Thanks for pointing towards those errors Yeroon. Tried it again at stock and mem at 1866 but still got only errors. Glad it helped Lightman!
Printable View
Thanks for pointing towards those errors Yeroon. Tried it again at stock and mem at 1866 but still got only errors. Glad it helped Lightman!
I've tried 2040, 4040, 8190, 12200, 12400 more crashes Intensities from 8 do 20 and even -g 2. There is no setting which will work properly. I suspect driver issue as it was giving HW errors on dedicated R9 290 too.Quote:
Originally Posted by Yeroon
Strangely now with 2x290 it's fine but I have suspicion it's to do with type of memory installed on gpu.
Hmm that A4-4000 i used for the delidding test and broke an pin thereby now only run's in single channel and only uses one dimm, spd info is show for both but only one slot is used. Population only that other slot results in an unbootable system.
http://abload.de/thumb/a4-4000-singlel3lo8.png
This has got to be the most comprehensive APU comparison ever, dude: very detailed.
It's kind of impressive how this chip answers the major complaints of Piledriver (Heat & Power Consumption)... ;)
I don't have any high Freq. mem to test with so I've only been focusing on core clocks in general.
Here's a CB r15 run with my best core clocks so far. No hard core stability test yet, but it was good for ~15 minutes of P95 blend:
http://i30.photobucket.com/albums/c3...ps21fade0a.png
The cool thing about this was, temps never went over 42c (water) and system draw stayed under 180w during the run!
I really hope AMD changes their mind and decides to make an AM3+ version of this uarch...
If they'd lose the iGPU and pack in the cores (8/10/12?), I think they'd have a nice CPU on their hands! :yepp:
guys, can someone with Kaveri testing PRIME for 30 minutes? I got some reports about underclocking in PRIME at stock settings...And seems not for overheating. Can you confirm it?
Any special settings for the test? Ran prime95 at many settings for up to 14 hours over- and underclocked and had not seen any oddities with temperatures. Older hwmon versions reported 30% higher temperatures and voltages at first but the latest version fixed that.Quote:
Originally Posted by FlanK3r
Update: After reading your request again i understand the issue abit better. With Turbo and APM enabled my chip sits at 3.7GHz idle, once i start prime95 it is usually at stock 3.4GHz/38C.
Nice chip, at that voltage's it's 4.4GHz teritorry on mine. What are your radeon temps during cb15 if you run the chip at 4.6GHz/1.55V? Here cpu temps stay in 50-55C area but radeon temps reach 65C and the system starts throttling.Quote:
Originally Posted by Daveburt714
Can it be cb15 likes tight memory timings?
I noticed some throttling with voltages higher than that too... Not sure what was causing it, but core temps were below 45c.Quote:
Originally Posted by justapost
HWMonitor doesn't show the GPU temps and it's the only monitoring I have installed so far. Does GPU-z show proper temps?
I haven't messed with the gpu yet other than raising Bclk which seems to get squirley @ anything >105.
I'm pretty familiar with this mem and know its good to ~2000 7-9-7 (Mushkin 1600 Black 1.35v @ 1.65v).
On the 9370 CB15 didn't seem to care much about mem timings, it really liked core clocks though!
Core voltage was set @ 1.35 with an offset of +.100 and LLC @ 60% on that run.
Can't remember if I bumped IMC v's to 1.3 or not. I made a few adjustments to it, but I'm pretty sure I brought it back down to 1.275 before this run.
Still haven't messed with any other voltages... :cool:
And you are using the iGPU?Quote:
Originally Posted by Daveburt714
http://abload.de/thumb/test4-m1526-a1299-p1476pop.jpg
You do not have "Radeon R7 Graphics" listed in HWMon like i have (see above screenshot)? I guess it's the NB or CPU heating up some part of the iGPU also containing that temperatur probe. However cpu always starts downclocking if that temp reaches 64C while cpu temps are at ~50C.
Using amd-sata drivers here and i have no 105MHz BCLK limit. With linux however i still have that limit and it's using the ahci driver. Was happy to get my adata kit running at CL9/1866MHz gotta try an lower CL. May i ask you to post screenshots of the SPD and Chipset aida64 pages here. I'm curious what subtimings your mem comes with.Quote:
Originally Posted by Daveburt714
At 4.4GHz i need around 1.425V to run prime95 for an few hours. At 1.3875V/+0/80% it's benchable with most stuff and also passes prime95 for 10 minutes.
Tested undervolting yesterday, with linux 0.8V was lowest for 2GHz but that did not get into windows, need around 1V there. Lowest idle requires C6 enabled. Like that with lowest fan and pump speeds my system requires 53W at the wall, overvolting/clocking has an minimal impact 0-3W+ on idle requirements. With C6 disabled and undervoltet to 2GHz/1V it reqires ~62W idle. Have not yet seen an negative impact of C6 on benchamrking results or oc stability so i usually turn it only off for overclocking tests to get rid off these low voltage reports C6 causes.
BKDG Info about P-State Registers
http://abload.de/thumb/p-stat-docs-176ibb.jpg http://abload.de/thumb/p-stat-docs-2cufto.jpg http://abload.de/thumb/p-stat-docs-3fhc6w.jpg
Third page mentions "Electrical Data Sheet for AMD Family 15h Models" has the info about CPU Vid, can't find that doc online but maybe it can be figured out by testing.
RW-Everything can be used to read the MSR Registers:
http://abload.de/img/p-statemsrhtd7a.jpg
P-State Current Limit:
http://abload.de/img/p-state-current-limitkff8j.jpg
P-State Control:
http://abload.de/img/p-state-control50fgy.jpg
P-State Status:
http://abload.de/img/p-state-statusirec9.jpg
P-State-0:
http://abload.de/img/p-state-0qlde8.jpg
CPUFid=22, CPUDid=0 => CoreCOF = 100*(CPUFid+16)/(2^CPUDid) = 100*38/1 =3800MHz
CPUVID[0-6]=24, CPUVID[7]=0, NBPState=0 (HighPerf) would be 1.4V = 1.55V - 0,00625V*CPUVID
IddValue=167, IddDiv=1 => 167A/10 = 16,7A
P-State-1:
http://abload.de/img/p-state-14wkjm.jpg
CPUFid=21, CPUDid=0 => CoreCOF = 100*(CPUFid+16)/(2^CPUDid) = 100*37/1 =3700MHz
CPUVID[0-6]=16, CPUVID[7]=0, NBPState=0 (HighPerf) would be 1.45V = 1.55V - 0,00625V*CPUVID
IddValue=26, IddDiv=0 => 26A
P-State-2:
http://abload.de/img/p-state-2leipu.jpg
CPUFid=20, CPUDid=0 => CoreCOF = 100*(CPUFid+16)/(2^CPUDid) = 100*36/1 =3600MHz
CPUVID[0-6]=36, CPUVID[7]=0, NBPState=0 (HighPerf) would be 1.325V = 1.55V - 0,00625V*CPUVID
IddValue=237, IddDiv=1 => 23,7A
P-State-3:
http://abload.de/img/p-state-3jwc39.jpg
CPUFid=18, CPUDid=0 => CoreCOF = 100*(CPUFid+16)/(2^CPUDid) = 100*34/1 =3400MHz
CPUVID[0-6]=16, CPUVID[7]=0, NBPState=0 (HighPerf) AIDA CPUVID 1.45V would be 1.45V = 1.55V - 0,00625V*CPUVID
IddValue=175, IddDiv=1 => 17,5A
P-State-4:
http://abload.de/img/p-state-4bvebh.jpg
CPUFid=15, CPUDid=0 => CoreCOF = 100*(CPUFid+16)/(2^CPUDid) = 100*31/1 =3100MHz
CPUVID[0-6]=62, CPUVID[7]=0, NBPState=0 (HighPerf) AIDA CPUVID 1.1625V = 1.55V - 0,00625V*CPUVID
IddValue=145, IddDiv=1 => 14,5A
P-State-5:
http://abload.de/img/p-state-52virv.jpg
CPUFid=12, CPUDid=0 => CoreCOF = 100*(CPUFid+16)/(2^CPUDid) = 100*28/1 =2800MHz
CPUVID[0-6]=74, CPUVID[7]=0, NBPState=0 (HighPerf) AIDA CPUVID 1.0875V = 1.55V - 0,00625V*CPUVID
IddValue=125, IddDiv=1 => 12,5A
P-State-6:
http://abload.de/img/p-state-6flcm1.jpg
CPUFid=8, CPUDid=0 => CoreCOF = 100*(CPUFid+16)/(2^CPUDid) = 100*24/1 =2400MHz
CPUVID[0-6]=94, CPUVID[7]=0, NBPState=0 (HighPerf) AIDA CPUVID 0.9625V = 1.55V - 0,00625V*CPUVID
IddValue=87, IddDiv=1 => 8,7A
P-State-7:
http://abload.de/img/p-state-7pffbs.jpg
CPUFid=24, CPUDid=1 => CoreCOF = 100*(CPUFid+16)/(2^CPUDid) = 100*40/2 =2000MHz
CPUVID[0-6]=106, CPUVID[7]=0, NBPState=0 (HighPerf) AIDA CPUVID 0.8875V = 1.55V - 0,00625V*CPUVID
IddValue=57, IddDiv=1 => 5,7A
wow, awesome post justapost !
Glad you like that info. Used rw-everything and crystalcpuid to tweak the p-states before k10stats was available. Seems to work the same way just switched into p-state-1 by writing 1 into bit 0 of msr 0xc0010062, the multi dropped to 31, so eighter my calculations and readings where wrong or things changed abit with kaveri. Should not be too difiicult to figure or the schema.Quote:
Originally Posted by FlanK3r
Hmm seems if i select p-state-0 (turbo p-state) via p-state-control register it uses the values of the first normal p-state (p-state-1). Flipping between p-state-1 and and two now while changing frequencies in the p-state msr registers and it seems to work.
Above settings where with default bios settings. If I switch to manual mode and select an cpuvid of 1.4V. p-state-1 and p-state-3 are now configured both at 3.4GHz/1.4V. p-state-0 is 1.6GHz/1.4V and p-state-2 remains at 3.6GHz/1.325V. Explains why turbo mode can not function well with manual frequency settings.
Will switch back to auto mode and try to tune turbo p-states. BTW amd overdrive also reports these odd turbo p-state settings.
Modifying turbo p-states works but sometimes the system freezes when i change an value in rw-everything.
Found another interesting register which can be utilized for easy power capping. It is possible to enable so called software p-state limits via Register 64 in PCI Device 18x3. Can be done with rw-everything.
http://abload.de/img/p-state-limit9oi46.jpg
Had to set bit 5 to enable software p-state limits. Bit's 28-30 define the maximum p-state, in my case p-state-6 so as you can see in cpu-z the frequency is at 2400GHz. For this to work apm and cpb should be disabled.
AMD Register Specs can be found here http://support.amd.com/TechDocs/4912...h-3Fh_BKDG.pdf.
RW-Everything here http://rweverything.com/
Looking into NB/GFX power states right now but have not yet found an easy way to access pci config space beyound FF.
Yes, I'm using the iGPU. Not sure why HWMonitor isn't showing the temp, just dl'd the other day it should be the newest version... :shrug:Quote:
Originally Posted by justapost
http://i30.photobucket.com/albums/c3...ps9b0eaa25.png
Thanks for the tip on drivers! I did a fresh W7 install so I'm running on W7's achi drivers I guess...Quote:
Originally Posted by justapost
I did install the newest catalyst package, guess that doesn't install AMD's sata drivers.
Sorry, I don't have a copy of the new Aida64. Here's a screen cap of the bios mem timing page though.
Hope it has the info you need:
http://i30.photobucket.com/albums/c3...ps58f9f377.png
Thanks for the tip on C6. I always turn it off for my HTPC box...
To be honest, I'm not sure exactly what it does other than cause the machine to enter a deep sleep mode.
Right or wrong, I was always afraid it may prevent my machine from waking to record TV shows. :p:
Yeah strange, same HWMonitor version here what biso version are you using? I'm on 2.81 (Beta).Quote:
Originally Posted by Daveburt714
Installed mine via my local os and software deployment server (opsi), used drivers from asrock website and ended up with amd-sata drivers. Also I'm able to flip between ide and ahci mode without boot issues.Quote:
Originally Posted by Daveburt714
I tried these settings but no boot at 1866 and even 1600. CL8 is the best my sticks can handle at 1866.Quote:
Originally Posted by Daveburt714
Aida64 does few more XMP profiles than the bios here, quite helpfull.
Adata XGP V2 2600 CL11 Kit
http://abload.de/img/adataxgpv2mdxmg.jpg
Crucial Ballistix Tactical 1866 CL9 Kit
http://abload.de/img/crucialballistixtactio8z0a.jpg
Like your kit the Crucials use relative low RFC's. Amd seems to like this.
In C6 cpu takes short nap's if idle. To save power i usually also turn on "Power Down Enable" on memory settings, saves 2-3W in idle.Quote:
Originally Posted by Daveburt714
Some bios settings mixed up my previous p-state settings. These look correct:
http://abload.de/img/p-statemsr-26vbu8.jpg
Register Name Frequency Voltage Current Power 0 Turbo 3 3800MHz 1,4375V 16,7A ~24W 1 Turbo 2 3700MHz 1,35V 26A ~35,1W 2 Turbo 1 3600MHz 1,325V 23,7A ~31,4W 3 P-State 0 3400MHz 1,25V 17,5A ~21,9W 4 P-State 1 3100MHz 1,1625V 14,5A ~16,9W 5 P-State 2 2800MHz 1,0875V 12,5A ~13,6W 6 P-State 3 2400MHz 0,9625V 8,7A ~8,4W 7 P-State 4 2000MHz 0,8875V 5,7A ~5,1W
Voltage Table (second table is easier to read CPUVID*2 are the hexnumbers 3-4 from the right of each p-state)
http://abload.de/img/voltages97oc7.png http://abload.de/img/voltages2clpn2.png
Frequency Table 1 (DID=0) Frequency Table 2 (DID=1) (hexnumbers 1-2 from the right of each p-state)
http://abload.de/img/frequencies1parqp.png http://abload.de/img/frequencies2b9pel.png
Did an quick test with the two offset modes. Both do not affect the voltages defined in the p-states directly.
With Chip intern offset, the cpuvid seleted in the bios gets applied to turbo p-state-2 and p-state-0.
With PWM offset the vid only applies to p-state-0.
So to overclock with cnq apm and cpb enabled using just an pwm offset and no change to cpuvid is the best way to go.
I'm running the newest release version (P2.70). Can you share a link for the beta? I'm not worried about stability right now, just havin fun testing... ;)Quote:
Originally Posted by justapost
PM me if you can't share publicly. I understand the risk and won't hold anyone responsible if things go south!
I'm REALLY glad to see you back for the for these new chips man!
Your still a guru in my eyes, and even though some of the stuff gets a little deep for me you were very helpful in teaching me to tune and understand the initial Phenom chips. :up:
BTW: Those settings from the CB15 run won't blend for more than 5 minutes today so they're obviously not stable...
They ran blend for ~15mins last night (weird).
As far as memory, I tried changing CL(CR) to 1T and my machine wouldn't even post!
Not sure if that'll help you or not, but my rig/mem prefers 2T... :)
Theres an Beta's section on the Asrock produkct page http://www.asrock.com/mb/AMD/FM2A88X...me6+/?cat=BetaQuote:
Originally Posted by Daveburt714
Nowadays you are the guru, you have been active all these years. :)Quote:
Originally Posted by Daveburt714
I just like to read these register specs and play around with em. It's the first APU i look a bit deeper into. Seems there are more interesting areas now. The NB root complex for example handles nb and gfx power states and there is an lookup table with nb and cpu p-state combinations used to find the right balanced p-states for an given workload. Wonder if these can be modified in an way that the apu always operates at lowest specs and thereby leaving more room for cpu oc's.
Yeah tough to track stable memory settings with prime95. With my kit it was easy to get 2133MHz CL9 (even T1) with ~1.625V. But even an 1MHz increase on BCLK required an huge increase on VDDA everything else did not work to make it stable. Max VDDA i had tried was ~1.3V and that blended over night. After an while i found that i also had to increase nb voltage at some poing (1150MHz mem), did not yet finetune the voltages again at 1150MHz so it's still in the fog what was really required.Quote:
Originally Posted by Daveburt714
http://abload.de/img/tweaking8bx2q.jpg
Howto: Optimize CPU P-States with RW-Everything
1. Download and Install RW-Everything http://rweverything.com/downloads/RwX64V1.6.5.9.zip
2. Edit C:\Programs\RW-Everything\RW.ini. Look for the Section [MSR_AMD_K8] and comment out the p-state related lines (see bottom right on Screenshot).
3. Disable Turbo (APM and CPB) and C6 in the Bios. Select Auto fpr Voltages and Frequencies and enable Cool and Quiet.
4. Boot into Windows and select the "High Performance" energy saving profile.
5. Launch RW-Everything and open the MSR Windows (Access->CPU MSR), scroll down to the bottom and you should see the p-state registers just commented out in the rw.ini file.
6. Now you can manually select an p-state. The p-states 0-2 are used for the turbo mode and can not be selected manually. The P-State-CTRL Register can be used to select p-states 3-7.
7. Double click on an number in the row "The P-State-CTRL" and column CPU (1-4 the changes will apply to all cores no matter which colum you choose).
8. Double Click on the number at the bottom right and enter an desired p-state (0 will select p-state-3, 1 p-state-4 and so on). Once you click on done the new values will be written to the register and an few seconds later the p-state will be selected.
http://abload.de/img/tweaking-ctrlncawv.jpg
9. Verify with CPU-Z that you are running at the voltage and frequency of your choosen p-state. Also you can compare the last four numbers of the COFVID_Status_register with those of the p-state you selected, they should match and represent the voltage and frequency.
10. To change the voltage of an p-state double click an number in the p-state row you want to change of any cpu column.
11. Double click the second number from the left at the bottom right and enter an hex value for the desired voltage (use the voltage table in the screenshot to find the right hex). If you also want to modify the frequency you can lookup the hex number for an frequency in the frequency table i posted earlier and enter that value as the first number at the bottom right. Click done and wait a second till the changes apply to the p-state registers.
http://abload.de/img/tweaking-p-stateb9lel.jpg
12. If you changed values for an currently active p-state you must switch to an different p-state and back to apply the new settings, otherwise just switch to our just modified p-sate and verify the changes where applied.
Board acts awkward when switching between manual and auto voltage and frequency settings. If i choose and cpuvid manual, reboot and the switch to auto mode the previous choosen voltage remains selected in p-state-0 and turbo p-state-2 (if it was greater than 1.35V).
After an cold boot it selects 1.45V for p-state-0 and turbo p-state-2, which can not be the defaul vid for 3.4GHz.
Did undervolting for all the p-states and the max turbo p-state. Assumen 1.25V as the default stock vid, bios thinks it's 1.45V (idle 84W load 169W)
Decreased the voltage in 0.0125V-0.025V steps and ran prime95 for at least 10 minutes. If an voltage was too low the system eighter froze or bluescreened with an clock interrupt error during the first few minutes blending.
30W-40W power savings possible at load this way.
P-State Frequency VID VID mix VID max Power idle Power load VID VID min VID max Power idle Power load Turbo-3 3800MHz 1.4375V 1.376 1.408 83W 166W 1.175V 1.128V 1.136V 68W 124W Turbo-2 3700MHz 1.35V Turbo-1 3600MHz 1.325V Pstate-0 3400MHz 1.25V 1.208V 1.224V 71W 133W 1.05V 1.008V 1.024V 64W 101W Pstate-1 3100MHz 1.1625V 1.128V 1.128V 67W 116W 1.0V 0.960V 0.968V 63W 95W Pstate-2 2800MHz 1.0875V 1.048V 1.056V 65W 101W 1.0V 0.960V 0.968V 63W 94W Pstate-3 2400MHz 0.9625V 0.928V 0.936V 63W 88W 0.875V 0.840V 0.848V 61W 83W Pstate-4 2000MHz 0.8875V 0.856V 0.861V 62W 82W 0.79375V 0.760V 0.768V 60W 77W
Modifications with RW-Everything work flawless if i stop things like CPU-Z, HWMon while running it.
Also modified the turbo p-states 3.8GHz->4.4GHz(1.425V) 3.7GHz->4.0GHz(1.25V) 3.6GHz->3.8GHz(1.05V) and enabled apm, cpb and C6. CPU sits at 4GHz most of the time and due to C6 the system requires 59W. Runninge CP15 multi test the cpu drops to 3.8GHz sometimes. Runnings CP15 single it slips into 4.4GHz but does not stay there regulary. Running CB15 OpenGL the cpu dropps to 2.4GHz and 2.8GHz. So for an future test i'll try to increase the frequencies of p-state-2 (2.8GHz) and p-state-3 (2.4GHz) and see if that makes an difference on the opengl score.
It may also be interesting to set p-state-0-4 to 4.6GHz(1.55V) to prevent frequency dropps if the cpu wants to throttle.