Could B3 Phenom be the rebirth of AMD....+1000MHZ OC done!

[wiak]

What we need is a G0 phenom

next revision can be a G hehe :p

[Particle]

Does having different speeds on cores actually help? Cause for example Crossfire, which I know is completely different though, it doesnt matter what card is faster as the speed being really used is from the slowest right?

If it does help I DEMAND NOW a BIOS which supports it cause I dont feel like getting into AOD every time. It's nice to do a few quick changes while testing but for me that's it. Unless they get everything working perfectly, but AOD has been out for a while and it still doesnt work perfectly so that might take a while

Simple answer: Yes*

I'm a programmer, so I appreciate some of the aspects about multithreading that a lot of people do not. So, why the asterisk? It's because the positive reply depends a lot upon the code being ran. Workloads that don't have to be synchronized can be parallelized rather easily and utilize any number of dissimilar cores to get the job done. Unfortunately, most real-world things must be either synchronized or in-order to be useful. Sure, you can encrypt 10GB of data using multiple threads and just write the data from one thread's output stream to disk as soon as it is done but good luck decrypting it. If you want to use tons of RAM, you could buffer the output in RAM either piecing it together during or after everything is processed, but in practice this is a bad methodology. However, most highly-threaded processing is done on "frames" of data, essentially a small chunk that isn't difficult to keep in memory. If well implemented, this allows even vastly different speeds of cores to work together to finish a common task that requires a serial output without using a terrible amount of memory in the process. If poorly implemented, you might be stuck waiting on the slowest core to finish at the end of each frame. Graphics problems like SLI and Crossfire are different in that the output is highly time-critical. To deliver frames that are consistently spaced, your faster card is frequently going to be waiting on your slower card to output so it can as well. Otherwise you'll end up with a jerky sense of action even if the framerate is modestly high if the cards are each just told "puke out frames as fast as you can and we'll write both of your outputs to the monitor". Yes, there are methods of doing multiple cards that don't entail frame interleave, but this makes for a good example of some of the complexity of the problem without being difficult to understand.

But that's just a view from a sugar-coated technical side. In a more consumerist way, it is safe to assume that yes--it will be faster/better than keeping them all at a similar, lower frequency. No, you won't always notice it. No, not everything will benefit.

Max them out.

[stocius]

Simple answer: Yes*

I'm a programmer, so I appreciate some of the aspects about multithreading that a lot of people do not. So, why the asterisk? It's because the positive reply depends a lot upon the code being ran. Workloads that don't have to be synchronized can be parallelized rather easily and utilize any number of dissimilar cores to get the job done. Unfortunately, most real-world things must be either synchronized or in-order to be useful. Sure, you can encrypt 10GB of data using multiple threads and just write the data from one thread's output stream to disk as soon as it is done but good luck decrypting it. If you want to use tons of RAM, you could buffer the output in RAM either piecing it together during or after everything is processed, but in practice this is a bad methodology. However, most highly-threaded processing is done on "frames" of data, essentially a small chunk that isn't difficult to keep in memory. If well implemented, this allows even vastly different speeds of cores to work together to finish a common task that requires a serial output without using a terrible amount of memory in the process. If poorly implemented, you might be stuck waiting on the slowest core to finish at the end of each frame. Graphics problems like SLI and Crossfire are different in that the output is highly time-critical. To deliver frames that are consistently spaced, your faster card is frequently going to be waiting on your slower card to output so it can as well. Otherwise you'll end up with a jerky sense of action even if the framerate is modestly high if the cards are each just told "puke out frames as fast as you can and we'll write both of your outputs to the monitor".

But that's just a view from a sugar-coated technical side. In a more consumerist way, it is safe to assume that yes--it will be faster/better than keeping them all at a similar, lower frequency. No, you won't always notice it. No, not everything will benefit.

Max them out.

Thanks for the answer. I wondered that myself.

[Suosaaski]

Heh. It doesn't work that way. It uses all four (three of which it uses quite a bit). You can't jumble the math any way you want it and say something like "If you add it up, you can see that it only uses one core at 270%." I'd also like to point out that three cores stay around 75% and the forth at 25% which averages a bit closer to three fully loaded cores instead of two if you do still want to look at it that way.

Yeah well, it all depends on how much one process stays on one core if one can do the math like that or not. For example, on most dual core setups when running an application of only single thread, the two cores usually (not always) average around 50% each.

And, my estimate what not entirely accurate, I just pointed out that it might not take use of all cores. Nevertheless it uses them nicely.

FSX is sometimes still a bit slow to respond, but after SP1 and SP2 it has improved a lot. I don't play it that often though.

[w00dzy]

What Prime test did you use? Small FFT, Large FFT or Blend?

blend.. is that the best one?

[Ben_Brown]

I was playing around with clocking the different cores to different speeds. The image below explains it all.



I have not tried to make it stable for anything other than benchmarks. May work on that this weekend. Just thought I would share.

[Particle]

Counter-Strike Source isn't quite as impressively threaded as UT3 and FSX:




---


On the subject of different core clocks, I've been mapping out my cores individually for the last few days. I've got one that only does 2.8GHz stable no matter the voltage, but the others can go higher. Three at least do 2.9GHz and maybe higher at 1.34V.

[SkullCracka]

ben your scores dont match on your 3dmark scores ????

strange.

[Trice]

ben your scores dont match on your 3dmark scores ????

strange.

lol oops

[CCUABIDExORxDIE]

ben your scores dont match on your 3dmark scores ????

strange.

no, they match up if you used the 3dmark calc and the scores of :

SM2.0 6373
HDR/SM3.0 8054
CPU 4180

i dunno if your not looking at the right scores and looking at the old score on the 3dmark06 big opening window itself or what.

[Rammsteiner]

Oh man.

Guys, AOD stability test does all actually, at least for me. I ran Prime and almost insta BSOD.

So all testing last two days, again wasted time.

[Brother Esau]

Oh man.

Guys, AOD stability test does all actually, at least for me. I ran Prime and almost insta BSOD.

So all testing last two days, again wasted time.

That AOD Stability test is crap atleast right now in the early stages I guess most of the App is crap aso no big surprise on that one Sorry Sammi I know you have put allot of time into this and it takes time to get to UGuru status but it needs allot of attention

[Rammsteiner]

Yeah,

I really like the idea of AOD, but it needs a lot of work!

Also, I think AMD should pay W1zzard a little, give him info about AMD cards and combine Atitool into AOD so you've just AOD to both OC PC and GPU's.

Completely bug free, using almost no recources etc... That's what I'd like to see. But ATM it's crap. Especially if you consider ~45 hours of testing has been useless AGAIN.

I stick with Prime Large FFT's again. AOD stability showed 1.2Vcore and 2.4Ghz NB stock Vnb was stable, but it was very far from stable. However, I won't give up. If CPU's won't do more than 2.7Ghz, ok, then Ill move onto the NB and try to get it superb high and try to get DDR1200 on RAM. Phenom IMC looks to handle RAM OC's, at least low latency wise, better than K8 IC's.

On my 6400 I couldnt get DDR800 to boot, I could with Phenom. Didnt test stability though, will do that later on after CPU and NB OC'ing is ready.

[Rammsteiner]

Im wondering, did anyone hit high RAM speeds yet with DDR1066 mode? Cause somehow my system does seem to like HTT more now. Im testing CPU stability right now, but if I would go from DDR800 to DDR1066 mode it should run on DDR1200. But Im wondering if anyone actually hit such speeds at all?

[Ben_Brown]

ben your scores dont match on your 3dmark scores ????

strange.

The other scores are from a previous run with Core0 running at 3100.

[Particle]

Running at various speeds is definitely worth investigation. I thought my chip was a total lemon since it maxed at 2.8GHz. Some cores are quite superb as I've found out, however. I'm still mapping where the cores top out, but this shows how a core might do a lot better than what the entire chip can max at when kept in sync.

[stocius]

Running at various speeds is definitely worth investigation. I thought my chip was a total lemon since it maxed at 2.8GHz. Some cores are quite superb as I've found out, however. I'm still mapping where the cores top out, but this shows how a core might do a lot better than what the entire chip can max at when kept in sync.

How is AMD overdrive working out for you? I hear it's still crap but I'd like another opinion

[Particle]

It's a bit quirky, but once you become familiar with it it's a very useful tool. It's really nice to know that if your OC settings don't work out, your board won't need cleared. It simply reboots and is using your old settings again.

Right now I have core0 at 3200MHz and the others at 2800MHz. I'm still trying to feel out the ceiling for core0 at 1.34V. I'm surprised it made the 30-minute prime run at 3100MHz even, considering how lack-luster my OC results were prior to running the cores async.

[leoy]

I think that if tony tried pushing the cores individually we would propably see 3.6-7 in such a good cpu...

[Rammsteiner]

Nice to see.

Although Ill only try different core speeds when it's put into BIOS. No AOD anymore for me for a while... crappy program.

Also I gave up on DDR1200+ RAM speed. I got DDR1136, but IMO it's not really worth it to be honest. The impossibility to get some normal timings instead of uber high ones ruines the fun for me.

Thus far it seems like my NB likes to run 2.4Ghz. Didnt test a very lot yet, but with 240Mhz HTT Ill run my RAM at 4-4-4, if possible 4-4-3/4-3-3. 3-3-3 is impossible though. From the DDR1000 tCL3 thread you see it requires around 3Vdimm to get that, so DDR960 would be a little less but still too much.

What's a good program to bench the real RAM bandwith with Phenom? Everest is showing crappy results. Im running unganged, but I want to see the real bandwith after all.

[Particle]

So far, I've been testing at 3.1x3.0x2.9x2.8 at 1.344V for an hour without a crash. I need an easy name for doing non-synchronous clock speeds. How about multiclocking?

[w00dzy]

looking good particle.. would like to see 3.2 on 2 cores and 2.8 on the others if you could get a 3dmark06 result with all at 2.8 and then with 2 at 3.2 that would be interesting....

good luck

[Particle]

3DMark06 ran at this resolution (1024x768 [monitor's max]) doesn't seem to benefit a lot from extra CPU.



Synchroclocked at 2.8GHz:





Multiclocked at 3100x3000x2900x2800:

[Seraphiel]

Counter-Strike Source isn't quite as impressively threaded as UT3 and FSX:

But both of those aren't near AC's level of multithreading. 40% to 70% usage on a C2Q @ 2667 MHz, just with Altair doing nothing:

http://img168.imageshack.us/my.php?image=acau6.jpg

Sorry for off topic.

[Particle]

But both of those aren't near AC's level of multithreading. 40% to 70% usage on a C2Q @ 2667 MHz, just with Altair doing nothing:

http://img168.imageshack.us/my.php?image=acau6.jpg

Sorry for off topic.

You might have missed the graph for FSX. It was reaching 100% a lot of times on a quad.