APU Wars - Ongoing APU Review by AtenRa

[YBS1]

Ok, not really happy with the temperatures I'm getting with the H60. So I have a few questions, what's the best temperature monitor for AMD chips, RealTemp doesn't work, and CoreTemp seems wonky on it. I'm currently planning to pull the H100i off the 3930k to use with this and buying a Swiftech unit to place back on the 3930. I may decide to just go air though, in that case which is the best mini-itx compatible heatsink? Performance and absolute minimal noise are the priorities, not cost.

Only real issue I'm having other than that so far is the machine doesn't wake properly after going into sleep (Windows 7 Pro FWIW). It powers back up and the monitor initializes but the screen remains black, any known issues? Not tried to work this out myself yet, just checking to see if it might be something obvious like certain sleep states are know issues with the chipset.

 

[YBS1]

Disabling Core C6 fixed the wake from sleep issue with Win7, it must be specific to 7 though as now that I'm running Windows 10 on this machine Core C6 enabled works fine.

Going to have to do something about the cooling issues though to get the best out of it. Managed a 3.99 Cinebench R11.5 run but it should have been much higher. All power management settings were disabled and the cpu was set to 4.6GHz but it throttled between 4.6 and 3.4 the entire way through the run. Seemed to happen at 60C, cpu internal thermal failsafe maybe? I've not tried beyond 4.6 as it throttles anyway, based on voltage needed so far I think I could certainly get 4.7GHz out of it if I could get the cooling in hand provided the mini-itx board can cope. BTW, Open Hardware Monitor seems to work pretty well for these A10 cpus, and gives tons of info.

 

[DrMrLordX]

Temp monitoring: CoreTemp will monitor thermal margin backwards. Normally you want 80 thermal margin (good). CoreTemp will show 0 thermal margin as being good, and 80 as being bad. But it will still monitor it.

SpeedFan will monitor CPU temp okay. Some claim that HWMonitor will do it as well, but it never works for me (just shows "package temperature" which is either nonsense or the board VRMs).

The main thing you need to watch is thermal margin. Once it starts going close to 80 it will start to throttle. It may start lower than that.

As far as cooling goes, I don't know what to recommend, since I use an nh-d14 that would not be suitable for your case in all probability. I do not know if you are going to do much better than an H60 given the circumstances, unless you can manage a better AiO cooler in there.

4.4 GHz @ 1.31v is excellent scaling btw. I strongly suspect that your board may be holding you back at higher voltages/clockspeeds. Cooling the VRMs may or may not help.

Another option is to use amdmsrtweaker to try and defeat that throttling. What you want to do is set the p4 state to equal the p3 state. But be careful, you may be forcing more power through the socket than the board's designers ever intended. I am also uncertain of exactly what temp you're getting as 60c, so it may be thermal throttling . . .

 

[YBS1]

Temp monitoring: CoreTemp will monitor thermal margin backwards. Normally you want 80 thermal margin (good). CoreTemp will show 0 thermal margin as being good, and 80 as being bad. But it will still monitor it.

SpeedFan will monitor CPU temp okay. Some claim that HWMonitor will do it as well, but it never works for me (just shows "package temperature" which is either nonsense or the board VRMs).

The main thing you need to watch is thermal margin. Once it starts going close to 80 it will start to throttle. It may start lower than that.

As far as cooling goes, I don't know what to recommend, since I use an nh-d14 that would not be suitable for your case in all probability. I do not know if you are going to do much better than an H60 given the circumstances, unless you can manage a better AiO cooler in there.

4.4 GHz @ 1.31v is excellent scaling btw. I strongly suspect that your board may be holding you back at higher voltages/clockspeeds. Cooling the VRMs may or may not help.

Another option is to use amdmsrtweaker to try and defeat that throttling. What you want to do is set the p4 state to equal the p3 state. But be careful, you may be forcing more power through the socket than the board's designers ever intended. I am also uncertain of exactly what temp you're getting as 60c, so it may be thermal throttling . . .

I think, *think* - it may be reading package temperature, since 60C core would be beyond good for a lowly H60 (and it wouldn't throttle). I'm pretty certain I can get the H100i in there without issue but I may have to mount the fans inside as pull rather than push in order to give the tubing a more graceful bend. I've just been reluctant/procrastinating ordering a Swiftech AIO to replace the H100i on the 3930K. If I do that I would certainly replace the Corsair fans, as I'm looking for whisper quiet. The other option that looked promising was the Noctua NH-C14S, but availability is sketchy ATM.

 

[YBS1]

A little update in regards to what DrMrLordX posted. I checked with core temp, it displays 65 when the throttling kicks in and that is definitely what is happening, as soon as it hits 65 (or 60 on Open Hardware Monitor) down it goes to 3.4. I figured out that I can delay the throttling by reducing voltage, I almost made it through at 4.5 before it throttled, I may be able to lower the voltage further, don't know yet. This is good though as it means the chip is good for a least a solid 4.6 or better if I can cool it off.

 

[TeknoBug]

Yup AFAIK 62.5C is the max temp before it starts to throttle. My old A10 6800K I had for a short time gave me fits when going over 60C.

 

[Abwx]

New Kaveri/Godavari SKUs :

Die neuen Prozessoren heißen A10-7890K und A8-7690K und unterscheiden sich von den Vorgängern durch einen höheren CPU-Takt.

4.1 and 3.7GHz base frequency respectively, turbo and GPU frequencies are unknown at this time..

http://www.computerbase.de/2015-08/...veri-prozessoren-erhalten-nochmals-mehr-takt/

 

[TeknoBug]

lol first the *K and now the xx90 numbers, why can't AMD do their own numbering like they used to instead of copying Intel?

 

[Abwx]

lol first the *K and now the xx90 numbers, why can't AMD do their own numbering like they used to instead of copying Intel?

That s logical since they step up by 20s, that s two binnings above the 7850K, but i guess that it s too difficult to imagine..

While you re at it, what about the 6600 and 6700 numbers..?..

Where did you state that Intel is copying AMD..?.

 

[DrMrLordX]

A little update in regards to what DrMrLordX posted. I checked with core temp, it displays 65 when the throttling kicks in and that is definitely what is happening, as soon as it hits 65 (or 60 on Open Hardware Monitor) down it goes to 3.4. I figured out that I can delay the throttling by reducing voltage, I almost made it through at 4.5 before it throttled, I may be able to lower the voltage further, don't know yet. This is good though as it means the chip is good for a least a solid 4.6 or better if I can cool it off.

Okay, you are getting throttling from high thermal margin. I would not indict the board just yet. It just needs better cooling, which, as you have observed, is a challenge in a mITX case.

New Kaveri/Godavari SKUs :

The 7890k is unexpected. Some buyers of the 7870k may be annoyed by that announcement, or maybe they'll just shrug their shoulders when they realize that both the 7890k and 7870k probably top out at 4.7 GHz anyway.

 

[Abwx]

That s indeed the same chips, and according to Kitguru Godavari power plan was enhanced in respect of usual Kaveris, that s supposedly the reason of their higher frequency ceilling although usual chips did obviously benefit from a matured process.

AMD’s latest accelerated processing units that belong to the “Kaveri Refresh” family feature a reworked power supply circuitry, which gives cleaner and higher volume power delivery to the die. This lets AMD to increase “official” frequencies of its highly-integrated desktop chips and which also provides overclockers a significantly better headroom for further tweaking. In addition, AMD is able to bolster clock-rates of its chips thanks to the fact that GlobalFoundries’ 28nm super high performance (SHP) process technology has got significantly better than it was early in its lifecycle.

http://www.kitguru.net/components/c...ster-a8-and-a10-accelerated-processing-units/

 

[YBS1]

Okay, you are getting throttling from high thermal margin. I would not indict the board just yet. It just needs better cooling, which, as you have observed, is a challenge in a mITX case.

The board and the chip have handled it like champs. Only crashing once when I went too low on the voltage at 4.6GHz. It just gets too hot, throttles and rolls onward. I actually expected the mini-itx board to cry and give up before it ever got to 4.6, I can't imagine an ~$85 mini-itx board was ever designed with crazy overclocking ambitions.

The 7890k is unexpected. Some buyers of the 7870k may be annoyed by that announcement, or maybe they'll just shrug their shoulders when they realize that both the 7890k and 7870k probably top out at 4.7 GHz anyway.

The 7850 may get there too if I get off my butt and order some parts.

 

[DrMrLordX]

Don't be so sure about the 7850k. Some of them will, but the vast majority top out at around 4.5-4.6 GHz unless you do some "extreme" stuff with them (delid, etc). The only way my 7700k hit 4.7 was with a delid.

 

[YBS1]

Weeeee! It's all under control now. I received my Swiftech H240-X to swap onto my 3930K today (which I'm much more happy with as it is so much more quiet than the H100i and it's default fans), so I ended up spending the first half of the night getting that rig outfitted. After sorting that out I moved onto my 7850K. I dumped the default fans on the H100i for some Cryorig QF120 Balance fans, and they seem to be excellent. I did manage after all to be able to fit it with the fans as push and no water line crimping this time. Anyways, the H100i with the Cryorig fans is kickin' tail. I finished getting it back together too late to try to max out the clocks or dial in a minimum voltage but the throttling is gone. Did a bit of thermal stress testing and got in a 4.6GHz Cinebench, will try to push it a little when I get some more time over the next few days.

 

[DrMrLordX]

Nice. Try running Prime95 28.5 custom 768k/768k mix/max FFT size, and run FFTs in place, to really stress that puppy.

 

[YBS1]

I got into Windows at 4.7 but I was unable to complete Cinebench @ 1.45v. Temperatures weren't a problem though. What is considered safe voltage with appropriate temperatures on these? It's possible I suppose this is about all a little mini-itx board is able to cope with too. Also, northbridge voltage?? 1.35 is default, I can run 2000 with that, a bump to 1.4 allowed me to run 2100, stopped there for now. Apu/pci-e frequency, does this give a notable increase in performance rather than simply going the easy route of changing the multiplier? If so, at about what point does this chipset tend to make it to without issues? Might as well pick the brains of people who are more familiar with it than me...

 

[DrMrLordX]

I think the temp limit is somewhere in the ballpark of 62-70C, though it's not a good idea to go that high. You will usually run into thermal margin problems before you get there.

As for voltages, 1.325v vNB is usually as high as you want to go unless you are really pushing it. Some people get away with 1.4v but I don't know how long that will last before your memory controller suffers stress/damage. vcore is officially listed as achieving "breakdown voltage" above 1.49v. Plenty of people have pushed 1.5-1.55v vcore without too many obvious problems. YMMV of course.

 

[YBS1]

Well I made it to 4.7GHz, I had been trying 1.35v with a +10% offset before. A switch to no offset 1.49v allowed it to run 4.7, drooping to as low as ~1.41ish under load. I was able to even get in one Cinebench run at 4.8 at the same voltage, but it crashed upon an accidental second run before getting a screenshot (was still slower than DrMrLordX's 4.7GHz run though, memory perhaps?). Seems to be cool and comfortable with 4.4@~1.30v, so that's probably where it will spend it's days.

 

[DrMrLordX]

Might be memory, might be Northbridge. Your voltage is lower than mine at that speed, though. Good on you for that.

And you didn't have to peel off the lid, either.

 

[Shivansps]

Im starting to use A8-7650K instead of A8-7600... did someone realised that the cooler on the A8-7650K is not only double in height than the cooler for the 7600 but is also about 25% higher than the cooler of a A10-7850K? wtf.

 

[burninatortech4]

Im starting to use A8-7650K instead of A8-7600... did someone realised that the cooler on the A8-7650K is not only double in height than the cooler for the 7600 but is also about 25% higher than the cooler of a A10-7850K? wtf.

Godavari (Kaveri revision) chips are boxed with larger coolers than the original Kaveri generation.

 

[DrMrLordX]

The 7650k is not a GV-A1 chip. Or, at least, it didn't start out that way. AMD may have jumped the gun on moving to new coolers with the 7650k when they released it.