955 overclocks/performs better @ DDR 2 or 3 ?

[NapalmV5]

now that 955s been out for a while.. what you guys experiencing ?

ive achieved higher clocks/performance/stability @ ddr2 1000 than @ ddr3 2000 on the q9650.. just wondering if 955 is any different

thanks for any input

[Oese]

dont think so..

[informal]

now that 955s been out for a while.. what you guys experiencing ?

ive achieved higher clocks/performance/stability @ ddr2 1000 than @ ddr3 2000 on the q9650.. just wondering if 955 is any different

thanks for any input

I think chew*(among others) did some comparo runs with ddr2/3 on 955. It seems ddr3 has some wall on ~1800MHz on current IMC rev. of Denebs. It isn't much of an issue since performance all around is better than with ddr2,especially when you ramp up the northbridge/L3 clocks to ~2.6Ghz(and upward) to avoid the bottleneck there.

So to recap,my opinion is that 955 with same NB/L3 clocks and DDR3 @ 1600Mhz cas 7 is faster some ~5% in real world apps than same 955 with DDR2 @ 1066Mhz cas5.

[Oese]

ok to be precise.. i dont think that nb/core oc is any different (at max some mhz but thats due to different overall setups/boards/stressing of the components).

ddr3 oc of course is limited by the imc now, but 800mhz cl6 is achievable with good ram and thats really fast even compared to ddr2-1200 cl5 (as my own testing goes) what you wont get stable with most rams.

so either go with your beloved am2+ mobo and ddr2 ram if you have some (i think you dont though ) and enjoy the higher oc margin the 955 has, or go with new am3 mobo and enjoy ddr3 speeds the cpu itself will do good on both platforms and ddr3 will make it even faster..

[NapalmV5]

i see.. thanks guys.. i guess id have to get both ud5p and see what max oc/performance i would get on both platforms..

again.. thanks for your input/time

[Particle]

Personally, I've seen better performance with DDR3. While both are at 3.2GHz, a 940 with DDR2-1066 and a 955 with DDR3-1600, the DDR2 system completes Cinebench R10 in 1m15s while the DDR3 system completes in 1m5s. Of course, these are on different mobos, but there's a potential advantage there.

[chew*]

ddr III is faster, you have to sit and tweak for hours to get 1200 5-5-5 and 1333 6-6-6 will still beat it......which took me about 10 minutes to acchieve prime stable with....

[NapalmV5]

Personally, I've seen better performance with DDR3. While both are at 3.2GHz, a 940 with DDR2-1066 and a 955 with DDR3-1600, the DDR2 system completes Cinebench R10 in 1m15s while the DDR3 system completes in 1m5s. Of course, these are on different mobos, but there's a potential advantage there.

yes.. default clocks perform better @ ddr3 both intel and now amd

but that changes the higher the clocks go.. basically the ddr3 parts just crap out while the ddr2 parts seem to have no limit

ddr III is faster, you have to sit and tweak for hours to get 1200 5-5-5 and 1333 6-6-6 will still beat it......which took me about 10 minutes to acchieve prime stable with....

what about max oc ? which offers more performance.. ddr 2 or 3 ?

same has been said ever since ddr3 came out about intel cpus.. and i agree.. but when i max oc time and time again i get higher clocks+performance on ddr2

in other words.. im interested what the 955s stamina is.. ddr2/3

[Tony]

an apples to apples comparison. New tech Vs old tech...DDR2 Vs DDR3 same CPU, same ram speed same timings (except TRAS)

CPU 955 @ check CPU-Z
Memory DDR2 1055Blade @1200 5-5-5-18 2T (1t was set in bios)
memory DDR3 1333Animal @1200 5-5-5-24 1T
motherboard DFI 790FX M2R
motherboard MSI 790FX GD70
SSD on both systems OS=WinXP32
Watercoling by OCZ flat base bowed blocks.

Test, encode DVD FILM 720/480 to Xvid 1280/960 1500MB file size mp3 160VBR audio with Auto GK...

Results: DFI



Results: MSI



Conclusion...MSI DG70 wins by 7mins and 52secs

Pretty conclusive to me running apples to apples GD70/DDR3 wins.Overall test time includes MP3 conversion etc, actual frame rate conversion are very similar.

I will now retest with MSI over 1333 6-6-6-24

[Tony]

Slight tweak to get close to same CPU CPU-Nb and HT speed with 1333+ ram.

This gains a little more speed:

[Particle]

Nice results, Tony. Looks like even at the high end, DDR3 is maintaining superiority.

[Tony]

test continues 1600+MHz 7-7-7-24 1T MSI



Still gaining...so pi is better with low ram speed tight timings...encoding is gaining with ram speed 7-7-7-24 timings. i think going to 8-8-8-24 will kill performance though.

Trying to get 1700+ 7-7-7-24 stable for another test, again CPU speed etc is real close

This test was 42 secs faster over the 1333+ test

[Tony]

4-4-4-15 just shy of 1000, stopped here to match CPU speed.




so 4-4-4-15 1000 is faster than 1200 5-5-5-18...

[iandh]

It's funny, with my DDR3 system, "stability" seems to be kind of a gray area. For instance, I can test dead stable in OCCT and memtest, but if I tighten up my timings too much it will actually hurt desktop performance... this sounds strange but the best judge I've found so far for seeing how good my system is actually running is watching how the tray apps start up. If I have a good stable OC they all snap open instantly, if not they kind of ramble one after another. This is all with tested stable settings too.

My DDR2 platform isn't as picky with my 720BE as the DDR3 system is, but I find that the DDR3 system does seem to have just a hair more oomph when dialed in just right.

IMO there is actually quite a bit of gray area in OC'ing. I've had sky high OC's that test rock stable and post higher synthetic numbers, but perform worse than lower/looser OC's in real world gaming. I've found overclocking to be very application specific.

[Tom128]

That is some impressive DDR2 ram there Tony. Just curious though (for my own knowledge), with the ram at ~600Mh are you really gaining anything by having the NB at 2.7Ghz? I think I remember someone (you even?) mentioning that once you go much more than twice the effective speed (in this case DDR2-1200*2= 2.4Ghz NB) you don't gain much. Did I misunderstand that?

[aGeoM]

Hi Tony

So for conclusion, DDR2 4-4-4-15 2T@1000 are slower by 50s compared with DDR3 7-7-7-24 1T@1600; 8s slower than DDR3 6-6-6-24 1T@1333; and 16s faster than DDR3 5-5-5-24 1T@1200.

I got it right?

Nice findings, thanks.

Be well...

[Tony]

I am trying to stay in the zone 2600>2800 so the comparison is pretty valid.

Here is the DFI running 1000 4-4-4-15 with slightly higher CPU clock and its still slower than the MSI... 15secs is 15secs and I had a slight advantage on the CPU clock on the DFI.



Full day tweaking here on and off, each test takes 1hr and it HAS to run 10 loops Linpack before i even think of running AutoGK...i need stable.

Forgot to say..1.93V for 4-4-4-15 1000 stable

[aGeoM]

Sorry, but can't see the picture, yes, testing is a PITA, time consuming wise.

Edit: Seeing it now.

[Tony]

A quick conclusion for DDR3 AM3 systems would be tight timings still matter, push as high with tight timings as you can and test this against slightly loser timings higher ram speed. I have no doubt that 6-6-6-24 1t is the sweet spot for timings on Am3 and 1500ish MHZ may be the zone where you need to be (if the ram will got that high) Reason being some of the boards are not doing 1700+ very easy...some even struggle at 1650mhz ram

For DDR2 boards...bragging you have 1200MHZ ram 5-5-5-15 means you are NOT a clever overclocker IMO...you do better tweaking NB clock and if you can staying with 4-4-4-15 timings around 900 to 1000MHZ.

DDr3 Vs DDr2...DDR3 does win, on the systems I tested anyway and i did NOT go to great lengths to dial in timings, just the primary and tRC were adjusted so end users can at least feel they have a chance of reproducing what i did.

AM3 DDR3 is the new king...Low voltage DDr3 is the way forward with 1.65Vish max...no need to ram volts anymore.

[aGeoM]

Yeah, my first step toward DDR3 system was to buy the 955 chip, but I'll gonna wait for RD890, if I don't sell any of the 4 DDR2 boards first.

I like the MSI board, but Bios is hard to mod, and SPI chip soldered to PCB doesn't help.

[Tony]

core8 bios tools and an spi flasher works wonders

I can just as easy ask MSI for tweaks though, they listen now.

[gOtVoltage]

core8 bios tools and an spi flasher works wonders

I can just as easy ask MSI for tweaks though, they listen now.

Dang Tony your the Bomb....nice work and yes MSI is on Top of the whole Oc ballgame again

[roofsniper]

tony what sticks are you using? and why ganged on the ddr2 but unganged on ddr3?

[afireinside]

tony what sticks are you using? and why ganged on the ddr2 but unganged on ddr3?

Was just about the ask the ganged question. I was under the impression that unganged was superior?

I really wish I wanted for DDR3

[Tony]

unganged on the dfi seems to make less difference...I still have boards hooked up so i can revisit

Also setting 1T still shows 2T, bios may need more work