Phenom 9850BE low voltage overclocking results!

[Kei]

I know a lot of people will probably hate me here right now as they haven't quite hit 3Ghz stable on some of their setups. Well to add more to the hatred I decided that until we get the new bios for our ASUS boards I won't be able to boot into windows at speeds higher than 3.1Ghz (passes post all the way up to 3.4Ghz on as little as 1.35v).

Anyway I proceeded to finsh my normal overclocking routine and now focus on how low I can get the voltage stable at high speeds. Well....................

I'll post screenshots later but the lowest I can boot is 1.20 volts at 3Ghz...yes you read that right 1.fukin20 volts! It wasn't stable at that speed but I had to get shots for proof, same with 1.21v booted with no problems but failed 100% testing rather quickly. So I decided I was hungry and it was time to get some food...dialed in 1.232v and went to get me a cheese steak. Well 1 hour 47 minutes later it was still chugging away just fine.

100% load on 1.232v without failing, no bsod, nothing...it's still running right now and I'm talking to you at that setting now.

Okay now edited to include screenshots.

K

Btw...cheese steaks are good

1.20v Successful 3Ghz Bootup

1.232v Everest 3Ghz Stress Test Results

 

[SirKeldon]

Woh, nice first tests ... i read about the issue on M3A32-MVP Deluxe (on ASUS Forums) and trying to clock this CPU higher than 3Ghz, i really hope this is solved soon just with a BIOS update, keep us in touch cause i really want to purchase a Phenom to get all the juice this mobo could give ... and i'm just waiting for the first results of 3.5Ghz with it, i'm wishing you really good luck!!!

And btw, really nice overclock at that voltage, can you post some shots of the temps too? If you're doing that with air, i'm trying to realize what i'll be able to do with water if the issue is solved =)

 

[a111087]

lol, thats quite nice and strange at the same time
I guess overtime Phenom will get better

 

[Kei]

Woh, nice first tests ... i read about the issue on M3A32-MVP Deluxe (on ASUS Forums) and trying to clock this CPU higher than 3Ghz, i really hope this is solved soon just with a BIOS update, keep us in touch cause i really want to purchase a Phenom to get all the juice this mobo could give ... and i'm just waiting for the first results of 3.5Ghz with it, i'm wishing you really good luck!!!

And btw, really nice overclock at that voltage, can you post some shots of the temps too? If you're doing that with air, i'm trying to realize what i'll be able to do with water if the issue is solved =)

Thanks, I've been testing on this since the day it came out and have been at 3Ghz since about 5-6 hours after having it. I did it at stock voltage originally and then did all the rest of the testing on the Northbridge, HT Link, and ram afterwards. I've had less than ZERO problems at 3Ghz since that day. It's simply perfect really.

I can post all the way up to 3.4Ghz on 1.35v but can't get into windows (though I tried 1.5v the other night and it actually tried to boot) yet due to the bios issues you've been hearing about.

Still no issues with 1.232v I'll post some screenshots a bit later on tonight (for real) of the test results and cpu-z shots.

As far as cooling I'm a 'die hard' air cooling guy so yep this is all done on air. But I'm using a ridiculous cooler (aka the best ) in the Xigmatek HDT-S1283 on low settings (never bothered testing on high...ever). Temps during the last stress test (1.232v) were max 46C, min 39C, and average of 44C. My room at the time was very warm compared to normal 23C normally it's around 16-17C so it'd be much cooler. And remember the fan is on low even though it's huge.

My temps at 1.30v (stock) are still awesome with this cooler max of 48C and load at about 38-39C. The reason the min temps during that last test were so high was cuz I was using it for a while before the test and doing tests with other things so I never let it get back to idle temps.

Also note the temps would be lower but I have CPU Tweak enabled which raises temps due to voltage tweaks that are done to make the processor even faster. On average in my testing this accounts for approximately 5-7C higher temperatures on both the X4 9500 and the X4 9850BE.

I may try for a 10 hour run when I sleep tonight (I'll just do something else for 2-3 hours while it runs) to see if it'll pass a full boar super stress test though after 1 hour for most intents and purposes that's as stable as it gets. I'm crazy though so I like to run torture tests to see if it'll pass everything completely.

K

 

[Kei]

Btw, I don't know if you've seen but 3.5Ghz has already been done on this chip though I haven't checked the status lately to see whether or not it's 100% stable yet. I know that 3.2Ghz has been done stable already though and obviously 3Ghz is more than perfectly 100% stable. Your mileage may vary as no two pieces are alike. Mine has tested at multiple 3Ghz configurations and not had a hiccup once.

K

 

[Psychlone]

Awesome!

Hey Brother...what's the stepping on that 9850 again? I know I read it somewhere, but now I can't find it.

My JAAFB AA 0810BPAW can't do what your's is doing as far as GHz and voltage - I've been able to crank out 2.9GHz at 1.3V (haven't tried anything lower as of yet, but now you've got my wheels crankin'!)

So far, (and I keep *meticulous* notes of what's worked and what doesn't, along with the math for *every* test) - 2.92GHz at 1.3V (215 X 13.5), HT at 1935, NB at 2135(? can't remember for sure now and my chicken-scratch math is hard to decipher sometimes!), and my RAM at 573MHz (1146MHz @ 5-7-7-25 2T and 2.2V)

But, after dicking around for hours, testing, rebooting, clearing the CMOS (I've actually superglued a toothpick to the CMOS jumper to facilitate the clearing process!) - I've found that 2.8GHz on the CPU with the RAM at 535MHz, 5-5-5-15 2T 2.1V and HT @ 2000, NB @ 2400MHz, I've got faster throughput (as per SuperPi, MaxPi, Everest and Sandra) than anything else I've come across as of yet.

I'm still tweaking though...I've got a long way to go before every latency is adjusted, but I've found the groove - it's just a matter of fine-tuning at this point...AND THEN YOU COME ALONG AND SAY YOU CAN DO 3GHz AT 1.23V just to throw me for an infinite loop! Now you *know* I have to test everything I've already done with lower volts, right???
(Good job!)

Psychlone

 

[jbunch07]

its nice to know these things oc well...ill be ocing mine soon...what kinda temps are u gettin at 3ghz

 

[jkatt12]

nice speed Kei! hope i can get up to 3ghz. i got the new m3n-ht deluxe 780a board put in with the 9850 BE, i should be able to get up to your speed shouldnt i? dang, been messing with the bios configs and still cant get it past 2.7 ghz. how do i get the 5,200 ht speed as they claim this board and chip can do?

 

[Psychlone]

The HT speed that you're seeing is only 1/2 of the total bandwidth you're using (kind of like DDR terms...)

So, 2000MHz really equals 4000MHz, etc. - and I've already proven on my M3A32-MVP, that it can reach 2600MHz (5200MHz total) - it's just not completely stable there with all my other overclocked settings (awaiting a BIOS update with bated breath!!!)

Psychlone

 

[Kei]

AMD's most efficient Dual, Tri, and Quad core?

Okay after doing some more stability testing I decided to try out multicore testing again (did it on stock 1.30v a while back) but to see what we can do to save power and still keep speed.

For those on a board that doesn't quite fit the 125w chip bill then you just may be able to still get the efficiency of the Phenom lineup and not have to really compromise by buying a 'lesser' model (they're all excellent but you know what I mean). You can still possibly buy the highest available and just change the 'architecture' so to speak and make it fit your board for the time being until you get another one. Of course that is ONLY if you have the option to downcore in your bios.

Also for those that just don't really need the 4 cores at this time but have one anyway you could also save yourself some heat, power, AND still upgrade at the sametime w/out having to buy one of the older X2 models and get something more efficient and 'future proof' at the sametime.

Anyway I decided to get a preview of the Kuma and Toliman line of processors early last night and what do you know...when you do downcore they actually have the names already on the chip as recognized by Everest Ultimate. Crazy huh?

If you are looking for a Dual-core Kuma in a Phenom style flavor then just power down 2 cores in the bios and there you go with one side effect...UBER POWER EFFICIENCY!!!!!! I ran stability testing last night on the Kuma (Phenom X2) setup last night at a ridiculously low 1.21v and it ran like a champ. I even decided to up the craziness and use the newest version of Prime for even further stress. Well 2 hours later it was still chugging along just fine. 3Ghz Dual Core on an INSANE 1.21v and more efficient than any other AMD Dual Core chip out there at 3Ghz and MORE than a match for the X2 6400. I'll run some Everest full benchmarks against the X2 6400 really soon (just built a system for my friend with that, plus Everest already has the stock numbers for one anyway) to see how they stack up. Using my Diamond 3870 (918/2646Mhz) and the Kuma setup I scored 11,294 in 3dmark06. Of course if I was using the old 8800GT I had the score would have been nearly 12,000 which is very good. You can play Crysis just peachy with this as I had no problems all day nor night (I was up all night testing last night). Temps on the Kuma X2 setup are sick at 40C under 100% load and only 32-34C at idle. Room temps were the same as earlier (~22-23C) and CPU Tweak was enabled as well as AutoXpress which generally causes about 5-7C higher temperatures when enabled (the CPU Tweak I mean, AutoXpress didn't really seem to have an big effect on cpu temps).

Now then onto the Tri-core Toliman setup, I haven't run this through Prime just yet but it passes Everest on 1.232-.25v which is just as good. If you've never used Everest Ultimate you should, awesome logging abilities and info, and sweet lord can you stress a system hardcore. Anyway just like with the Kuma X2 setup you get all the sweet memory performance of the Agena X4 setup due to the architecture being the same. As for multitasking you'll more than likely NEVER notice the difference between X3 and X4 (and in many instances even X2 to X4). Temps ranged approximately 36C idle at 1.25v (what I'm on right now), and under 100% approximately 41-42C which is still extremely low. CPU Tweak and AutoXpress are also being used so temps would be far lower still. I also ran 3dmark06 on this setup and scores 12,407 with the same Diamond 3870 (918/2646Mhz).

I'm currently running on the Toliman X3 setup right now to see how it fares over the course of time for now. Since the bios will allow multiple setups (2 saved plus the current setting) you could have all 3 setups if you wished and switch between them with ease. This can save you plenty of power, heat, etc. on your system and is better than CnQ in my opinion because you get the voltage benefits though not quite as much but you still get the speed.

These results compare VERY well to overclocked and stock AMD 64 X2 Dual Core model chips like the 6000+ and 6400+. This is especially true when you think I'm using a less powerful graphics card than I was previously so the scores are noticably lower. I just pulled up AMDGeeks.net and grabbed a few results for comparison sake only.

X2 6400 3.5Ghz 3dmark06 run (water cooled cpu 1.50v w/8800 GTS)
http://service.futuremark.com/compare?3dm06=5080755

X2 6000 3.2Ghz SuperPi run (air cooled cpu 1.38v)
http://www.amdgeeks.net/overclock/706

I couldn't find any of them stock (it's not that kinda site) but that's still a good comparison as both of them are overclocked and at a higher speed yet neither is siginificantly (if at all) faster than a 3Ghz Kuma/Phenom X2 setup with slower ram. This if course is attributed to the fact that the Northbridge and HT Link can be both faster and the architecture is far more efficient as well. The 3.2Ghz X2 6000 SuperPi run is only as fast as the 3Ghz Kuma/Phenom X2 setup using ram clocked at 960Mhz (vs. 1068Mhz on the X2 6000 setup) which is very impressive.

The 3dmark06 test against a monster 3.5Ghz X2 6400 setup is only 123 marks higher despite a huge clock advantage and a more powerful video card! That is rather amazing if you ask me.

Okay famliy discussion time...

K

Btw, I don't really need to say please keep this a positive conversation as Intel doesn't neccessarily need to be mentioned in here as this isn't one of those conversations. Very strictly an AMD efficiency conversation. Thanks

 

[Kei]

Deeper Testing with Kuma X2 platform

Okay as promised I did some further testing with the Kuma/Phenom X2 platform to see how it compares to the Athlon 64 X2 6000 and 6400 chips. In an attempt to get the numbers a bit closer I changed the Kuma X2 settings to the following...

Kuma/Phenom X2 Platform
3Ghz (15*200 instead of 240*12.5 3Ghz setting)
2.4Ghz Northbridge (same as previous test up from 2.0Ghz stock)
2.0Ghz HT Link (4.0Ghz effective, stock setting up from lower 1.92Ghz of previous test)
800Mhz RAM (4-4-4-12 2T down from 960Mhz 5-5-5-15 2T)
CPU 1.216v

Athlon 64 X2 6400 Platform
3.2Ghz (16*200 stock setting)
1.0Ghz HT Link (2.0Ghz effective, stock setting)
800Mhz RAM (4-4-4-11 1T)
CPU 1.35v

The settings are all pretty close with both platforms having certain advantages that I can't change right now (I'd like to test the Kuma with 1T ram but it just won't run stable at those timings without large volts). Either way both platforms are very impressive in their own respect, but due to the new Phenom architecture you can adjust more which will give you greater/same speed at lower clocks and voltage.

Now then for the EVEREST Ultimate results...
----------------------------------------------
X2 6400-----------Kuma X2

Memory Read
8891 MB/s--------8375 MB/s

Memory Write
8855 MB/s--------6018 MB/s

Memory Copy
7245 MB/s--------8402 MB/s

Memory Latency
47.2ns------------54.5ns

CPU Queen
11,185------------11,115

CPU PhotoWorx
7,898-------------9,606

CPU ZLib
38,065------------40,100

CPU AES
8,292-------------10,081

FPU Julia
2,595-------------3,674

FPU Mandel
1,540-------------2,628

FPU SinJulia
1,429-------------1,313

I was shocked at some of those results but not by others. The latency thing is pretty obvious who was gonna win that one with the higher clock speed and CR1 timing vs the Kuma's lower clock and CR2 timing (I'll see about getting some CR1 able ram again). Write speed was another area I was not surprised to see the Kuma lose for the same reasons.

Out of the 7 CPU tests I was shocked to see the Kuma win 5 of them and by large margins as well! The X2 6400 eeked the Queen test by only 70 points which was very close and surprising. I was also surprised to see the Kuma lose the SinJulia test by 116 points a very good victory for the X2 6400. The rest of the tests though....wow...just wow Kuma is a beast!

I also ran a SuperPi run at this speed and scored an excellent 25.896s (tied with that 3.2Ghz X2 6000+ from the previous post http://www.amdgeeks.net/overclock/level2/706_2.jpg?nocache=120925425298393300).

Kuma/Phenom X2 SuperPi

Kuma/Phenom X2 3dmark 06

That 3dmark run lines up extremely well with the 1.50v 3.5Ghz run of the X2 6400 from the previous post (http://service.futuremark.com/resultComparison.action?compareResultId=5080755&compareResultType=14). It is now only 42 points lower than the X2 6400 3.5Ghz run! I can only imagine what score I could've achieved if I was still using the old 8800GT I had...not to mention if I put another 3870 in here (maybe Tuesday)!

This just keeps getting better and better...I wonder what the Toliman/Phenom X3 platform would do against the Athlon X2 6400 platform. Maybe that'll be my next set of tests...

Well then your turn guys, discuss...anyone else able to get close to what I can voltage wise (using 2,3, or 4 cores)?

K

Btw, I'm now officially scared of what the Phenom FX chips will be like.

Edit: All of the testing results were done with ganged mode enabled so the numbers would be higher if unganged was used. They were never retested as such.

 

[Kei]

More Kuma X2 testing

Okay so I started testing it with the memory timings worked over this time to see how close I could get to that 3.5Ghz X2 6400 run and the 3.2Ghz X2 6000 SuperPi run. Well...I beat them both.

Amazing how large the difference when you tweak the memory timings and settings just a bit. That's 26 marks above the X2 6400 3dmark06 run, and .307s better than the X2 6000 on the SuperPi run.

Kuma/Phenom X2 3dmark06

Kuma/Phenom X2 SuperPi

I ran the Everest benchmarks again to see how (and what) things changed after the tweaks.

Kuma/Phenom X2 Everest Results

When I did the full report for Everest the results came back as follows...

Memory Read
9,203 MB/s (up from 8,375 MB/s)

Memory Write
6,245 MB/s (up from 6,018 MB/s)

Memory Copy
9,444 MB/s (up from 8,402 MB/s)

Memory Latency
48.0ns (down from 54.5ns)

CPU Photoworx
10,847 (up from 9,606)

The rest of the tests were the same as before within a few points. The jumps that were made where HUGE jumps though and the latency is now withing striking range of the X2 6400 and I'm still on CR2 timings! I'd love to see what the latency and others would be if I was using CR1 timings. The voltages are still the same as before. I raised the Northbridge to 2.5Ghz (up from 2.4Ghz) and then adjusted the memory timings and settings (what made the biggest change), the ram is also now running at 1000Mhz again (up from 960Mhz), HT Link is still at 2.0Ghz.

I wonder what this would do if I raised it to 3.2Ghz................

K

 

[SirKeldon]

I wish you could go till 3.5Ghz as some XS guy did ... i'd love to see the results and benchmarks cause as i'm watching ... the benefits to migrate to a Phenom are not so significant if i get stuck on 3Ghz, cause not all the chips will do 3.2-3.5Ghz even if you have watercooling ... anyway the results are pretty good and they're building a balanced road AMD must follow, i don't know if they will release another series of Phenom's before the 45nm jumps to street ... but if they don't and the clock results of the ppl are getting stuck on 3 and can't go further (3.5-3.7 will be an acceptable clock for me) i think i should maintain my 6400+ BE @ 3.54Ghz 100% stable (3.58-3.6 for benching), at least, for now.

I'm adding two screenshots @ 3.54Ghz, SuperPi and Everest benchmark:

More results Everest Benchmarks @ 3.54Ghz

CPU Queen - 12294
CPU Photoworxx - 30907
CPU Zlib - 41985
CPU AES - 9147
FPU Julia - 5992
FPU Mandel - 3370
FPU SinJulia - 9363

 

[Kei]

Those are some EXCELLENT scores you've got! I don't have anything ram wise that I can turn up to a higher speed with having the other settings still clocked to what they are. I may pick up some 1066 ram (from GEIL of course) and that'll let me clock even higher without having to sacrifice any of the other clocks just to get the ram higher (highest I've went by itself is 1224Mhz).

I wonder how close I can get to your scores if I add some faster ram to the system. I know I won't be able to get your exact numbers due to the fact I'm using 3Ghz and you're far higher at 3.54Ghz but I should be able to come close. The everest numbers I believe I can beat (read and copy speeds), and maybe some of the CPU tests as well...that is a hard bit to bite off though because you're clocked higher than the stock 3.2Ghz model I used for comparison.

Also most people with your processor don't usually get above 3.4Ghz (and some can't get that stable) so I guess we're both in the same boat. Once the bios things get fixed I'm extremely confident I can boot to higher than 3Ghz (I'm not limited to that now that's just what I use...I've done 3.125Ghz on low volts too).

Thanks for the reply and especially for the comparison numbers (your clocked higher than the 3.5Ghz 6400 from the test and have faster ram too). I see you have a 2900XT...would you mind running 3dmark06 for a comparison as well? I know you're on 2 cards and I'm not asking you to take 1 out to make it more even (that'd be awesome though) but I'd love to see what score you get in comparison. Maybe I'll pick up a 2nd 3870 in a few days and then we'll have an even better comparison. I wish I knew what was truthfully going on with the 4xxx series cards so I could decide on buying another 3870 card or just wait. Meh...I'll probably just buy another 3870 instead anyway cuz I don't wanna try to sell this card as I love it.

K

 

[SirKeldon]

Yups, RAM bandwidth and latency means it all on AMD systems as far as i saw, so i think if you were able to run some 3.2-3.3 tests with a RAM speed close to 1150-1200 you'll own my results pretty hard and will convince me to get a 9850

Btw, i wish i had 2 x 2900 XT's, they're 2 x 2600 Pro which are far far away from the performance of just one 2900 XT ... anyway, you should go XFire for sure (maybe wait for the 4 series though) ... with one card at 3,57Ghz and HD2600Pro at 729/522 the best result were ~3.8k, with the crossfire and both cards clocked at same speed i hit about 6.9k (1280x1024 - Optimal) ... as i'm not a hard gamer that's a pretty result for me

 

[Kei]

I really wish the bios issues would get dealt with soon as I can EASILY post all the way up to 3.4Ghz (3.2Ghz on stock 1.30v). It'd be nice to see what I can do at 3.2Ghz without upping the voltage to a high number (3.4Ghz will start loading windows at 1.50-.55v).

The other guys over at XS (I've been following that thread since it was born and with the 9500 I owned) are using different boards than we are so they can get other things we can't, then again other things we have they can't get. They're using pretty low voltage for those speeds as well (3.4Ghz at 1.45v).

I'm almost 99% certain that I'll just go with a second 3870 as those have matured far more than the not even released 4xxx cards and I'll probably have less problems with them. Also I remember hearing something about the 3xxx cards not being able to Xfire with the 4xxx cards which would suck but I'd understand that (you can't Xfire 2xxx with 3xxx either from what I remember).

3870 in Xfire is extremely good anyway and more than enough for pretty much any game out now either way. Even Crysis runs very well on Xfire 3870 cards depending on processor and other settings. Now I have to decide on whether I really want matching cards or go with the Visiontek card for the 2nd card due to the following...

1. I can't use 2 rebates from Diamond so I won't get another $15 back
2. The Diamond card with overnight express shipping is $193.14
3. The Visiontek is available at BestBuy for $199
4. They're the same card minus the VIVO not being on the Visiontek (according to newegg)
5. The Diamond card will probably clock to insane speeds as mine does w/o hardware mods
6. The Visiontek card may not clock the same w/o hardware mods...

I ALWAYS ALWAYS ALWAYS use overnight express shipping (no matter the cost) so I have to weigh that into my pros and cons which makes the price difference between the 2 cards null and void now I'd say. The only reason I'd go with the Diamond card now is for better peace of mind in believing it will clock the same as the one I already have, and to have 2 matching cards. The Visiontek is (my original choice over the Diamond before I bought it) my usual ATi graphics card company and it's easier to get and RETURN if anything is wrong or I don't like it....decisions decisions

K

 

[cdawall]

just by the way that puts the chip at only 158w TDP which if AMD has some luck could ewasily be done on the boards ok with 125w TDP

 

[jonmcc33]

You all call those overclocks?

 

[jbunch07]

You all call those overclocks?

considering the voltage he is using hell yea i do
besides this is 9850be not intel

 

[Bytor]

You all call those overclocks?

They always seem to chime in sooner or later eh!!!

It's like they have radar for AMD posts..


Kei, Thanks for all the help through our PM's.. I'm still running great at 2.860 @ 1.26v...

 

[hat]

You all call those overclocks?

WOW. A whole 200MHz more. Oh, and you're at 1.4v! He's at 1.2. On a Quad. Owned?

 

[Kei]

You made my evening

You all call those overclocks?

I finally got a good laugh tonight thanks to you...I'll just pretend you're joking cuz seriously....did you even read your own cpu-z data? Not to mention you're using 1.4+volts compared to my 1.21v and my clock is damn near the same as yours.

Thanks though.

K

 

[Kei]

WOW. A whole 200MHz more. Oh, and you're at 1.4v! He's at 1.2. On a Quad. Owned?

Actually...you made me laugh even more.

I take back my thank you statement to him cuz you joke was funnier.

K

 

[Kei]

Btw, jonmcc33 do you have anything to contribute to the post or just being the Intel newbie of the day? j/k

Got any screenshots of your 3dmark06 runs to add to the thread?

K

 

[jbunch07]

Actually...you made me laugh even more.

I take back my thank you statement to him cuz you joke was funnier.

K

ill have to agree with u!
i lol when i heard that!
thanks hat

and thanks to u Kei for these post! ill be ocing mine soon so if i need to ill come to u for some help!