Phenom 9500 w/ MSI K9A2 Platinum

Quote:

---

Originally Posted by Oldguy932

Here's what I remember getting last time I was running benches

Everest Cache and Memory

L1: 1.0ns
L2: 3.0ns
L3: 6.5ns
Memory: 57.4ns

---

Your NB speed is 9x looking at your RAM latency, it should be below 48ns if NB was actually plus 2.4G but your L2 cache is too low, lower than real 2.91G Phenom gets. 2.91G gets 3.3s which means you're at plus 3GHz CPU speeds as that's all that effects L2 latency. ;)

What RAM MHz/timings does CPUZ show at those settings?

Quote:

---

Originally Posted by Oldguy932

And I probably won't be able to hit 3.5ghz, my htt maxed at 300 when I tested it on this cpu the other day. That would only give me 3450 at most, which would be amazing for a phenom, but even just 2.9 at 1.3v is giving me 50-51C load with small fft's, meaning around 56-57C true max.

---

Cold limits are many times different ;)

Quote:

---

Well, I guess I'm one of the few that can play with the nb multi.

---

You see, so could everyone else I've seen incl. myself and look at tictac's post above mine. It's just that Phenom 9500/9600 did not actually change NB multi plus 9x. Only CPUZ would show it higher speeds than 9x and real software would show you the speeds.

BUT... but... divide the NB speed AOD shows by the HT ref. you set in BIOS at bootup and I bet it shows the NB multi at 9x. ;) ;) :p:


BE does change NB multi and shows NB speed having big difference on perf and you need high NB volts for that.
*If you want to find out and see for yourself, like I said, just bootup at 266x11.5 3.059G at 9x NB multi and your scores will be like you're getting above, probably even faster.
-Then set to 239x11.5 2.749G and 11x NB multi in BIOS and compare 1M/wpime with your above "2749" scores. It will be a lot slower.
-Then set 217x11.5 2.495G and 11x NB multi in BIOS and compare with above "2530" scores. They'll also be a lot slower.

Not because you didn't oc your NB, you did here too but because your NB was always at 9x and your CPU speed was much much lower in these cases than the shots you've already posted.

You're not setting 2.7G/2.5G in BIOS, you're setting 266x11.5, that's why CPUZ is reading 266x11.5 3.059G.
If you set BIOS at 2.7G and 2.5G, I'd like to see that.

TRUE NB multi moved up above 9x is shown by one clear way: Sandra mem/1M/32M/wprime/RightMark MTMT and most importantly, you can validate the CPUZ (you can't otherwise) and you'll also be able to move HT up to those speeds too. Like this:

Attachment 72123


If no NB multi change above 9x, AOD does not allow higher than 9x HT multi on any Phenom.

Easy tests to prove NB/CPU speeds:
Run Cinbench 10 and POV-Ray 3.7 BETA at (BIOS values):
228HT ref. x 11.5 CPU multi (2622MHz) and 9x NB multi (2052MHz) ->
THEN
Again run at 228HT ref. x 11.5 CPU multi (2622MHz) and max NB multi (if 12x then 2736MHz) ->

Post back the images and scores.
Show both, AOD and both CPUZ CPU/MEM tabs in the image.

:)

Quote:

---

Originally Posted by tictac

This is Phenom 9500 (max nb multi = 9x) with 5.6GHz NB speed... it boot up ... Biostar TA770.... i guess it is not working.. because if it work it wont post at 5.6GHz NB speed....

http://www.lejabeach.com/pics4/TF770A2+020908-2.JPG

but benchmark show improvement over slower NB speed.... i am still wondering why... :confused:

---

Hehe.. that's crazy. What benchmarks have you tested? Can you run the above two tests for me at 9x NB and at your max NB multi?

Quote:

---

Originally Posted by Oldguy932

What's it showing for your cpu speed though?

Edit: I just started trying 267*11.5. @1.3v it failed prime on the first core in 6 seconds, @1.325v it failed in 55s, and @ 1.35v the last two cores don't even start. I just realized I had processor to nb voltage left on auto, set it to 1.3 and I'm gonna see what happens next. Oh and if this setting doesn't work I'll at least get a screenie of 3ghz stable with 1.3v.

---

That is polygon aka rebelheaven rig... you can refer his post here...
http://www.rebelshavenforum.com/sis-...;t=000463;p=23

I think he's just facing the strange issues I dealt with 3 months ago. :D

I can't tell the CPU/NB speeds from those ss though. AOD and CPUZ ss side by side would clear it all out for me. :)

You're not listening. Here is the complete breakdown.

HTT=HTT set in bios
NBM=NB Multi
NB=NB Speed
MEM=Memory speed
CPU=CPU Multi
CPUS=CPU Speed

NB=9*HTT
MEM=HTT*divider
CPUS=(NB/NBM)*CPU

I know the actual NB speed cannot go above 9*HTT on a regular Phenom. What happens on my board is that it calculates the NB speed before the cpu speed, and because it lets me change the NB multi above 9, the htt the cpu is based off of is different than what is set in the bios. I'll make some tables in excel in the next couple days or so to show you what I mean.

yeah ... please explain it more detail oldguy... maybe add some screenshot. Graph comparison table would be nice too. Thanks :)

I just need a list of benches I should run. I tried doing a table with my 9500, but its htt limit was alot lower than this 9600, so I didn't bother.

Here's what I was thinking:

Super-Pi 1m
Pifast
wPrime 32m and 1024m
Everest benches, all of them
3DMark06, Overall Score and cpu fps
PCMark05 if you guys want it

I can't do game benches cause my most recent game is CNC 3, which I don't have a benchmark for, unless I did a timedemo and used fraps. Its what happens when you become an hwbot addict...

Benches I requested above please mainly POV-Ray 3.7 BETA(multi-cpu)/Cinebench10 (single-multi)/wPrime 32M/Superpi 1M/32M/Sandra Memory Bandwidth if you may and better are the first 3. Enough to show differences. I already know what you're mentioning because I said similar things when I thought NB MHz were high and it all proved incorrect once you learn and understand what is happening. Simple POV-Ray and Cinebench 10 runs with the settings I gave (easy settings for your CPU) will show your CPU/RAM/NB speeds quite clearly. ;)

And yeah, I did this 3 months ago. A comparison of approx. 1980MHz NB Vs 2860MHz NB at the same CPU/RAM speeds with the 9500. So I'm looking forward to what you'll present as I can re-run those exact settings with the BE pretty easily.

Also, I don't think you should take one software speed reading over the other when you don't have official word on such matters and when you have no way to verify any unless you bench a known setting in a known software which scales with CPU/NB/RAM and someone can cross-check the actual speeds by running the same settings.

It is extremely painful to run these at stock, so extremely slow in wPrime. I'm going to try and do comparisons from 8x to 12x, although I dunno how well 12x is going to run, it probably won't hit 266 like the others. That and what would be the easiest way to post the tables, cause a screenie ain't gonna cut it for this. I'm not doing Pov Ray or Sandra, but I do have cinebench. I'll do screens of wPrime with the htt at 266 for each nb setting.

Edit: Ok, wPrime was ok, single core Cinebench is forever.

For any of you folks who were worried about the power section of your board getting too hot, I think I found a pretty elegant solution...

I went to MicroCenter today looking for some heatsinks for the inductors, but couldn't find anything that would do the job. So I cruised over to CompUSA's going out of business sale and found this:

http://www.provantage.com/antec-spot-cool~7ANTF037.htm
(I got it for $10 :) )

You just remove a mobo screw and put the hold down under it, then plug it into a fan header. It's actually very quiet on medium and still moves a good amount of air, heres a shot of it installed.... (click for larger image)

http://i30.photobucket.com/albums/c3...lerClipped.jpg

Dave: Thanks mate. I use an 8000RPM 60mm fan only when I push the system as in oc/bench/stability test. Works fine like the above and it's fanmate controlled. ;)

Oldguy932: 266HTx11.5 comparison would be a completely flawed test methodology. Those are not the settings I stated to test and it'll leave you seeing the same picture you're seeing right now. Also wPrime 32M at 2.6G is not much slower than 3G, the difference is 3secs at most. Even 1M is +2x as slow and 3D is more than 30x as slow.

I'll make another table for that where the cpu is set the same but the nb are at their respectful speeds. I was just trying to show what happens when you overclock with the higher nb multis. And 32m isn't slow, 1024m is though.

Old Guy, I wish I could figure out what your doing... :D

Unless you know someone who cherry picks your chips at the fabs, it's kind of hard to believe you hit the CPU lottery twice, with the 9600 and the BE...

Regardless of possible wrong readings on monitoring software, those SuperPi & wPrime numbers absolutley impressive! I can't get anywhere near those... :(

With these settings:
http://i30.photobucket.com/albums/c3...0130AOD_25.jpg

I do 1M SuperPi @ 30.399s

And my wPrime is

http://i30.photobucket.com/albums/c3...210_wPrime.jpg

These are Very Stable settings though... Those weren't suicide shots were they? Even if they were, they aren't shots my chip could hit!! :D

Quote:

---

Originally Posted by KTE

Benches I requested above please mainly POV-Ray 3.7 BETA(multi-cpu)/Cinebench10 (single-multi)/wPrime 32M/Superpi 1M/32M/Sandra Memory Bandwidth if you may and better are the first 3. Enough to show differences. I already know what you're mentioning because I said similar things when I thought NB MHz were high and it all proved incorrect once you learn and understand what is happening. Simple POV-Ray and Cinebench 10 runs with the settings I gave (easy settings for your CPU) will show your CPU/RAM/NB speeds quite clearly. ;)

And yeah, I did this 3 months ago. A comparison of approx. 1980MHz NB Vs 2860MHz NB at the same CPU/RAM speeds with the 9500. So I'm looking forward to what you'll present as I can re-run those exact settings with the BE pretty easily.

Also, I don't think you should take one software speed reading over the other when you don't have official word on such matters and when you have no way to verify any unless you bench a known setting in a known software which scales with CPU/NB/RAM and someone can cross-check the actual speeds by running the same settings.

---

KTE... this is super pi 1m from Polygon Biostar TA770. NB at 5.4GHz :eek:

http://www.lejabeach.com/pics4/TF770A2+020908-3.JPG

Yeah 1024M is a very awfully long bench. Await the results and take your time to make it as accurate as possible please. ;)

Anyway, just a few factual notes:-

-Should see some v.good BIOSes with many fixes and available options coming out very soon, simplified oc... so get ready to test max oc again :D
-Same with AOD, should fix many issues and make oc'ing in Windows much easier and stable.
-Phenom 9600 BE is multiplier unlocked on CPU and NB, upwards and downwards.
-At >1.5V on the best air cooling, Phenom will not be actually idling <40C, truly. Ever. (Achim you better read this :p: )

MSI Asia: Happy New Year is now over, hope you enjoyed it. Now please just give us those BIOS options many of us have asked for many many times. I know you can do it.

I propose the following; Display in Cell Menu (in decimal digits) ->

Attachment 72133

All will finally allow people to overclock/run a Phenom fully to it's potential. Best options for a BIOS, people will flock to buy the board!

I know you're reading... :)

tictac: thank you to both of you. But that's showing a slower perf. than 8x NB :D

Quote:

---

Originally Posted by KTE

tictac: thank you to both of you. But that's showing a slower perf. than 8x NB :D

---

KTE this is what gotvoltage got from comparing NB speed

Quote:

---

CPU@2300mhz Phenom 9600 BE :

SuPi NB@2800mhz HT1800mhz = 33.235s
SuPi NB@5600mhz HT1800mhz = 33.203s

NB@2800mhz HT1800mhz
Crysis @1280x1024 LowQ = 72.1fps Average/9/loops

NB@5600mhz HT1800mhz
Crysis @1280x1024 LowQ = 75.35fps Average/9/loops

Just For kicks,,
CPU@2400mhz Phenom 9600 BE
SuPi 5600nb ht1800 = 32.109s
Crysis @1280x1024 LowQ = 78.30fps Average/9/loops

---

I hope your right KTE... ;)

Personally, I say screw the dividers, just give us CPU/NB Multis @ .25 intervals, Ex: 11.25, 11.5, 11.75 etc... That should be good enough resolution to dial things in pretty well. Especially on the NB...

I just tried setting my NB FID to 06 (10x) in P-States and it actually booted!
"latency.exe" showed rediculously low latencies (L3 of 39ns), but when I ran SuperPi it was almost 40s for a 1M calculation, AOD showed actual NB speed at the 9x multiplier though, and when I tried to up the HTT Ref clock it locked...

Here's to hoping MSI releases a decent BIOS soon, without the TLB patch FTW... :toast:

I'm a happy camper!! :rocker:

Best $10 investment I've made in a long time..., It's only 5Mhz on the HTT Ref clock, but it's enough to tell me that part of the board was way too hot!!
Maybe I should put the fan on high, it ain't that loud... :p:

http://i30.photobucket.com/albums/c3...h_0210_AOD.jpg

I'ts been good for about 4 hours, and I've been trying to break it.. ;)

tictac: these guys are getting badly confused over a small NB issue, they need to understand how different apps are working out values differently and they're only picking one application over the other to show false speeds which CPUZ won't verify because they're bugged. Problem is, I've ran higher NB speeds on 9500/9600 and 9600BE, only the last is where you can run it higher than 9x real and Super Pi 1M/32M is noticeably quicker with plus 9x NB multi than those times those guys are showing for 2300MHz. You can get better than those times at only 2.3GHz CPU, 1.8GHz NB, 400MHz RAM without any tweaks. Look here:

http://img91.imageshack.us/img91/6789/23001meq3.th.png

That's not even quick, you can beat it more with tweaks at 1,8G NB and you can get it even quicker @ 4-4-4-4 800 at higher NB multi. Meaning those times are just slow and that's why the guys can beat them at the same MHz. At 2.3GHz CPU and 3GHz real NB you should have around 31.5s in 1M and 6GHz should be far lower. But you might aswell ask AMD, I already know it, that you can't get 3/4/5/6GHz NB but false readings with no real change above 9x. CPUZ will show the HT ref and CPU/RAM speed.

This is the same as the people who had 4GHz and 8GHz Phenoms (at stock volts!) on ASUS M2N and Crosshair just because it had BIOS problems. :D

Daveburt714: Your wPrime time is slow because you didn't set it for 4 core usage. And no I don't get my cpu's cherry picked, my 9500 and 9600 regular I got from the local Microcenter and the BE I got from newegg.

I'm not so sure about those .25 multis, the memory speed already gets screwed up enough from the different multis on the cpu, but I would love to get half multis on the NB at least.

Currently the benches that are being run include:

Super-Pi 1m
wPrime 32m and 1024m
Everest, Cache Latencies and All Benchmarks
3DMark06, Overall Score and fps from the cpu tests
Cinebench, single and multi cpu scores

Tests are being done for NB multis 8x to 12x, HTT 200-260 in 10mhz increments. There will also be a table that I will try to get everything as close to 2.0ghz as possible to see the differences between the NB multis.

Sound good?

Edit: Tests for 200-240htt done for nb 8x

thanks oldguy.. waiting for your presentation.

I'm with KTE regarding NB speeds.

No way of running NB above 9x on locked Phenoms...

I've played with this for a week or so some time ago and you can re-read Gigabyte DQ-6 thread where my results are. Theoretically higher NB speeds was yielding minimally better results from test apps, but that might be just due to slightly different timings set somewhere in BIOS or skewed RTC for NB settings over 9x.

Sorry guys! Time to buy Phenom BE :yepp: (mine en route from warehouse :p: )

Finally... nb 8x table is done. Only 5 more tables to go...

YAY... 8x and 9x tables done now. Such a long process...oi... For those screenshots, what exactly did you want in them? And I took a look at CPU-Z for the first bootup at 10x nb, it shows cpu speed at 2.3, ht link at 1.8, and nb at 2.0. I'll post what I get on the next bootup as a comparison. I need a break, this crt is making me realize that I really need to get that lcd so my eyes hurt less...

Edit: Yeah, it locks the ht-link at the 1.8ghz divider according to cpu-z, but my nb multi does go up cause with 210 htt and 10x multi it gives me 2.1ghz nb speed. It does however show 11.5x210=2415 rather than the speed from my formula, although everything else, including everest cpuid has it based off my formula.

Edit 2: I thought it was a fluke, but the 10x so far seems the best for doing pi, its consistently better than 9x for 200 and 210 htt.

screenshot with 2 cpuz tab 1 showing cpu speed and 1 showing nb speed. plus final score from benchmark test.