Phenom 960BE and M4A89TD

I am not sure this is the place for this but, I have got it set up just like in my sig. 3900 OC with ram at 800 mhz 7-7-7-20. You might think that it would be really fast at some benches but it just isn't. It scores lower then many of the AM2 Athalons with DDR2. It was really easy to get this OC, just a couple of settings in bios and she booted right up, so cudo's to the mobo but still it is blown away by many of the setups in the Phenom thread. In Cinebench it takes 110 seconds scoring only in the 12750 area, SuperPi 1m is high 17 or low 18's beaten by many other lower clocked CPU's and ram at only half the MHZ's. Granted I have not yet figured out the Ht and HTT settings as they both just are set to 2000 by the auto settings but increasing them has had very minimal results so far. I think that I must be missing something very basic or I wasted a lot of money on ram and CPU that I could hgave got for a lot less had I known that the results were so small. I am fine with the 3.9 OC because I beleive that any higher would push it to the temperature limit but I am thinking that it should perform a little better. Any advise would be helpfull.

Please stack the program windows on the desktop differently or downsize the pic before uploading so that we don't have to pan from side to side to see the info.

Also, please show us the "SPD" tab in CPU-z.

You can leave your ht link where it is since your stable and won't gain anything by raising or lowering it. You could try dropping the ram to 1333 and see how low you can get those timings. the phenom ii doesn't support 1600 because it requires an overclocked cpu/nb freq of 2400mhz or higher to take advantage of the extra bandwidth, so you could either drop the ram down or up the cpu/nb freq.

I found 2 sweet spots for my system using 3.9ghz. 1 is having my ram at 1333 with a 2400mhz cpu/nb freq the other is having my ram at 1600 with a 2600mhz cpu/nb. I prefer the 2400mhz setting because the temps don't increase much over the stock 2000 but at 2600mhz it takes a good boost in voltage to hold stable and pushes my temps to around 51c.

The biggest lack I see is the cpuNB speed - it's still running at 2000 MHz. If you can bump that to 2400 MHz, maybe even more, I'm sure it would help a lot. Both the L3 and IMC (memory controller) are on the cpuNB so even though your RAM is running 800 MHz you're not getting anywhere near that kind of use from it since the cpuNB is running so slow.

You are probably going to have some issues trying to do that, though. The cpuNB is on the processor and it could easily raise your load core temp when you increase it. Since you're already past the usual stable OC temp of 55°C you may be forced to decrease something else.

Odd that your temps aren't better with that heatsink. How's your case airflow?


PS
A little trick I learned from IMOG a few months ago. When you attach a pic you can go up to the top (where the B I U) buttons are and click on that paper clip symbol. This will show your attachments and, if you click on those attachments they'll be embedded in the post where the forum will re-size them for you ...

Thanks guys, raising the cpu-nb must be what is wrong. I weill have to try that, I have raised it before to 2400 and it did help some of the benches but I dedn't run that many to see a big difference. I will go ahead and give it a shot and see what happens, and thank for the paper clip tip. Here is another screen shot with the spd in it.

Well I tried lowering ram to 1333 and tightining timings.
Raised the CPU-NB to 2600,2400 and a few other things.
It did run a little hotter but only changed benches slightly.
Set it back to original settings except cpu-nb at 2200, at least it is stable.

I'm not sure what scores you're looking at but these are the SPi 1M times from our Bench Team thread. ChanceCoats is a member and this is his score with his 720 at just over 4 GHz, I've added some others for comparison as well.

4.ChanceCoats123 - 17.000 - Phenom II x3 720 @ 4050Mhz - DDR2@506 5-5-5-16-20 (Air)
5.wanner - 18.002 - Phenom II X4 955BE @ 3850Mhz - [email protected] 7-8-7-20. (Air)
6.reddeathdrinker - 18.782 - Phenom X6 1055T @ 3808MHz - 725.3MHz/9.0-9-9 24, Air cooled.

Not sure I see a problem with your scores?


Cinebench is too card specific for my taste when trying to compare CPUs ...

{PGA}AfterShock said:

You can leave your ht link where it is since your stable and won't gain anything by raising or lowering it.

Click to expand...

So there isn't any performance increase in games raising the ht link?

Bakercake said:

So there isn't any performance increase in games raising the ht link?

Click to expand...

Not unless you're running a pair of high-end cards there isn't. From the tests I've seen even 1800 MHz covers most single card installations and that can drop to 14-1600 for older cards like the 8800 or 9800. On FX chipsets (38 PCIe lanes plus 4 command lanes, IIRC) the HT Link can move a HUGE amount of data ...

QuietIce said:

I'm not sure what scores you're looking at but these are the SPi 1M times from our Bench Team thread. ChanceCoats is a member and this is his score with his 720 at just over 4 GHz, I've added some others for comparison as well.

4.ChanceCoats123 - 17.000 - Phenom II x3 720 @ 4050Mhz - DDR2@506 5-5-5-16-20 (Air)
5.wanner - 18.002 - Phenom II X4 955BE @ 3850Mhz - [email protected] 7-8-7-20. (Air)
6.reddeathdrinker - 18.782 - Phenom X6 1055T @ 3808MHz - 725.3MHz/9.0-9-9 24, Air cooled.

Not sure I see a problem with your scores?


Cinebench is too card specific for my taste when trying to compare CPUs ...

Click to expand...

Perhaps you are right, I just thought that 1600 would be faster then 1333. It appears that the platform utilizes all ram by timings and not speed. It would seem that 1333 ram with the same timings scores the same as 1600 speed ram. Actually I can score a little higher by lowering speed and tightning timings and the examples that you posted bear that out as my timings fall between 4 and 5. I guess I just had the wrong idea that a higher clocked ram would equal some gain in performance. My only conclusion is that tighter timings mean everything and (ram) MHZ means nothing on this platform.

That's true to a certain point but if the speed of the ram slows down too much the CPU becomes starved for bandwidth. If the bandwidth need is being satisfied then latencies become more important.

shadowdr said:

Perhaps you are right, I just thought that 1600 would be faster then 1333. It appears that the platform utilizes all ram by timings and not speed. It would seem that 1333 ram with the same timings scores the same as 1600 speed ram. Actually I can score a little higher by lowering speed and tightning timings and the examples that you posted bear that out as my timings fall between 4 and 5. I guess I just had the wrong idea that a higher clocked ram would equal some gain in performance. My only conclusion is that tighter timings mean everything and (ram) MHZ means nothing on this platform.

Click to expand...

We try to tell people that over and over and it just doesn't sink in until it's right in front of their face. People get so caught up in speed, speed, speed that they don't stop to consider the other half of the RAM equation - delay of signal. Since most RAM operations are a lot of small chunks instead of fewer large chunks that delay is a killer for most programs. It's the difference between downloading 10x10 Mb files compared to 1x100 Mb file. If your Internet connection had a comparable delay to RAM latency compared to speed (and I hope it doesn't!) then that delay time would become vital with the 10x10 Mb transfers.

As far as I know it's not a matter of platform, just simple physics. I'd bet that even with Intel (though I could be wrong) 1333CL7 is darn near the same as 1600CL9 with 1600CL7 being faster than both, just as it is on AMDs if the cpuNB is running fast enough.

trents said:

That's true to a certain point but if the speed of the ram slows down too much the CPU becomes starved for bandwidth. If the bandwidth need is being satisfied then latencies become more important.

Click to expand...

+1

QuietIce said:

We try to tell people that over and over and it just doesn't sink in until it's right in front of their face. People get so caught up in speed, speed, speed that they don't stop to consider the other half of the RAM equation - delay of signal. Since most RAM operations are a lot of small chunks instead of fewer large chunks that delay is a killer for most programs. It's the difference between downloading 10x10 Mb files compared to 1x100 Mb file. If your Internet connection had a comparable delay to RAM latency compared to speed (and I hope it doesn't!) then that delay time would become vital with the 10x10 Mb transfers.

As far as I know it's not a matter of platform, just simple physics. I'd bet that even with Intel (though I could be wrong) 1333CL7 is darn near the same as 1600CL9 with 1600CL7 being faster than both, just as it is on AMDs if the cpuNB is running fast enough.

+1

Click to expand...

Well I haven't been around much for a while and having just left ddr 3200 it would seem that it is indeed a different ballgame. I did a lot less research before this upgrade then any that I have done before and the memory section has never been one of my favorite forums, so I am a little behind.
The thing that really puzzles me is that 1600CL7 is not faster the 1333CL7, it is the same and CL9 at any speed is slower then CL7 at any speed. Again your example above bears this out because the fastest time is CL5 DDR2 @506 while the DDR3 CL7 is slower @733. I am curios if there is a fastest timing that can be acheived before the latency is to short for the full memoy cycle. Kinda like the way PC3200 @7-3-3- was as fast as it could get and still still allow all data to be fully cycled. Guess I have some reading to do.

The thing that really puzzles me is that 1600CL7 is not faster the 1333CL7 ...

Click to expand...

If you were running the stock 2000 MHz cpuNB speed when you tested it then this wouldn't surprise me at all. cpuNB speed is very critical to RAM performance. Crank up the cpuNB to 24-2600 MHz and it should make a difference ...