pretty much the best you can expect if the stars align. No more getting lucky with a part that outperforms the marketing.

However, the bitching about not hitting 4.6Ghz when you hit 4.55 is kind of pointless. AMD could put out a microcode update that just fakes the sensor data to say 4.6 and since performance would be the same either way, it wouldn't matter.

The bigger problem that I'd like to see fixed is being able to use PBO ...since PBO doesn't seem to work at all on these parts. You never exceed the PB speeds/perf on the 3900x ...and I haven't tested it as much on my 3800x but it looks to be the same.
 
Darth Ender said:
You also have to consider this, at 160 watts or all core 4.25Ghz as in your example, at 1.5v cpu voltage which is probably the max the cpu would auto-pb and pbo to, you're pulling almost 107 amps. If your voltage is lower than 1.5 volts at that speed/tdp then your current is even higher given the same frequency.

Technically, we would expect to be able to push as much current as possible until heat stops us. But I also think as die shrinks, max limits for currents decrease, just as as components are squished closer together, max voltages decrease.


Obviously though, something is incorrect as amd is making changes. Some part of the algo is being a bit too conservative or ignoring the fact that we want to exceed some specific limit. But i think we're also being impacted by a change in how these parts are being marketed. Where when the number says "this can go up to" etc...they mean ...that's pretty much the best you can expect if the stars align. No more getting lucky with a part that outperforms the marketing.

However, the bitching about not hitting 4.6Ghz when you hit 4.55 is kind of pointless. AMD could put out a microcode update that just fakes the sensor data to say 4.6 and since performance would be the same either way, it wouldn't matter.

The bigger problem that I'd like to see fixed is being able to use PBO ...since PBO doesn't seem to work at all on these parts. You never exceed the PB speeds/perf on the 3900x ...and I haven't tested it as much on my 3800x but it looks to be the same.
Click to expand...

No way its pulling 4225 at 1.5v. That is insane and chip killer. Its pulling 4200 little over 1.35 volts with PBO, without PBO its around 4100 around 1.3v. All core manual oc is 4350 with 1.31v under load with memory at 3733mhz 16-16-16-32 with subtiming tweaks.
 
so at 1.3v, you're pulling 120amps. I'd expect we should see the chip able to handle a bit more than that, but I dont know how much more and if it would be significant.

Assuming we could get the cpu to hit 200 watts sustained, that's almost 150 amps at 1.35v. Through tiny little traces almost half the size we've previously put that kind of current through.

Perhaps at normal ambient temps, you can't trust that the temp sensor can respond to temperature changes fast enough to avoid turning a core trace into a fuse above certain currents...so it limits you even if temps look good.
 
Last edited:
Darth Ender said:
so at 1.3v, you're pulling 120amps. I'd expect we should see the chip able to handle a bit more than that, but I dont know how much more and if it would be significant.

Assuming we could get the cpu to hit 200 watts sustained, that's almost 150 amps at 1.35v. Through tiny little traces almost half the size we've previously put that kind of current through.
Click to expand...

Yep thats why I was saying it should be able to sustain around 4.4ghz in games given I have 4350 all core with bit more voltage.
 
NKD said:
Yep thats why I was saying it should be able to sustain around 4.4ghz in games given I have 4350 all core with bit more voltage.
Click to expand...


I think you would, if you put your ram to 3200 ..you'd probably see core frequency rise under these same workloads. Though I've seen variations in max core frequency change by over 300Mhz just by varying the type of load. Be it normal all core, stress avx all core, or just memory bound all core. and all core temps will vary significantly with each of those, as well as the wattage it's operating at.
 
Darth Ender said:
I think you would, if you put your ram to 3200 ..you'd probably see core frequency rise under these same workloads. Though I've seen variations in max core frequency change by over 300Mhz just by varying the type of load. Be it normal all core, stress avx all core, or just memory bound all core. and all core temps will vary significantly with each of those, as well as the wattage it's operating at.
Click to expand...

I can definitely try setting it to 3200mhz only for testing purposes though.
 
Just to let you guys know the older BIOS/Beta BIOS on motherboards could hit CPU 1.55v on Idle and light loads such as gaming/Single Cinebench and My motherboard did hit 1.57 but I just assume HWINFO64 was off a little.With 1.55v my 3600X can do 4525Mhz.
I shouls also say RAM speed had no effect on the boost for me.

With new BIOS I can not hit higher than 1.5v and 4250Mhz but performance is same or better than 4525Mhz.

Stilt Quote from thread but fun to play around with it for now.
There is not much longevity in SMU swapping, especially at this stage of the life of the platform.

As soon as the new SMU FW starts implementing a new feature(s), which is implemented in AGESA the old FW versions preceeding the change will be useless.
They'll probably still be working, but there can be various issues.

For example the new non-public FW (46.44.0) I tested yesterday has both CPPC and C0 residency improvements, which are obviously nice to have.
I cannot see the 50MHz higher peak and 41MHz average frequency being worth of not having the other improvements and bug fixes.
Click to expand...

Video showing voltage 1.57v RAM CL 14 3600Mhz running game If interested on beta BIOS.Maybe AMD going back to this method ,J/K noway they would do that.
 
Last edited:
noko said:
After hours of memory testing, watching videos, editing images running memory test, looking at HWiNFFO64 Core Clock, the Maximum frequency obtain; 4 cores were clocked at some point at 4.616ghz. I had 3 cores that did not go above 4.2xxghz until I ran AIDA64 Cache & Memory Benchmark then those stubborn cores went up 4.3ghz+. Note the max core voltage of 1.5v - many were complaining about the high voltage which may have prompted AMD to reduce that and thus Maximum Boost as well with the newer AGESA - I don't know. AIDA Benchmark shows the BIOS Version 0002+, Gupsterg added the + to designate the bios was modified (in this case with an older SMU package). The 400% MemTest, it is good standard, had zero errors, I normally like to go above 1000% but I will be tweaking the memory more over time.

Click to expand...

The complaint of high voltage was in regards to high voltage under light/no load conditions.
 
RamonGTP said:
The complaint of high voltage was in regards to high voltage under light/no load conditions.
Click to expand...
There was various complaints and an initial attempt by many to manually control the voltage to less than 1.35v. Confusion on PB2 behavior. That has been cleared up. First time I saw anything above 1.48v was with a modified bios with older SMU software.

Anyways those locking in voltage then saw zero boost which I take confuse many at the time. It would be better if AMD clearly show by example boost with PB2 and PBO. Showing that potential 200mhz boost over PB2 as in 4.8ghz on a 3900x.

I wish Kyle did an article over AMD Marketing BS. Definitely miss that in general. Most sites rather poor attempts at real journalism is tiresome.
 
noko said:
There was various complaints and an initial attempt by many to manually control the voltage to less than 1.35v. Confusion on PB2 behavior. That has been cleared up. First time I saw anything above 1.48v was with a modified bios with older SMU software.

Anyways those locking in voltage then saw zero boost which I take confuse many at the time. It would be better if AMD clearly show by example boost with PB2 and PBO. Showing that potential 200mhz boost over PB2 as in 4.8ghz on a 3900x.

I wish Kyle did an article over AMD Marketing BS. Definitely miss that in general. Most sites rather poor attempts at real journalism is tiresome.
Click to expand...

honestly he gave a general example. It wasn’t about any particular processor just clock speed in general. PBO does help in all core boost. I can see that on my board. Not a whole lot in single core but multi core it definitely does for me.
 
NKD said:
honestly he gave a general example. It wasn’t about any particular processor just clock speed in general. PBO does help in all core boost. I can see that on my board. Not a whole lot in single core but multi core it definitely does for me.
Click to expand...
I see about a 30 point increase in CB20 multithreaded, at least a few runs, that does not indicate PBO is increasing multithreaded performance in my case that is.
 
You must log in or register to reply here.
Back
Top