Dropping the CPU Multiplier overclocks the 975X Chipset [Credit to OCZ Tony and FCG]

[Super Nade]

Dropping the CPU Multiplier overclocks the 975X Chipset [Credit to OCZ Tony and FCG]

As a follow up to the good work by OCZ Tony and FreeCableGuy, here are a few results for the 975X chipset.
​

The gist of the matter is:

Actual Bus Speed = (Default Multi/User Set Multi) x User set FSB

So, Sandra shows the actual bus speed under the mainboard info screen. Note a few discrepancies under Sandra, where under the CPU screen, it shows CPU Freq as 3.2GHz. There are other ways to resolve this and I'm working on a few tests to decipher this.

More to come.

TESTBED:

• ASUS P5W-DH DLX BIOS 1403 beta
• Allendale E6400
• 2 x 36Gb raptors
• OCZ Gold PC5300 2 x 1 Gb
• MSI x1800xl 512Mb (modded to an xt)
• Supermicro PWS-0056 650W PSU

CPUz does not show the actual Bus speed. For the record, I cannot post if I increase the FSB by 5 MHz. This is probably because the actual FSB is so high.

 

[eva2000]

ah so it is true for 975x as well!

 

[Super Nade]

These are scores under Default Multi at daily working frequencies. It has nothing to do with the tests I'm running, but it serves as a benchmark to see how various numbers scale.

 

[Super Nade]

Well, since this is going to take a while, let me outline the questions I have in mind (Answers will be provided in the last post):

• Does dropping the Multi overclock the chipset?

First hints are provided by Sandra (not fully reliable because of the discrepencies I pointed to in the first post) and by the fact that I cannot POST at say 440 MHz FSB.
​

• Does this entail an automatic change in internal latencies of the chipset?

Quite possibly so (referring to what Tony said about PAT/Turbo mode in this case). The way to sniff this out would be indirectly, i.e by benchmarking the system with different FSB's and Multi's. I may have to fully saturate the busses to get my results, but I'm not sure how to go about it, yet.

​

• What is the most efficient configuration for Core2Duo overclocking?

Answering the above two questions will answer this one.

​

• Real world tests with Games (FPS and MMPOG's), Scientific computing (scilab) and WinRar.

 

[Burdman27911]

Looks like the start of something great... going to keep my eyes on it!

 

[MadMan007]

Good to see your own thread about this topic here

At the beginning of the Conroe craze everyone saw the 975x was limited to ~400FSB. Then the 965 came and people saw much higher FSB. OCZ Tony and FCG found out about the strap settings and how they work on the 965, apparently these help make possible the high FSB. Then I saw reviews of the Asus P5W64 WS Pro...975x but the same higher FSB range as the 965 boards so it's not just based on the chipset. My theory is Asus has enabled the 965 strap adjustment scheme on the P5W64, is that possible?

 

[Super Nade]

I believe every ASUS Intel board since the 8xx has had the AUTO strap adjustment feature. However, I believe the Workstation boards have loose chipset timings to begin with, hence they are able to scale higher. Running bus intensive benchmarks should reveal these differences. At this point I am not sure if there is a significant difference between the 965 and 975 chipsets apart from the latencies (in the context of performance, not architecture).

A quick update#
My system stutters and sputters when I'm at 400 x 6, whereas this completely disappears when I hit 8 x 300. A sure-shot indicator of bus issues is the much maligned wireless card on my P5W-DH-Deluxe. The moment the bus freq goes far out of spec i.e the 534 MHz regimn, the internet connectivity goes to hell. I know it is not my router, because I am checking connectivity with a laptop.

 

[Super Nade]

The ss for the update in the above post. Note that Set FSB=400 and performance mode is Turbo.

 

[Super Nade]

Here are the results for the 6x Multi 401MHz user set FSB. The lower scores clearly indicate something is different. This confirms Tony's prediction that at 400MHz FSB, the bus latency is lower, leading to increased bench scores (THOSE WHICH ARE BUS DEPENDENT) but massive instability.

 

[iamjcl]

Don't know if this applies here, but I always found that Sandra didn't respond if changing FSB within Windows (SetFSB / ClockGen, etc...).

Over the last several boards (955X, 2 X 975Xs) I found this to be the case.

I know at least the memory bandwidth would remain essentially unchanged over wide changes in FSB within Windows.

I would have to re-boot and set the FSB I wanted to test in BIOS for Sandra to "notice" it - then I would see big changes in Sandra results.

I may have not had the most current version of Sandra, and I may be missing something, but thats the way I remember it.

 

[Ross]

Two things: first, the latency change with different straps is not a theory that needs to be proved yet again...AFAIK, it's a fact and has been proven time and again. Your time would probably be better spent just finding where the "sweet spot" is for the chipset. Keep in mind, it will only take 1 BIOS update to change something in the chipset though

Second, like iamjcl says, don't waste your time running benches in Sandra for proving results if you are going to use SetFSB, Clockgen or any other Windows app to alter clocks after booting...they will be inconclusive to say the least. If you can't boot at a clock, don't do it.

Boot at 565 FSB and clockgen down to 510FSB:



Look what happens when I boot at 510FSB instead of booting high FSB and clocking down:



In short, don't use anything to alter the FSB/clocks after you boot or Sandra is going to give you some potentially WAY off results.

 

[Epox4life]

Thanks for that info Ross. Those subtimings are tight.

 

[Super Nade]

I'm Old skool. I have eschewed clockgen for quite a while. All values are solely set in the BIOS. Note that I'm using the beta 1403, where I can change the multi.

 

[Ross]

No problem S_N, just wanted to make sure you (and everyone) is aware there are potentially very different results using Sandra and any kind of desktop OC I think it bases all measurements compared to boot settings only...pretty sure there's no way I have 109% BW efficiency

Epox, yeah, this mem/NB combo is still a bit weird to me. I can't get some secondaries as tight as others I've seen (the most important ones of course), but can get others tighter. The weird thing is, if I set them too slow, it will error out just as fast as if they were too tight...go figure

 

[Gautam]

I'm keeping my eyes solely on the memory bandwidth from Sandra, because nothing else that it reports is worth looking at. I'm sure all those FSB numbers are bogus.

From looking at the memory bandwidth alone, I don't see anything inconsistent, and I don't even see bandwidth changing to any significant degree between 400MHz and 401MHz. Or any of the other benchmark results for that matter. Correct me if I'm wrong.

Ross that looks fuggin weird. But whats this about booting at 565MHz?? I'm guessing you're changing multis right? I know that CPU is good, but no way it can be good enough to pull off 565x9...right??

Either way my head is spinning.

 

[Super Nade]

I'm keeping my eyes solely on the memory bandwidth from Sandra, because nothing else that it reports is worth looking at. I'm sure all those FSB numbers are bogus.

Quite possible. But how come no other program reports any discrepancy?

From looking at the memory bandwidth alone, I don't see anything inconsistent, and I don't even see bandwidth changing to any significant degree between 400MHz and 401MHz. Or any of the other benchmark results for that matter. Correct me if I'm wrong.

The first clue I go was the huge system instability at 400 MHz. My wireless just kept crapping out and the system got really sluggish. Everything was smooth as silk on the 401MHz try. I made three attempts with 6 x 400MHz and every time the rig was sluggish and potentially unstable (I'm pretty sure of it).

Either way my head is spinning.

Likewise

 

[Ross]

Ross that looks fuggin weird. But whats this about booting at 565MHz?? I'm guessing you're changing multis right? I know that CPU is good, but no way it can be good enough to pull off 565x9...right??

Maybe 9x565 on DI? Yes, boot at 8x565 and then drop to 9x510 on the desktop to make the comparison...8x510 is about 50MB/s less than 9x510. I didn't need to do it to make my point, but 97xx looks better than 96xx

There is DEFINITELY an FSB spot where mem BW tanks for me...and that's ~510FSB. It drops over that and even at 570FSB isn't as much as it is at 510FSB. I would actually need to go back and double check, but IIRC, from 510FSB and up, BW continues to decrease as FSB increases. I am not sure if it's something in the chipset, if it's heat or some kind of inefficiency from the high MHz, but it drops just a hair from 510->515FSB and after 515FSB it drops like 50-70MB/s and just gets worse.

 

[Gautam]

I don't know, but Sandra has always sputtered BS like that with every CPU I've had, which is why I've never even fired it up for the past two years. It just calculates the values it displays, doesn't actually read em off the PLL. If you want to know what speeds you're really running, Clockgen SHOULD pin them perfectly, though the 80MHz PCI-E that it displays is still a head scratcher.

But the only way to truly prove anything like this is to show a real performance difference, not a stability difference or reported speed differences. Right now your Sandra benchmarks all look equal to me. My favorite though is SuperPI 8M...should usually take 3-5 minutes and is a very accurate gauge of whats going on.

 

[Ross]

Duh, disregard what I said about the BW drop...just realized that was using SetFSB, LOL. It drops "relative" BW when raising FSB on the desktop just like lowering FSB with it increases it