Benchmarks with Asus P4P800 'PAT' On and Off

[prometheus]

Effects of Asus P4P800 Memory Acceleration Mode (PAT)

Asus has inplemented on the 865PE-based P4P800 what they claim is the equivalent of the Performance Acceleration Technology implemented in the 875p Canterwood chipset. In the P4P800 thsi feature is called Memory Acceleration Mode, and it can be Enabled on turned Off (by setting to Auto) in the BIOS. To determine the effect of MAM benchmarks were run under the exact same conditions with the same components. The only difference was wheter MAM was Enabled or Turned OFF.


3DMark2001SE - Asus P4P800 Standard 800FSB, 3.0C, DDR400 at 2,5,2,2-4 timings, Performance at 'Auto'(Turbo OFF), Video OC 441/752(376). DEFAULT, STOCK, best memory timings.

P4P800 MAM ENABLED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P4P800 MAM OFF

http://home.rochester.rr.com/hessfink/18706.jpghttp://home.rochester.rr.com/hessfink/17522X.jpg

Compare link for P4P800 Memory Acceleration Mode ENABLED at http://service.futuremark.com/compare?2k1=6552704 Compare link for P4P800 Memory Acceleration Mode OFF at http://service.futuremark.com/compare?2k1=6552890

[prometheus]

Sandra2003 MEMORY Buffered (Standard) - Asus P4P800 Standard 800FSB, 3.0C, DDR400 at 2,5,2,2-4 timings, Performance at 'Auto'(Turbo OFF), Video OC 441/752(376). DEFAULT, STOCK, best memory timings.

P4P800 MAM ENABLED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P4P800 MAM OFF

http://home.rochester.rr.com/hessfink/800b400.jpghttp://home.rochester.rr.com/hessfink/800b400X.jpg

Sandra2003 MEMORY UNBuffered - Asus P4P800 Standard 800FSB, 3.0C, DDR400 at 2,5,2,2-4 timings, Performance at 'Auto'(Turbo OFF), Video OC 441/752(376). DEFAULT, STOCK, best memory timings.

P4P800 MAM ENABLED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P4P800 MAM OFF

http://home.rochester.rr.com/hessfink/800u400.jpghttp://home.rochester.rr.com/hessfink/800u400X.jpg

[oldfart]

What I would like to know is.....

Does PAT or MAM work @ 5:4?

[prometheus]

Memory Acceleration Mode (PAT) at 5:4 Memory

While we can clearly see the significant improvement PAT/MAM makes in 1:1 or Syncnchronous memory, I have NOT seen those kinds of improvements - or any real improvement with PAT at 5:4 or Asynchronous memory. With the ability to turn on and turn off MAM/PAT, we should finally be able to test whether PAT is infact working with memory at 5:4


Sandra2003 MEMORY UNBuffered - ASYNCHRONOUS 5:4 Memory on Asus P4P800 980FSB, 3.0C, DDR392 at 2,5,2,2-8 timings, Performance at 'Auto'(Turbo OFF).


[b]P4P800 5:4 MAM ENABLED Buffered . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P4P800 5:4 MAM OFF Buffered

http://home.rochester.rr.com/hessfink/980b392.jpghttp://home.rochester.rr.com/hessfink/980b392X.jpg


P4P800 5:4 MAM ENABLED UNBuffered . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P4P800 MAM OFF UNBuffered

http://home.rochester.rr.com/hessfink/980u392.jpghttp://home.rochester.rr.com/hessfink/980u392X.jpg

It certainly appears that MAM/PAT IS NOT WORKING IN ASYNCHRONOUS MODE. The differences in benchmarks that we are seeing - 0 to 0.7% - are well without normal variations in Sandra benchmarks. I only say appear because while the differences are tiny with and without MAM/PAT, the MAM enabled scores are in all cases equal or slightly higher. Looking at these results I do think we can say with some confidence, that if PAT is working at 5;4, the Performance difference is negligable.

[oldfart]

Many overclockers will have to run a 5:4 or even 3:2. If so, the choice between 875/865 and the Asus with MAM or another 865/875 board is not that significant. If you run 1:1, it is.

[Tony]

Running 1:1 is the key,A 3C is a must and running 250fsb+ 1:1 will really fly.

[Ardy]

[It certainly appears that MAM/PAT IS NOT WORKING IN ASYNCHRONOUS MODE.

So...why does PAT only work in synchronous mode?Is it only working in synchronous mode in the 875 Canterwood??
That's supposedly why Granite bay boards weren't given asynchronous ratios.Because intel said the boards wouldn't perform as well......

[RotorHead]

So running a 3C with the higher multiplier will allow for a better OC and Higher memory performance at 1:1 with PAT on?

[Ardy]

I think it all depends on what your memory is capable of. Bigtoes answer is based on being able to run your memory at 500 .My Corsair wouldn't come close.

[oldfart]

Originally posted by bigtoe
Running 1:1 is the key,A 3C is a must and running 250fsb+ 1:1 will really fly.
Yeah, but for us regular folk, 3.0 GHz and <gulp> DDR500 are WAY out of the price range right now. My Corsair XMS 3200C2 is good for ~ 426 MHz. 1:1 is not a reality for me at this time. Maybe down the road when it is commonly available and prices are reasonable.

My 2.6C should be here tomorrow or Friday. I'll run it @ 2?? FSB and a 5:4 ratio.

[ErikaeanLogic]

great info, BT!

question for ya: have you located a sweet spot, the detente where the MAM/PAT no longer shows a difference? (for example, when I was using a P4m on my P4B533 I was unable to run a 4:3 mem ratio over 178fsb, found through trial and error.) I understand this would be a real PIA but I think that the information would be useful, so I will undertake this next week once me P4P800 arrives:); it would be interesting to witness either a decline in the difference in performance or a period, from 1 fsb setting to another, where it just stops showing a difference.


:cool:

[Tony]

What ever cpu you run with 875 or 865 you would be better running 1:1.Howecer 5:4 is still VERY good but you may find more vdimm is needed to run the same timings you can run at 1:1.

[ErikaeanLogic]

and while we're on the subject. . .;)

what RAM would you recommend for high-fsb on 865/875? I am tempted to run TwinMOS PC3700 (or PC3200 CH5 chips) but I've had to RMA 2 of 4 sticks I've received of their product in the past.

what's faster, 1:1 at conservative timings or 5:4 at aggressive timings?

[ThugsRook]

why does this just seem like Turbo mode (CMD 1T) to me?

or maybe its 1T mem write?

what is it?

[prometheus]

I think it is something more than just a 1T turbo mode, Thugs. The P4P800 ALSO has a Turbo mode selection in the Frquency voltage section that seems to do the same thing as setting aggressive manual mem timings.

[ThugsRook]

on a p4g8x "turbo" enables 1T cmd.
if i used wpcredit i could also enable a 1T write cmd.
all those settings together would look like this, cas 2-2-2-6-1-1

in sandra it would look like cas 2-2-2 (1T CMD)
in cpuz it would just look normal, cas 2-2-2-6

so ~ what is turbo doing and what is pat doing on springdale?

granitebay vs canterwood showed canterwood is slower @ 179fsb.
why?
it sounds to me like intel had problems and tuned the chip down a bit. now theyre getting better yields and all of a sudden its a feature?
so ~ is pat a feature, or something we already had/have on granite bay?!

:scratch:

[oldfart]

Only thing missing from the review is Canterwood @ 5:4. From what I see,

PAT 1:1 ~= MAM 1:1
MAM 5:4 ~= MAM disabled 5:4 (MAM doesn't work @ 5:4)
PAT 5:4 ???? MAM 5:4

Any chance that could be added?

[DAPUNISHER]

What Intel says about PAT
PAT technology CANNOT be disabled in a 875P based motherboard running at 800MHz system bus with dual DDR400 memory. However if memory population increases to 2 DIMMs per channel, PAT is designed to balance maximum performance with robust operating stability, and you won't see the maximum benefits that PAT can bring to a single DIMM per channel environment.


After some further inquiry we were provided with some additional information:

PAT is hardwired within the chipset and cannot be turned off by a motherboard vendor. Certainly a motherboard vendor can minimize its benefits by setting inefficient memory timings, but not sure why they would want to do that. There is no way to test within Windows to see if PAT is enabled within the chipset. However companies with the right tools can check PCI registers to see if the motherboard vendor has optimized their own memory timings. I think this is the basic confusion, motherboard vendors are setting memory configurations in certain ways and then claiming they are turning on or off PAT. This is incorrect, all they are doing is optimizing or unoptimizing their memory performance. Trying to relate these BIOS switches to turning PAT on or off is incorrect. Discuss among yourselves

:rotflmao:

[ThugsRook]

no wonder they named it PAT ~ noone can figure out what it is!!!

:look: :look: :look: