Phenom 9500 w/ MSI K9A2 Platinum

Ok, Now that my head has cleared, I figured out what was screwing me up last night (besides Beer ;) )

You said "Plus 9x NB and 11.5x CPU on Phenom 9600 doesn't work."
And I read it as "Plus , 9x NB and 11.5x CPU on Phenom 9600 doesn't work."
If you would only have said "Over 9x NB and 11.5x CPU on Phenom 9600 doesn't work." :rofl:

I actually do have a legit question though.. You said these were stock VID settings for the 9600 (non BE)?
CPU VID = 24
NB VID = 24

When I first opened P-States after flashing back to 1.13, these were the Pre-Sets...
CPU VID = 28
NB VID = 36
Which I assumed to be Stock...

I'm currently running @ CPU=24 / NB=36... If stock NB is 24 then I'll set it to that....

Yep, that's how they load from disabled->enabled... but, why not see what FID/DID are presets. ;)

Stock is 1.2000-1.250/1.250 for both parameters. You can disable P-States and bootup at stock all auto and check in AMD PowerMonitor.

On a side, you don't need high MHz for high bandwidth on Phenom...

http://img90.imageshack.us/img90/6962/fkfkfkfbw2.jpg


Tried CAS3 1000 today, my sticks can do CAS3 1066 on P35, but with this system = no go. The system will boot but it will boot with bad CMOS: reset and everything reset.

Also, I've been experimenting. Two things to note, two very important things:

* Phenom is starved for memory bandwidth. Yes, damn, I've been playing with high clocks and low clocks and Phenom performance is low at low RAM clocks (same CPU/NB) compared to at higher RAM clocks, say 400 3-3-3-3 compared to 500 4-4-4-4.

* Phenom is very IMC dependent. As you may know, running Super Pi at 70k efficiency when the best efficiency is around 75k is impossible. On an Intel chip, best you can manage is 49k when you can get 50k very easily without any tweak. But with Phenom I've tried this and you gain quite a bit of performance in applications through just IMC overclocking. Number calcs will show this up very quickly, like wPrime, Pi apps and so on. You'll be quite shocked at what you observe if you do it the right way. ;)

Check this. My system runs 1-core 2.8G bootup fine-> http://valid.x86-secret.com/show_oc.php?id=306352

1-core 2.9GHz fine but with high volts. Minimum needed to run 2.9G is 1.45V-> (click for full image)

http://img240.imageshack.us/img240/4308/29goh2.jpg

3GHz was max, it froze soon after-> (click for full image)

http://img88.imageshack.us/img88/9835/dsc02079ew0.jpg


BUT... I met very bad performance going from 2.75G to 2.8G. 2.8G was 2.2s slower than 2.75G in SuperPi 1M and 2.6G was 0.3s faster than 2.8G :confused:

http://img90.imageshack.us/img90/335...8250xs9.th.jpg

Yeah, it's probably called instability, I gather. I better stability test my RAM to see if it's okay, this is very typical of dying RAM. :(

Quote:

---

Originally Posted by KTE

Hmm..

---

^ That was your comment about MaxVid/MinVid.
Tried it here on the m2a-vm with 1,325V vcore and MaxVid is 1,3V here. Gonna inspect that register further.

Quote:

---

Originally Posted by KTE

Those RAM/L3 are directly comparable, they're the same NB/RAM. ;)
The L1/L2 caches aren't comparable because my CPU is 100MHz faster.
I'll run some 200x11 today and post here.

However you only ran one single threaded bench which doesn't support Phenom fully yet. Try and run Sandra/WinRAR as it supports Phenom multi-threaded memory access and that'll show the real bandwidth of Phenom better. Unganged will perform better than Ganged in multi-thread, EVEREST can't be used to show the real bandwidth but can be used to show the single-threaded memory bandwidth. ;)

---

Yep, I tried RightMark's multi threaded memory benchmark whom SocketMan recommended.
Unganged mode is faster here. Even more faster (ganged vs unganged relation) with ram at 1066 than at 800.

Quote:

---

Originally Posted by KTE

Here ya go Achim, everything's matched now even the CPU speed. Only thing different is you have the theme service switched off, I'm on XP Pro SP3 and RD790 chipset. ;)

---

Yep, theme service and some other useless services are deactivated by default. No international SP3 available I guess.
Beside that I see the biggest differences in latencies. MSI seems to be a tight better in unaganged, Asus in ganged mode.
Thank you for running those benchmarks, I expected bigger differences.

Quote:

---

Originally Posted by KTE

Interesting thing to note-> people really haven't worked out this platform properly yet, let alone a week or a few days for a review. Look below ;)

Power Consumption:

Code:

---

CPU 11.5x 2.3GHz @ 1.160V, 1.175VID
NB 9x 1.8GHz @ 1.050VID
HT 9x 1.8GHz @ 1.1V
2x 1GB RAM 800 @ 2.2V
HSF 1290RPM @ 5V - 400mA
Sapphire HD 2600 XT 800/700 @ 1.26V/DK
2x SATA II WD Caviar SE 80GB
Antec Earthwatts 430W

---

Hibernate: 9W AC
Standby: 55W AC
Idle: 94W AC
Load: 161W AC


Compare with stock system findings 15th Jan: Power Draw

Code:

---

CPU 11x 2.2GHz @ 1.160V, 1.175VID
NB 9x 1.8GHz @ 1.050VID
HT 9x 1.8GHz @ 1.1V
2x 1GB RAM 800 @ 2.2V
HSF 1290RPM @ 5V - 400mA
Sapphire HD 2600 XT 800/700 @ 1.26V/DK
2x SATA II WD Caviar SE 80GB
Antec Earthwatts 430W

---

Ganged DCT Mode:-
Idle AOD Green: 92W AC
Idle AOD Yellow: 95W AC
Idle AOD Red: 119W AC
Load: 156W AC

Unganged DCT Mode:-
Idle AOD Green: 94W AC
Idle AOD Yellow: 97W AC
Idle AOD Red: 118W AC
Load: 158W AC

;)

---

AMD should have merchendised the tlb-fix as a power saving feature. :rolleyes:
Is the load consumption also affected by the tlb-fix?

System CPU load power consumption stays the same, hence why I only left one value for both conditions.

Importantly, if any reviewer clicked that or opened AOD without changing it back to yellow/green, they will give higher power figures, 30-38W higher idling than they should be. ;)

Quote:

---

Originally Posted by KTE

System CPU load power consumption stays the same, hence why I only left one value for both conditions.

Importantly, if any reviewer clicked that or opened AOD without changing it back to yellow/green, they will give higher power figures, 30-38W higher idling than they should be. ;)

---

That's an pretty interesting find. I really want to know what is going on behind the scenes in yellow and red mode.
Have you ever tried CrystalCPU to disable the fix? If so how did it behave in terms of power consumption.
Looking at the linux kernel fix, the software workaround seems to be no such big thing, I wonder if MS will release an hotfix for this issue, I guess never. ;)

Quote:

---

Originally Posted by justapost

^ That was your comment about MaxVid/MinVid.
Tried it here on the m2a-vm with 1,325V vcore and MaxVid is 1,3V here. Gonna inspect that register further.

---

Hmm, had not translated hex to binary correct MaxVid is 1,25V. Are you able to set higher CPU VID's than 1,25V with an 9500 on your mobo?
M2A-VM allows higher VCore settings, AMD PM always shows 1,25V VID. That seems to be the correct handling and behaviour to me with higher vcore's.

Quote:

---

Originally Posted by justapost

That's an pretty interesting find. I really want to know what is going on behind the scenes in yellow and red mode.
Have you ever tried CrystalCPU to disable the fix? If so how did it behave in terms of power consumption.

---

I don't use a TLB patched BIOS though and the CrystalCPUID fix only changes it to the yellow state, which isn't much increase over the green.

Quote:

---

Looking at the linux kernel fix, the software workaround seems to be no such big thing, I wonder if MS will release an hotfix for this issue, I guess never. ;)

---

Microsoft will never do such a thing unless it profits them in a large way and hence, if it's perceived as a supplementary need to boost their sales. They have enough Vista issues to deal with yet frankly.

Here's some more bandwidth for ya. :D

http://img30.picoodle.com/img/img30/...wm_0cbcd28.jpg

Quote:

---

Originally Posted by justapost

Hmm, had not translated hex to binary correct MaxVid is 1,25V. Are you able to set higher CPU VID's than 1,25V with an 9500 on your mobo?
M2A-VM allows higher VCore settings, AMD PM always shows 1,25V VID. That seems to be the correct handling and behaviour to me with higher vcore's.

---

Nope. 1.25VID is max option you get, just like 9x NB and 11x CPU multi is max you get with 9500. Although they are downward unlocked and on the BE, the NB multi is downward locked. :(

Just a little posting...

1-core action: http://valid.x86-secret.com/show_oc.php?id=306513

Attachment 71638

Attachment 71639


Memories nothing, good BIOS like DFI one which runs 1250 4-4-4-4 can murder that. But look at the L3/IMC b/w... wanted to see how much MHz it takes to break that 10k barrier. :D

Again another fact about Phenom: memory/IMC bandwidth, latency and performance is not the same at X settings and Y IMC/RAM clock speeds, if the CPU speed is different. Lower CPU speed has lower IMC/RAM performance and higher latencies at the same timings/MHz than a higher CPU speed, big difference. You will see this difference in applications too, when scaling is plus 100%.

I also ran 2596MHz CPU and 2596MHz IMC on that just to check with the same RAM timings. B/W hardly broke the 9k barrier until I topped 2.7G on the CPU whilst keeping everything else constant. ;)

yeah.. agreed.. the higher cpu speed you have. higher memory efficiency you get. the higher memory bandwith will show.

Quote:

---

Originally Posted by KTE

I also ran 2596MHz CPU and 2596MHz IMC on that just to check with the same RAM timings. B/W hardly broke the 9k barrier until I topped 2.7G on the CPU whilst keeping everything else constant. ;)

---

Congrats now go for 10k Mem read an 100k L1 read. :D
I'm happy if I can reach ~8k here.

tictac->

Attachment 71649

:yepp:

That lady in your av is making me look at her... :p:

Quote:

---

Originally Posted by justapost

Congrats now go for 10k Mem read an 100k L1 read. :D
I'm happy if I can reach ~8k here.

---

Problem is BIOS. We don't have a BIOS which runs plus 1:2 divider without big RAM/IMC performance penalty, even when the AOD red button is chosen. What I had when I ran 520 5-5-5-5 2T 15tRC 59MAL 1.9G NB with my 9500 is still higher than these new BIOSes give with 600 4-5-5-15 2T tRC11 MAL51 2.5G NB.

On this BIOS, oc is harder since you're limited to 2.66 RAM divider and the lowest RAM timings available are 4-5-5-15-12 and only 2.5V which is 2.41V real. Furthermore, NB does not boot over 2.596G and while it's easy to get that alone (high volts/cooling), you can't have high CPU, high NB and high RAM clocks with low timings synchronously -> you need monstrous voltage on all of them and loose timings. And to make it worse, all the new BIOSes don't allow over 2.5V DIMM and all give me only 4-5-5-5-15 as the lowest timings at 2.66 divider. Even worse, my chip can't do plus 250 HT ref. so I'm stuck at max 500. That's why I revert to the older BIOS, it has far better performance although only 1:2 RAM divider. :(
My RAM is 800 3-2-2-1 at 2.0V, 1000 3-4-4-4 at 2.2V, 1200 4-4-4-4 PL5 tRFC35 10-10 at 2.2V and 1325 5-5-5-15 PL5 at 2.35V already verified on 3 different Intel P35 boards. With Phenom/790Fx, plus 1220 hasn't booted up even with 2.5V at 6-5-5-15 and 2G NB. :(

Here's some trial runs for you with different RAM latency/speed. Look at the trend...
(click for full size images)

http://img118.imageshack.us/img118/9...6002475gc0.jpg

http://img352.imageshack.us/img352/9...0247548zj0.jpg

http://img205.imageshack.us/img205/865/13gbhh1.jpg

http://img352.imageshack.us/img352/1...6002475ko4.jpg


That last Sandra would've been over 9800 in EVEREST Ganged but it was very hard to get stable and Sandra is tough on the RAM too, loads everything full so it fails quick with a little instability.

100k L1 is not possible with this chip, volts needed are too high, above air cooling and it needs plus 2.95G, which this chip won't do sub 1.53V.

Quote:

---

Originally Posted by KTE

tictac



That lady in your av is making me look at her... :p:

---

twice :D:rofl:

Quote:

---

Originally Posted by KTE

Yep, that's how they load from disabled->enabled... but, why not see what FID/DID are presets. ;)

Stock is 1.2000-1.250/1.250 for both parameters. You can disable P-States and bootup at stock all auto and check in AMD PowerMonitor.

---

Thanks KTE, don't know why I didn't think of that... You were right, when I turned it to AUTO it booted with 1.20v on the CPU, and 1.25v on the NB.. :D

I'm not afraid to turn the NB up a little now, not trying to be a wimp, but with the reviews I've been reading about B3 this chip might have to last me until the 45nm parts come out...

Quote:

---

Originally Posted by jonspd

twice :D:rofl:

---

Shes looks like someone I know from Cali. called Jaymie, that's why. ;)
She has a pic that was taken in her car as she sat in the drivers wearing the seatbelt and that one looked similar.

Here's another plus 9.7k EVEREST read at different clocks, found it on the 2nd drive, must have saved it whilst messing around.

Attachment 71667

Phenom doesn't boot plus 626MHz RAM, no matter what volts/latencies/cooling (plus zero), it just BSODs at Windows load-up. With RAM that is known to run this flawless at 2.3V. :(

Quote:

---

Originally Posted by Daveburt714

I'm not afraid to turn the NB up a little now, not trying to be a wimp, but with the reviews I've been reading about B3 this chip might have to last me until the 45nm parts come out...

---

1.3VID? Should be easily safe if temps stay good. However, have you tried to see if VID actually changes when moved up and down one notch?

Quote:

---

Originally Posted by KTE

1.3VID? Should be easily safe if temps stay good. However, have you tried to see if VID actually changes when moved up and down one notch?

---

Yep, according to AMD Hardware Monitor it does.. :)
Man you've been so much help, I really appreciate it!

I'm testing @ 1.30 (VID=20) on the CPU/NB right now. Actually booted at HT REF of 260, it's never done that before.. :up:

Use Memset to check NB speeds because it works perfectly fine and is a very light and quick app, won't reboot or freeze the system even when it's extremely unstable. When you reach near 2.1G NB, you need to give it more volts for sure. When you then reach 2.2-2.25G NB, you need to feed it even more volts for stability. Had I known this with my Phenom 9500, you'd probably have seen higher clocks, but I didn't. You're welcome BTW. :)

Well, right after I typed that message my machine got really squirley... :(

Had to reset CMOS a number of times, even the Insert/Delete combination wouldn't work... I had to set both VIDS back to 24 in order to get it running again.

This is a knee jerk reaction, but it seems like the CPU doesn't like much over 1.25v's... I'll try again, upping the NB only.

Here's an ss of Memset:

http://i30.photobucket.com/albums/c3...0201MemSet.jpg

Well, moving either one of the VIDS, CPU or NB results in a No Post now...

I tried NB=24/CPU=23, and NB=23/CPU=24, both required a CMOS jumper reset... :(

Man this thing is wierd!! Just when you think your getting somewhere it kicks ya in the azz.....

Don't change CPU VID, it's locked. ;)

KTE, you would really be my Hero if you could post a couple photos of your BIOS (Highest/Most Stable Settings).. I know thats alot to ask, and I'm not trying to be a pain...
I realize you have a BE, but at least it may give me a point of reference.. :)

Cell Menu (below all the DOT stuff), and P-States... Would be perfect. ;)

I can actually lower my CPU VID to 24 (1.25v), So at least I can get a little out of it (stock was 1.20v/28)...

I actually have no clue how your putting more V's on the NB/IMC, if I even touch NB VID, my machine won't boot...

Quote:

---

Originally Posted by KTE

tictac->

Attachment 71649

:yepp:

That lady in your av is making me look at her... :p:

---

Me too... :rofl:

Well... macci said that too?...hehe... back in nforce2 days we used to play with sandra bandwith memory efficiency. We got higher memory bandwith effeciency when the cpu clock higher.... :yepp:

where can you download the latest BETA version?any changes?

Quote:

---

Originally Posted by madfaze

where can you download the latest BETA version?any changes?

---

A guest user on this forum recently added me to Windows Live Messenger and drew my attention to this post on a German MSI forum:

http://www.msi-forum.de/thread.php?t...user=0&page=13

The user is awaiting validation (and has been for roughly 2 weeks) so if anyone can help with that, that would be great - he's really anxious to get in here and help out.

Anyway, about 2 quarters down that page there are some beta BIOSes... If anyone's feeling brave (KTE, here's looking at you =P) you could try them out. I would, but I've just got my system working satisfactorily and I don't want to ruin it now.

Just an idea, anyway. :)

Quote:

---

Originally Posted by Hosser

A guest user on this forum recently added me to Windows Live Messenger and drew my attention to this post on a German MSI forum:

http://www.msi-forum.de/thread.php?t...user=0&page=13

The user is awaiting validation (and has been for roughly 2 weeks) so if anyone can help with that, that would be great - he's really anxious to get in here and help out.

Anyway, about 2 quarters down that page there are some beta BIOSes... If anyone's feeling brave (KTE, here's looking at you =P) you could try them out. I would, but I've just got my system working satisfactorily and I don't want to ruin it now.

Just an idea, anyway. :)

---

thanks!:D

Please let us know how you get on =)