ASRock FM2A88X Extreme6+ / A10-7700K observations

[Daveburt714]

Hi Just,

I had all intentions of buying a Kaveri combo Sunday, but some problems arose...
I was shoveling snow on Saturday and got double vision and a loss of balance/dizziness. It went away after a couple hours, but started again on Sunday. It was so bad that I went to the hospital and have been there until today!
Turns out I had a number of mini-strokes... Now I have to take 5 different medications a day, but I'm feeling better, so hopefully everything will be alright. Enough of my problems though...

Thanks for all your testing man! I'll probably still pick-up a combo, but from looking at your numbers (other than iGPU) I don't really see where it will offer a huge improvement over the 6600K for an HTPC unless software gives it a serious boost down the road.

What are your thoughts? Improvement for my use, or just something new to play with?

Thanks,
Dave

[l0ud_sil3nc3]

^With AMD it's always something new to play with

Going to have to stop by Microcenter and grab one as well

[justapost]

Hi Just,

I had all intentions of buying a Kaveri combo Sunday, but some problems arose...
I was shoveling snow on Saturday and got double vision and a loss of balance/dizziness. It went away after a couple hours, but started again on Sunday. It was so bad that I went to the hospital and have been there until today!
Turns out I had a number of mini-strokes... Now I have to take 5 different medications a day, but I'm feeling better, so hopefully everything will be alright. Enough of my problems though...

Was wondering where you had left. Mow that sounds really serious, take your time and get well!

Hi Just,
Thanks for all your testing man! I'll probably still pick-up a combo, but from looking at your numbers (other than iGPU) I don't really see where it will offer a huge improvement over the 6600K for an HTPC unless software gives it a serious boost down the road.

What are your thoughts? Improvement for my use, or just something new to play with?

Thanks,
Dave

Yeah, the results look abit mixed on the cpu side. Few areas where the kaveri performs abit slower but also some with huge gains. I replaced my old athlon II 430 which has still fast and snappy for my daily work but it was five years old. In opposite to FM1/2 FM2+ seemed to be more future proove, as long as carizo will be compatible.
Last years i mainly used Athlon II's afterwards 4100's and 4300's for office builds, also tried i3 once but the gains are neglectible due to slow networks and such. Had to build two cad workstations and used 2500k and then an socket 2011 platform with an sb xeon. Tried to oc both of them abit, was fun too but i'm just used to amd.

[justapost]

^With AMD it's always something new to play with

^^this

[justapost]

Yesterday i struggled with the memory at 2400, no matter how much i relaxed the timings and upped the voltage it aloway got stuck during windows startup. That also happened with 2133 and BCLK's at around 102MHz. Now i found that the APU PCIe VDDA voltage has to be increased alot. It's stock settings are at 1.05V things started go get abit more stable at 1.2V and now at 1.3V it's finaly running prime95 for more that 10 minutes at 2210MHz 10-12-12-32-42. Had optimized timings and voltages at 2133MHz yesterday and it ran prime95 for 3 hrs at 9-11-10-28-38 at 1.6V. Really strange that it needs an 0.25V voltage increase on the APU PCIe VDDA to run stable at 80MHz faster speed.

[justapost]

APU VDDA must be in an good balance with the memory frequency it seems. At 2280MHz 10-12-12-32-42 it passed an hour prime95 with 1.258V. Tested 1180 afterwards, but did not get that stable yet. Odd thing using more realxed timings like 11-13-13-36-46 (clsoe to the 1300MHz settings in XMP) the board does not even post. Also while beeing rated for 1.65V the memory behaves much nicer at 1.6V.

[Lightman]

Hope you get an better replacement chip soon, that was bad luck at the wrong time. Did your board also choose an richland type of nb voltage in manual mode? Mine picked 1.275V but when i ran it at auto settings 1.1375V was used.

TBH I'm still testing this CPU as now it seems to be OK with PCIe 3.0 stable at least with BIOS default loaded, but I'm still getting HW errors in cgMiner for Kaveri GPU. One thing I've changed was I installed proper memory, initially I only tested this righ with some old cheap Kingstons DDR3 1333MHz 1GB 1.7V modules x4 but finally putted my proper Kingstons HyperX Predator 2133MHz DC set and it just works. I haven't tried OC yet but ran 3D games and benchmarks for few hours and now 24h of mining on 2xR9 290 with not a single crash or error. The only worry I have is these HW errors on Kaveri while mining, but this might be purely because of AMD new OpenCL driver included in beta driver supporting new APU's.

If someone can test for me cgminer on Kaveri I would be grateful. I can provide the whole command line to one of my workers so it would be as easy as click and run

Edit:

Good question about NB voltages! I'm going to check it now ...

I'm back! Default NB voltage is set to 1.10V on Auto.
I now had no problem increasing BCLK to 102MHz with A10-7850k in the socket. Works great, will test more, including dGPU clocks over weekend as I'm quite busy right now!

Thanks for your feedback and great testing!

@Daveburt714 I wish you quick and full recovery!

[FlanK3r]

7700+7580K on the way guys. Im looking forward too. And for some LN2 fun

[justapost]

If someone can test for me cgminer on Kaveri I would be grateful. I can provide the whole command line to one of my workers so it would be as easy as click and run

Edit:
I'm back! Default NB voltage is set to 1.10V on Auto.
I now had no problem increasing BCLK to 102MHz with A10-7850k in the socket. Works great, will test more, including dGPU clocks over weekend as I'm quite busy right now!

Odd thing with the memory and the BCLK, my Adata 2600 kit also causes such issues at 2400 and only worked fine at 1866 after adjusting timings manually. Crucial ballistix tactical had no problems running this speed with stock XMP settings. They have much lower tRFC but higher tRAS and tRC settings.
Thank you for checking NB voltage, so ~1.1V is the default. 2GHz worked stable here with 1.2V.
I missed the miner hype but can handle an one liner.

[justapost]

7700+7580K on the way guys. Im looking forward too. And for some LN2 fun

I'm for sure not the only one waiting for you to try ln2 on these.

[Daveburt714]

After going to the doctor and getting even MORE medication , I stopped buy MC and picked up a 7850k/ASR EX6+ combo on the way home! ($250 tax and all (not bad!))

I went with the 7850K because it was only $20 more than the 7700. Plus if Mantle/Huma /gpGPU and all the other stuff that shares processing loads with the iGPU work out the extra gpu cores may offer a nice boost on compute performance...

Here's a couple pics:






This board does have a lot of nice features!
I especially like the fact that it has an Atheros NIC, I've never been a big fan of Realtek's LAN chips.
I'll post some results from the bench (water cooled) when I get a chance...

[justapost]

@Daveburt712: Congrats to your decision. Had to rewire my water colling setup here completely somehow it got choked and the 430 at stock still worked. That liquid smelled undescribable brackish after beeing in an closed loop for 5 years.

Finaly found stable voltages for 1150MHz mem. It had failed at 1.6V mem and 1.25 or 1.285V APU VDDA so i upped mem as well and with 1.62V mem and 1.3V apu vdda it passed prime95 for 14 hrs. Going to lower vdda now till i it's getting unstable again. Seems these two volates have an sweetspot for each frequency and need to be balanced well. Tried 1.05V and 1.1V for apu vdda and p95 failed within seconds, now at 1.16V after an minute.



Background info to VDDA voltage: http://www.hardwaresecrets.com/print...otherboard/995

VDDA: This is the voltage used by the clock multiplier circuit inside the CPU, also known as PLL (Phase-Locked Loop).

[FlanK3r]

VDDA is something similar as PLL at Intel boards

[Lightman]

Odd thing with the memory and the BCLK, my Adata 2600 kit also causes such issues at 2400 and only worked fine at 1866 after adjusting timings manually. Crucial ballistix tactical had no problems running this speed with stock XMP settings. They have much lower tRFC but higher tRAS and tRC settings.
Thank you for checking NB voltage, so ~1.1V is the default. 2GHz worked stable here with 1.2V.
I missed the miner hype but can handle an one liner.

Interesting findings, I'm going to test it more on my config with memory set to 1833MHz and 1600MHz tomorrow or on Sunday.

If you have time to test cgMiner for me then below is all you need to run it. Just start Kaveri.bat and let me know if you will get any HW errors.
https://www.dropbox.com/s/evgtfegaca....2-windows.zip

In bat file you can play around with Thread concurrency and values up to 12200 should work and you can try to remove --no-restart from bat but on my motherboard it causes blank cgminer window (on Gigabye and 5800k it wasn't needed and it's to do with GPU FAN monitoring).

[Daveburt714]

@Daveburt712: Congrats to your decision. Had to rewire my water colling setup here completely somehow it got choked and the 430 at stock still worked. That liquid smelled undescribable brackish after beeing in an closed loop for 5 years.

I changed my pump after about 3yrs of use and the water was quite fresh. If it helps, I just use distilled water and a few drops of PTNuke in my system.

What are the folks with the EX6+ using for their LLC settings? I also have an AM3+ EX9 and LLC was a little troublesome on that board so I thought I'd ask...
I don't have any high freq. Mem (2000-2200 @ best), so I'm still playing with CPU clocks.

[justapost]

Oh my sensitive memory. Will tune that at an other day. Did a few benchmarks at 3640MHz CPU 1.525V+0.05V PWM Offset LLC 80% NB 1938MHz/1.275V MEM 1149MHz 10-12-12-32-42 1.65V VDDA 1.31V.
CB15 383

CB11.5 4.37

PiFast 28.86

SuperPi 1M 15.6

wPrime 1.55 32m 10.658 1024m 324.245

Finaly shabby image of the system

[justapost]

I changed my pump after about 3yrs of use and the water was quite fresh. If it helps, I just use distilled water and a few drops of PTNuke in my system.

What are the folks with the EX6+ using for their LLC settings? I also have an AM3+ EX9 and LLC was a little troublesome on that board so I thought I'd ask...
I don't have any high freq. Mem (2000-2200 @ best), so I'm still playing with CPU clocks.

Odd i used an coloured mixture with corrosion protection, terrible smell.

Not so sure about LCC 60 and 80 are good, 100 is abit too much on higher oc's and 0 causes too high spices on my side.

[justapost]

Interesting findings, I'm going to test it more on my config with memory set to 1833MHz and 1600MHz tomorrow or on Sunday.

If you have time to test cgMiner for me then below is all you need to run it. Just start Kaveri.bat and let me know if you will get any HW errors.
https://www.dropbox.com/s/evgtfegaca....2-windows.zip

In bat file you can play around with Thread concurrency and values up to 12200 should work and you can try to remove --no-restart from bat but on my motherboard it causes blank cgminer window (on Gigabye and 5800k it wasn't needed and it's to do with GPU FAN monitoring).

No hw errors so far if they look like this usually https://bitcointalk.org/index.php?PH...614#msg3849614

[justapost]

http://www.techpowerup.com/197338/ek...rocessors.html

EK Water Blocks, Ljubljana based premium water cooling gear manufacturer, is proud to introduce the EK-Supremacy PreciseMount Add-on Naked APU. This product is world-first of it's kind - it is the mounting mechanism upgrade for existing cooling device, designed to support de-lidded 1st- (Llano), 2nd- (Richland / Trinity) and 3rd (Kaveri) generation AMD A-series APU processors.

Wonder if these will work with my d-tek block.

[justapost]

Having an A4-4000 here, so delidded it with an razorblade, removed the remaining glue and replaced the thermal past with simple as5 and put the lid back on. Broke an pin by accident but got luck and the cpu does still work. Launched linux from usb with the voltage set to 1.6V and only an standard heatsink. Ran primesieve which usually generates as much heat as prime95 and reached 35? max. Was more an prove of concept so i have no before numbers to compare with. Will try this with the A10-7700K on sunday. Lid is very thick, might like some lapping as well.

[Daveburt714]

I'm doing some testing tonight concerning LLC...

With the CPU voltage set @ 1.375 I got the following results (idle/load):

0% = The highest overvolt (~.064)
60% = Slight Overvolt (~.016)
80% = Mostly flat w/ brief OV's of (~.008)
100% =Undervolt of (~.016) under load

Granted, these numbers will probably change depending on core voltage, but I would assume the ratio would stay the same...
I'll continue testing @ 60% LLC. I'd prefer a slight OV under load, rather than flat or an undervolt.

Now I'm just curious about the difference between a "set" Vcore and adding a Vcore offset...
For now I'll assume it has to do with CnQ (as on the C5F), where setting a high Vcore stopped CnQ from working properly, but using offset mode allowed CnQ to function...

I haven't pushed hard yet, but so far power consumption is looking decent on my chip.
Even @ 4.4Ghz with Prime95 torture test running I haven't seen system draw go over 200w yet...

Glad to see ASRock did a better job with early bios on this board compared the the EX9!

[justapost]

@Daveburt714: Thank you for the detailed LLC test, it's similar here. What is new to me is that there is an CPU intern and and PWM offset. Guess the CPU intern offset is an offset applied to the voltages in the p-states and the pwm offset is applied via the pwm without knowledge of the cpu. Could make an difference with apm handling.

Delidded the A10-7700K today, first i tried direct mount but that made things alot worse, difficult to mount it in a good way with my big tubbing. Chip did only boot at stock and radeon temps where so hight the cpu downclocked even at that settings and light loads. So i tried it with the ihs inbetween and still got abit worse results, line 5C higher radeon temps, which are my only concern, cpu temps are still at ~50C with ~1.5V 4.5GHz.
Did abit of reading and made a three layer carton frame ~1.2mm think. Temps where now at IHS level. Removed one layer and i'm back at my original temps. Removing another layer would result in an thickness of 0.4mm so too low.

[Yeroon]

No hw errors so far if they look like this usually https://bitcointalk.org/index.php?PH...614#msg3849614

All you're getting is HW errors - second line, A:0 (accepted shares) R:0 (rejected shares) HW: 179 (hardware errors)
Granted I'm pretty new to the crytomining stuff but that there is not working.

[Lightman]

No hw errors so far if they look like this usually https://bitcointalk.org/index.php?PH...614#msg3849614

As already said, only HW errors, no submitted WU's.
Thank you very much for testing this for me! At least now I know this is not my setup issue but broader AMD driver issue. I hope Cat 14.2 will solve it

[Yeroon]

Interesting findings, I'm going to test it more on my config with memory set to 1833MHz and 1600MHz tomorrow or on Sunday.

If you have time to test cgMiner for me then below is all you need to run it. Just start Kaveri.bat and let me know if you will get any HW errors.
https://www.dropbox.com/s/evgtfegaca....2-windows.zip

In bat file you can play around with Thread concurrency and values up to 12200 should work and you can try to remove --no-restart from bat but on my motherboard it causes blank cgminer window (on Gigabye and 5800k it wasn't needed and it's to do with GPU FAN monitoring).

What do you have the work size and T-C set to for kaveri? I've found 128 for work size and 3(640*4) for thread count worked best for my 7770, and intensity=15. This gives me an acceptable kH/s to W/U ratio. I'd expect something similar from kaveri since its only down 2 gpu cores.