Phenom 9500 w/ MSI K9A2 Platinum

Those should be more than stable. If its Memtest stable, I'm at loss why 232HT isn't but due to CPU requiring more volts, better cooling or just the CPU limit itself. :shrug:

But Phenoms are a little different so a few additional tests should be, power cycling, hot/cold reboots, standby/hibernate, let system idle. Overnight is best case for lengthy jobs.

NB speed is nothing, anyone should be able to get that and more. I've had 2.6G with the last one and this one should do it too. ;)

Need to test max HT ref, max NB, max stable on both as well as with high/low CPU multi and then max stable CPU frequencies, max benchable, best performance, sweet spot for 24/7 (if it can be better than the previous 2.7G one), temps/power at that sweet spot, lowest voltages for that sweet spot, lowest voltages/temps/power at stock speeds possible yet. I find more of an enthusiasm in undervolting than in overvolting actually. ;)

Temperature discrepancies
I realize some Phenom temperature sensors are reading near ambient values on air, looking around and that makes them quite untrustable if anyone has such a chip (one of my 9600s was affected). What I'm at loss with is, why do they shut down at 74C when that is the chips actual shut down limit? Also why does it not make it unstable at 60C, if actual temps were plus >10C? 70C real will make Phenom totally unstable in operation and damage it. Furthermore, why does the stock HSF able to just dissipate 95W not keep it cool and keep it quite warm, at 50C load if we know that can't be 60C actual?

The only way we can check is by crosschecking with the chips physical limits. 70-75C will assert PROCHOT and shutdown Phenom instant as its beyond tolerance. Then we can work from there onwards to finding the actual temps (approx) by using TDP/cooling/ambient/load/HSF thermal resistance values.

Quote:

---

Originally Posted by jesteronyer6

One last Q is what is the best method for flashing these MSI boards, in the past I used winflash.

---

If you've got a flash drive you can try MSI's tool here too. I used it to update mine and it was completely painless.

http://forum.msi.com.tw/index.php?topic=113295.0

You'll need to create an account there to get to the tool, if you don't already have one.

BTW I thought setting the light in the upper right hand corner of AoD to red/orange defeated the TLB patch. Is that wrong?

Quote:

---

Originally Posted by Blacklash

BTW I thought setting the light in the upper right hand corner of AoD to red/orange defeated the TLB patch. Is that wrong?

---

In theory, it was supposed to be correct, but in practice, performance is still a margin of the actual patchless. Decompression/compression/ encryption/decryption/ encoding/decoding/ calculations seems to be affected the most severely.

Synthetic tests may still run giving normal performance output but real workloads won't. Super Pi is usually 2-3 seconds slower with the method you mentioned over a non-patched BIOS.

Looking at the BIOS state, it seems AMD might have (most likely actually) pushed MSI to implement the patch at any cost. It is AMD who decides most of their platform decisions in case you didn't know.

Quote:

---

Originally Posted by KTE

In theory, it was supposed to be correct, but in practice, performance is still a margin of the actual patchless. Decompression/compression/ encryption/decryption/ encoding/decoding/ calculations seems to be affected the most severely.

Synthetic tests may still run giving normal performance output but real workloads won't. Super Pi is usually 2-3 seconds slower with the method you mentioned over a non-patched BIOS.

Looking at the BIOS state, it seems AMD might have (most likely actually) pushed MSI to implement the patch at any cost. It is AMD who decides most of their platform decisions in case you didn't know.

---

I think you're right. 3mark06 CPU test is down with the latest official BIOS even if I toggle the TLB fix off with AoD. It's only on the order of 40pnts and I still hate this big brother crap.

well 2.6 is stable

11.5x227 is about the same as 9.5x272

230 was ok about the same as 274 was

maybe I'm maxed and need to drop to 2.58ish and be settled until b3 :D

Just tested 1.212VID CPU 1.2V idle/1.192V load 2.3G - 1.3VID NB 2.4G = easily passed. ;)

I'll go lower next-> 1.2VID 1.192V idle/1.184V load CPU 2.3G
That'll need many testing hours.

Guys with Phenom: can you help out Felix here to test his new Memset builds please: http://www.xtremesystems.org/forums/...=114754&page=2

Feedback is very welcome (in his thread). Try and make it precise and test it properly to find any bugs so that we can work to fix them. One of the best apps around and he puts a lot of hard work into it for nothing back when he deserves much more back :)

Specifically, change the CPU multi from stock and see if the values are picked up correctly in it (BIOS change is always best, but try AOD too).

This goes out to fellas using any Phenom on all boards.

Quote:

---

Originally Posted by KTE

Just tested 1.212VID CPU 1.2V idle/1.192V load 2.3G - 1.3VID NB 2.4G = easily passed. ;)

I'll go lower next-> 1.2VID 1.192V idle/1.184V load CPU 2.3G
That'll need many testing hours.

---

Lol, that's what i did in my spare time the last week. The odd thing is sometimes you can use a much lower nb voltage if you increase cpuv a little.
With cpuv 1,2V I can go down to 1,025V on the nb here. Looking for the min nb with 1,185V for the cpu atm. I want to annotate that I use relaxed memory settings 800-5-5-5-15. Can be that a higher nb voltage is required for opzimized settings.

EDIT: You might also try 1,185V/1,075V for CPU/NB at stock, lowest possible on the M3A with a 9500.

Got one question here, hope the pros can help me ans it! Really appreciate it guys! Thanks!

Im now running the 5000X2 on my K9A2 plat while waiting for my 9500 to come.

I change the multi to 15x, stock voltage, ram is 1:2, DDR2 800.

Its running at 3Ghz, but my ram is running at 375 instead of 400. Why is it this way?

Is it due to the 15X multi?

Quote:

---

Originally Posted by justapost

Lol, that's what i did in my spare time the last week. The odd thing is sometimes you can use a much lower nb voltage if you increase cpuv a little.
With cpuv 1,2V I can go down to 1,025V on the nb here. Looking for the min nb with 1,185V for the cpu atm. I want to annotate that I use relaxed memory settings 800-5-5-5-15. Can be that a higher nb voltage is required for opzimized settings.

---

NB voltage? We can't see NB volts, we can only see the VID value. Though I know NB volts rise and fall with VID selection, but they don't rise and fall much by just VID changing unless you push to extremes. I actually want lowest CPU voltage first because it's the biggest temp/power hog, NB voltage can be taken near 1.2VID at least (1.192V) at stock stable, which is OK for me.

Right now, I've just gamed 30 minutes and then primed stable >5.5 hours (still going) at:
202 x 12 = 2423MHz CPU @ 1.225VID 1.216V idle/1.208V load
202 x 10 = 2020MHz NB @ 1.275VID

http://img220.imageshack.us/img220/6...tablbt8.th.png

Consider, stock was 200 x 11.5 = 2.3G @ 1.25VID 1.240V
That was just to test headroom at those volts. An easy and quicker way to tell if any lower VID/V is unstable.

Quote:

---

EDIT: You might also try 1,185V/1,075V for CPU/NB at stock, lowest possible on the M3A with a 9500.

---

1.2VID gives me 1.192V idle/1.184V load but 1.075VID will probably give near 1.0V for the IMC. That might b a little too unstable I'll try it though. :)

Quote:

---

Originally Posted by ragnarok95

Im now running the 5000X2 on my K9A2 plat while waiting for my 9500 to come.

I change the multi to 15x, stock voltage, ram is 1:2, DDR2 800.

Its running at 3Ghz, but my ram is running at 375 instead of 400. Why is it this way?

Is it due to the 15X multi?

---

Read up on some guides for X2 oc'ing and you'll do much better overall (google). ;)

With X2: (HTT speed) x (CPU multiplier) / (memory divisor) = RAM clock

200 x 15 (2000) / 8 = DDR2-750

Use this tool for oc and it'll give you the different frequencies at different settings: http://www.xtremesystems.org/forums/...d.php?t=120432

Although you don't need it if you understand it. :)

Quote:

---

Originally Posted by KTE

NB voltage? We can't see NB volts, we can only see the VID value. Though I know NB volts rise and fall with VID selection, but they don't rise and fall much by just VID changing unless you push to extremes. I actually want lowest CPU voltage first because it's the biggest temp/power hog, NB voltage can be taken near 1.2VID at least (1.192V) at stock stable, which is OK for me.

---

I have two settings in my bios prozessor and northbridge voltage. VCore (AOD,CPU-Z) seems to be prozessor voltage and same as CPU VID here (AOD), NB voltage is NB VID(AMD PM).

Quote:

---

Originally Posted by KTE

Right now, I've just gamed 30 minutes and then primed stable >5.5 hours (still going) at:
202 x 12 = 2423MHz CPU @ 1.225VID 1.216V idle/1.208V load
202 x 10 = 2020MHz NB @ 1.275VID

---

Was the system unstable at 202x10 with a higher VID/Vcore?
With 1,2125V Vcore I can reach 204x11 max here. With 1,225V it's 208 (or 206)x11.

Quote:

---

Originally Posted by KTE

1.2VID gives me 1.192V idle/1.184V load but 1.075VID will probably give near 1.0V for the IMC. That might b a little too unstable I'll try it though. :)

---

I use a 9500 so you might try it with an 11x multi. For 2310Mhz I need 1,25V/1,05V (prime95 stable for 17 minutes). Can be that my low nb requirement has todo with the relaxed memory timings.

Yeah how VIDs work is like so:

HEX = VID -> Voltage set Idle/Load

28 = 1.2 -> 1.192V/1.184V
27 = 1.212 -> 1.20V/1.192V
26 = 1.225 -> 1.216V/1.208V
...

That voltage is default for those VIDs unless you manually change the voltage.

I've passed 1.2VID 1.192V Idle/1.184V Load for 2.4GHz CPU (200x12) and 1.1VID 1.8GHz NB

I've just finished benchmarking and now started priming 1.050VID NB 1.8GHz at the above CPU settings (20mins).

http://img222.imageshack.us/img222/8...minsbb1.th.png

Unganged is much harder on the IMC than Ganged, so you need higher volts/higher RAM timings for a setting at Unganged than in Ganged DCT mode. Usually CR, tRFC, tRC and MaxAsyncLat is what suffers in Unganged mode compared to Ganged.

However, I cannot get CnQ to work or kick in at all even at stock. :confused:

Quote:

---

Originally Posted by justapost

Was the system unstable at 202x10 with a higher VID/Vcore?

---

For the NB you mean? Yep, even with 1.275VID 2.5GHz runs fine.

Quote:

---

With 1,2125V Vcore I can reach 204x11 max here. With 1,225V it's 208 (or 206)x11.

I use a 9500 so you might try it with an 11x multi. For 2310Mhz I need 1,25V/1,05V (prime95 stable for 17 minutes). Can be that my low nb requirement has todo with the relaxed memory timings.

---

This chip is doing much less. See what I'm priming now... although I've learnt many lessons for Phenom and it's even more paramount now, that prime is not the all end for stability. I will have to power cycle, cold/warm boot, idle, overload by opening many applications at once, trying to open many tabs in Firefox at once, 2k6 looping, load different parts of the core (FPU/caches/IMC), Systools 32M, wPrime 1024M, and then do some gaming. ATM, I'm running my stock fan at its lowest speed just to have highest temps and see if its unstable. I haven't tried dropping CPU VID/V more than 1.2VID/1.192V at 2.4G yet.

CPU - 2300MHz (200x11.5) 1.175VID 1.160V Idle/1.152V Load
IMC - 1800MHz 1.050VID (near 1.0V)
Idle - 98W AC
Load - 165W AC ;)

28min gaming+power cycling+app. over-loading+2k6+2k3+website surfing+30min idling+Memtest+wPrime1024M+Systools 32M+>4hrs P95, still going strong.

http://img91.imageshack.us/img91/736...0180au2.th.png

That's the IMC limit before it starts failing. 1.0VID failed after 40 minutes P95 on one core.

I'm going to undervolt the CPU more yet and then see what I reach. Let's try 1.0V ;)

Thanks KTE!

Quote:

---

Originally Posted by KTE

CPU - 2300MHz (200x11.5) 1.175VID 1.160V Idle/1.152V Load
IMC - 1800MHz 1.050VID (near 1.0V)
Idle - 98W AC
Load - 165W AC ;)

---

Yep, does not help that much in idle but under load it can save ~10W.:yepp:

Quote:

---

Originally Posted by KTE

28min gaming+power cycling+app. over-loading+2k6+2k3+website surfing+30min idling+Memtest+wPrime1024M+Systools 32M+>4hrs P95, still going strong.

---

Wish I could do more testing but i'm looking for rought numbers at ref HT's in 2MHz steps so it whould take forever. So my current criteria is 20 minutes prime95, then i run cinebench10 and note the power consumption during rendering and the time the benchmark tool and the temps during the benchmark and in idle. That way I can calculate the "render enegry" required at given ref HT's. I will be finsihed in a few days.

That's the IMC limit before it starts failing. 1.0VID failed after 40 minutes P95 on one core.

Quote:

---

Originally Posted by KTE

I'm going to undervolt the CPU more yet and then see what I reach. Let's try 1.0V ;)

---

He He, I found if you put a little more vid on the cpu you can go lower with the nb.
Assuming the cpu can handle 100A and the nb 20A the average voltage needed can be calculated like this:

u = u1*4/5+u2*1/5

This average is very close for two different cpu vid/nb vid settings at the same ref HT, the diference is ~0,005V.

KTE

What ram are you using in your system?



I placed the ram in the two differnt slots 1 and 3. Both stick are now seen as Single Channel. 2GBs. System posted with no memory timings tweaks

Using POH bios with TLB fix disabled. Seems like the best bios with my ram.
Still cant get it dual channel.

Going to run that winrar test on this bios, On the Phenom.

I ran some tests on my BE 5000x2 with the k9A2 2gigs I get 970kbs
also on my 939 system 3800x2 2gigs RAID 0 I get 690kbs

A phenom gettting in the 700s sucks.

What is the Xpress option in this bios???
Right next to TLB fix.

in order to run ram in dual channel it has to be different color slots. it may still show that it's in single channel when you post (bug).

1+2 or 3+4

Quote:

---

in order to run ram in dual channel it has to be different color slots. it may still show that it's in single channel when you post (bug).
1+2 or 3+4

---

Yes but if you paid attention to my whining :)... I cant get a post with my Cosair Dominator 8500 with either in 1+2 or 3+4

I have to take the second stick out everytime. I have a resetup my bios because if gives a bad checksum. Then I play the with more timing voltage NB volt. Divider. Nothing.
So I want to get the exact sticks that you or Kte has so I can have some more fun

thats cause ddr2 1066 has issues...


I think the 1.33 betas may help you out alittle.


http://www.xtremesystems.org/forums/...&postcount=859

Quote:

---

Originally Posted by ragnarok95

Got one question here, hope the pros can help me ans it! Really appreciate it guys! Thanks!

Im now running the 5000X2 on my K9A2 plat while waiting for my 9500 to come.

I change the multi to 15x, stock voltage, ram is 1:2, DDR2 800.

Its running at 3Ghz, but my ram is running at 375 instead of 400. Why is it this way?

Is it due to the 15X multi?

---

how high can you go on those platinum and BE proc?

Max 100% stable found at stock MHz was my last post. ;)

VID (Idle Volts/Load Volts)

CPU - 2300MHz (200x11.5) 1.175VID 1.160V Idle/1.152V Load
NB - 1800MHz 1.050VID (near 1.0V)
Idle - 98W AC
Load - 165W AC

NB VID 1.000 booted, never froze, surfed web easy, but failed Prime95 inside 10 minutes.
CPU VID 1.138 (1.120V) and 1.162 (1.152V/1.144V) booted and idled perfectly running Thunderbird, WinRAR, SuperPi 1M and Firefox. Ran Prime95; the first failed inside 2 minutes and the second setting core3 failed after 40mins. The rest ran fine till the 3hr mark (never tested beyond).

BIOS/CPU Problems:
-Cannot boot below NB 9x multi on Phenom 9600 BE (both of my last ones - can do with Phenom 9500 and 9600 though)
-Cannot boot plus HT 10x multi.
-AOD is extremely buggy if HT is moved above 9x through BIOS (it'll freeze/reboot).

Max HT POST: 268MHz
Max HT full bootup: 265MHz (got excited before capturing, moved it up and system froze - not worked since then)
*Max HT Validated: 248MHz (stable easy - there's a black hole after this 1MHz)
Max CPU POST: 3240MHz
Max CPU full bootup: 3040MHz (froze on capturing screenshot, damn PNG compression -volts makes no difference, never happened since)
**Max CPU validated: 2812MHz (locks up idling)
Max CPU benchable: ~2760MHz
Max CPU stable: 2756MHz (with sub 1.4V, not 1MHz more stable is possible)
Max NB stable: 2500MHz
Max HT stable: 2480MHz

*: http://valid.x86-secret.com/show_oc.php?id=304234

http://img239.imageshack.us/img239/3783/cpumaxvt2.jpg

**: http://valid.x86-secret.com/show_oc.php?id=304222

http://img239.imageshack.us/img239/2219/maxhtnbrq0.jpg


I will later post the BIOS AM2+ P-State mappings and what volts they give you to the limit of my testing so far. I am too short of time to test yet though and haven't even tried plus 1.4V oc limit yet. OC stability has gradually started decreasing. What I required 1.32V to do before (2.726G) now requires 1.368V minimum or else it'll lock up idling. I'm going to just Prime95 cycle it days on end to see if there is degradation, which I expect as seen with my last Phenoms. What happens is, they start requiring more volts, usually +0.1V more for 100MHz less after 3 weeks at a particular MHz/V than when first stability tested. So say I have 2.7G stable at 1.328V... it becomes 2.58-2.6G at 1.4V. :(

Quote:

---

Originally Posted by justapost

Yep, does not help that much in idle but under load it can save ~10W.:yepp:

Wish I could do more testing but i'm looking for rought numbers at ref HT's in 2MHz steps so it whould take forever. So my current criteria is 20 minutes prime95, then i run cinebench10 and note the power consumption during rendering and the time the benchmark tool and the temps during the benchmark and in idle. That way I can calculate the "render enegry" required at given ref HT's. I will be finsihed in a few days.

That's the IMC limit before it starts failing. 1.0VID failed after 40 minutes P95 on one core.

He He, I found if you put a little more vid on the cpu you can go lower with the nb.
Assuming the cpu can handle 100A and the nb 20A the average voltage needed can be calculated like this:

u = u1*4/5+u2*1/5

This average is very close for two different cpu vid/nb vid settings at the same ref HT, the diference is ~0,005V.

---

Are you sure about this Achim? Currents are separately regulated and limited. I'm not sure how NB VIDs/Volts affect CPU VIDs/Volts you know, they are totally separate and independent of each other, because the IMC has it's own separate voltage from the cores, decided initially by the NB VID; on the 5 phase MSI RD790, CPU gets 4 phase PWM regulation with a maximum of 110A, so 27.5A in each phase and the IMC gets 1 separate phase for 20A maximum. Both have separate VRMs controlling their supply voltages, and both attain independent voltages. This is why that fella burnt up his K9A2 CF PWM area, it can't handle 110A+20A through the VRMs.
So, if I test 2.3G CPU/1.8G IMC at 1.2V/1.1VID and it fails, moving it to 1.4V/1.1VID or 1.4V/1.05VID doesn't get me any more stability at all, it still fails. ;)

Quote:

---

Originally Posted by Brains1

KTE

What ram are you using in your system?

---

2x 1GB Crucial Ballistix Tracer PC2-8500 (D9GMH) 5-5-5-15-30 2T 2.2V.

Quote:

---

I placed the ram in the two differnt slots 1 and 3. Both stick are now seen as Single Channel. 2GBs. System posted with no memory timings tweaks

---

That's single channel but did you change it to unganged mode?

Quote:

---

Using POH bios with TLB fix disabled. Seems like the best bios with my ram.
Still cant get it dual channel.

Going to run that winrar test on this bios, On the Phenom.

I ran some tests on my BE 5000x2 with the k9A2 2gigs I get 970kbs
also on my 939 system 3800x2 2gigs RAID 0 I get 690kbs

A phenom gettting in the 700s sucks.

---

IDK which settings you had on the 5000+BE, but 2.64G/1.2G HT/880MHz 4-4-4-4 1T RAM gets 1332KB/s with 5000+ BE. Want to see what Phenom gets on a rough run with CPU 35MHz lower and high RAM timings?

http://img81.imageshack.us/img81/7572/winraree4.jpg

That's ~150KB/s lower than a 3.6G Q6600 running 1200 5-5-5-15 RAM.

Quote:

---

What is the Xpress option in this bios???
Right next to TLB fix.

---

AutoXpress? I didn't try the new BIOSes properly apart from just RAM/TLB patch because that's all that I needed more and they had next to no BIOS OC options I wanted. I always left that option disabled, sorry.

Quote:

---

Originally Posted by Brains1

Yes but if you paid attention to my whining :)... I cant get a post with my Cosair Dominator 8500 with either in 1+2 or 3+4

I have to take the second stick out everytime. I have a resetup my bios because if gives a bad checksum. Then I play the with more timing voltage NB volt. Divider. Nothing.
So I want to get the exact sticks that you or Kte has so I can have some more fun

---

Have you changed DRAM timings and RAM volts to what it's rated for with just one stick in and then reboot>add 2nd stick>boot?

Quote:

---

Originally Posted by jonspd

thats cause ddr2 1066 has issues...

I think the 1.33 betas may help you out alittle.

http://www.xtremesystems.org/forums/...&postcount=859[/

---

Yep, the new P0G/P0H and 133 BETA can run RAM at all dividers perfectly.

Looks like my testing with Phenom is now about to end. ;)
Only some benchmarks left now.

Max CPU stability, NB stability, HT ref. stability and HT link stability has been reached.

Max HT and NB combined, stable: http://valid.x86-secret.com/show_oc.php?id=304383

http://img352.imageshack.us/img352/7573/25922os3.png


When 1GHz HT was out, people said that's it! More HT makes no difference, and some assume this to be reflected over on Phenom too, when that 'aint so. As I have shown earlier, GFX perf. increases with higher HT clocks from 1GHz to 2GHz. However, I thought there may not be much more to gain with high HT clocks after 1.9GHz.

I've tested 2.043GHz and 2.497GHz HT in AquaMark 3 at the same settings, stock GPU, but changing the HT link speed. See the in-detail results below:

1. http://img153.imageshack.us/img153/51/2043um7.png
2. http://img352.imageshack.us/img352/2316/2500zw4.png

http://img84.imageshack.us/img84/8889/altogetherkp7.png

Just FYI, AM3 shows higher GFX score with more CPU MHz, always. As a comparison, 2.268GHz CPU/2.592GHz HT/NB get 16347 GFX score with Sapphire HD 2600 XT at stock.

In the tests, scaling is hardly there but still, look at the AV.triangles/second. ;) You'll see the GFX perf. drops overall but at the same time the CPU perf. jumps a little and enough to warrant a slightly higher end score.

Thus, there isn't much to gain from 2GHz to 2.6GHz HT (I did bench 2.592GHz HT with a lower CPU too) but there is a slightly increased overall graphics performance. :)

Another Bust for Phenom Tye?

Bust? I think mines doing quite well actually. ;)

However, there comes a time when you have to move on... => :p:

Hey, you have done a great service for everybody and you should be commended for such a fine job Tye:clap: :clap: :clap: :up:

I will stick with my original statement and say that Phenom Currently is not suited for Overclock as I watch this data being posted regularly and that what I see from a collective view point of whats been posted from start to current.

It can only get better and I am looking forward to one myself maybe 4GB of 1066 SPD Ram if JDEC gets the Hooks out of their Ass:D that and Windows XP64 should prove interesting:yepp:

Quote:

---

Originally Posted by KTE

Are you sure about this Achim? Currents are separately regulated and limited. I'm not sure how NB VIDs/Volts affect CPU VIDs/Volts you know, they are totally separate and independent of each other, because the IMC has it's own separate voltage from the cores, decided initially by the NB VID; on the 5 phase MSI RD790, CPU gets 4 phase PWM regulation with a maximum of 110A, so 27.5A in each phase and the IMC gets 1 separate phase for 20A maximum. Both have separate VRMs controlling their supply voltages, and both attain independent voltages. This is why that fella burnt up his K9A2 CF PWM area, it can't handle 110A+20A through the VRMs.
So, if I test 2.3G CPU/1.8G IMC at 1.2V/1.1VID and it fails, moving it to 1.4V/1.1VID or 1.4V/1.05VID doesn't get me any more stability at all, it still fails. ;)

---

Yep i'm sure but that formular is only valid for the next lower cpu vid if you found one you can run stable with a really low nb vid. Can be a temperature issue. The power consumptions is the same with my power meter between those two alternatives.
Formular should be u=u1*11/13+u2*2/13, had 100A in mind.
Does your mobo also use the L6740L chip?