The 1:1 vs. 4:5 results thread

[n7]

So anyway, here's what we've learned so far from this thread:

On nF 680i:
1T is way better than anything 2T (unless it's insanely high clocked 2T)
When running 2T, 4:5 is indeed faster than 1:1 in the majority of situations.

On 975X:
4:5 > 1:1

On P965:
4:5 is far better than 1:1 in basically everything.

Obviously, CPU speed is generally the main factor (save for maybe when right around the P965's strap change area), so running 1:1 + higher CPU speed will nearly always be better than 4:5 w/ lower CPU speed, so keep that in mind as well.



Please feel free to add your results from your systems


n7's results

Specs:
E6300 @ 2807 MHz (7x401)
P5B Deluxe
Team Xtreem
X1900 XT stock

For these quick benches, i ran:
7x401 (2807 MHz) 1:1 DDR2-802 (4-4-3-4)
or
7x401 (2807 MHz) 1:1 DDR2-802 (3-3-3-4)
or
7x401 (2807 MHz) 4:5 DDR2-1002.5 (5-5-5-4)
or
7x401 (2807 MHz) 4:5 DDR2-1002.5 (4-4-4-4)

Results were averaged from a few runs

SuperPI 4M
1:1 (4-4-3-4) - 104.875s
1:1 (3-3-3-4) - 103.964s
4:5 (5-5-5-4) - 102.667s
4:5 (4-4-4-4) - 101.990s

3DMark01 SE game tests only
1:1 (4-4-3-4) - 37653
1:1 (3-3-3-4) - 38040
4:5 (5-5-5-4) - 38377.5
4:5 (4-4-4-4) - 38645.5

Q4 bench 800x600 high quality AA/AF off SMP enabled
1:1 (4-4-3-4) - 97.91 fps
1:1 (3-3-3-4) - 99.53 fps
4:5 (5-5-5-4) - 100.85 fps
4:5 (4-4-4-4) - 102.4 fps

WinRAR bench multithreaded
1:1 (4-4-3-4) - 1184 KB/s
1:1 (3-3-3-4) - 1225 KB/s
4:5 (5-5-5-4) - 1283 KB/s
4:5 (4-4-4-4) - 1326 KB/s

FEAR 800x600 0x AA 2x AF everything maxed
Min - Ave. - Max fps
1:1 (4-4-3-4) - 55 / 111.67 / 231.67
1:1 (3-3-3-4) - 55 / 111 / 229.33
4:5 (5-5-5-4) - didn't bother with. FEAR's results are too inconsistent for my liking
4:5 (4-4-4-4) - 55.33 / 112.67 / 233.33

UT2k4 Primeval 1024x768 no AA/AF
1:1 (4-4-3-4) - 91.965 fps
1:1 (3-3-3-4) - 92.975 fps
4:5 (5-5-5-4) - 93.691 fps
4:5 (4-4-4-4) - 94.635 fps

DVD Shrink Battlestar Galactica s1dvd1 transcoded (folder to folder) from DVD9 to DVD5 everything default settings
1:1 (3-3-3-4) - 356.33 seconds
4:5 (5-5-5-4) - 375.00 seconds
4:5 (4-4-4-4) - 366.33 seconds
I'm not so sure how accurate these are though, since even though i ran a couple/few runs for each, the results were somewhat all over the place.

DVD Rebuilder + CCE 5-pass Rome s1dvd1 DVD9 to DVD5 re-encoding
1:1 (3-3-3-4) - 103 minutes
4:5 (4-4-4-4) - 101 minutes
_________________________________________________________________________________________

cmdrdredd's results

Specs:
E6400 @ 3200 MHz (8x400)
P5B Deluxe
Crucial 10th anniversary
ATI X1900XT @655/775

same settings as N7 for each game. 1:1 = DDR2-800 4-4-4-8 and 4:5 = DDR2-1000 5-4-4-8

Quake4
1:1- 112.3fps
4:5- 116.7Fps

F.E.A.R
1:1- 62Min 129Avg 296Max
4:5- 63Min 129Avg 300Max

3DMark01
1:1- 42335
4:5- 43524

Winrar
1:1- 1369KB/s
4:5- 1540KB/s

SemperPi 4M
1:1- 1m 36.469s
4:5- 1m 32.953s
_________________________________________________________________________________________

Yoxxy's results

Specs:
X6800 @ 3200 MHz (8x400)
680i (eVGA or Asus Striker?)
Corsair 6400C3
8800 GTX/GTS?

3DMark01
1:1 (3-4-3-9) 2T - 62251
4:5 (4-4-4-8) 2T - 63232
_________________________________________________________________________________________

lopri's results

Specs:
Xeon 3060 @ 3600 MHz (9x400)
eVGA 680i
Team Xtreem
8800 GTX

SuperPi 8M
1:1 (3-3-3-8) 1T - 2m 53.922s
8:11 (4-4-4-9) 2T - 2m 57.672s
__________________________________________________________________________________________

Yellowbeard's results

Specs:
DFI 975 X/G, Rev 1.0, bios 12-01
E6300
2 or 4 x 1gb Corsair Dominator PC8500 w/Dominator Fan
EVGA 7800GT

Aquamark
400mhz FSB 1:1 112,544 (4,4,4,12)
400mhz FSB 4:5 112,928 (5,5,5,15)
430mhz FSB 1:1 114,557 (5,5,5,15)
430mhz FSB 1:1 114,998 (4 x 1gb Dominator PC8500 5,5,5,15)

3DMark 01
400mhz FSB 1:1 35,498
400mhz FSB 4:5 36,030
430mhz FSB 1:1 36,821
430mhz FSB 1:1 36,484

SuperPI 1M
400mhz FSB 1:1 20.250 sec
400mhz FSB 4:5 20.141 sec
430mhz FSB 1:1 19.031 sec
430mhz FSB 1:1 18.79 sec

Sandra Memory Bandwidth, Buffered

400mhz FSB 1:1 7373/5091
400mhz FSB 4:5 7612/5091
430mhz FSB 1:1 7864/5091
430mhz FSB 1:1 7999/5091

*Apparently there is an issue with this chipset and the latest Sandra reporting the Integer benchmark correctly as it never changed from 5091.

This is obviously by no means complete but, the results are consistant.

 

[Idontcare]

n7, can you hotlink your test cases to the benchmark software download site?

I want to contribute but want to make sure I use whatever you used. Just a request, if too much work then I will hunt them all down myself.

I haven't been in the benchmarking biz for some years now, still climbing back into the saddle.

 

[mattb2696]

what is the meaning of the fsbram ratio. i know its the ratio of the clocks, but what is the significance and what is ideal, what do you want to shoot for?

 

[The-Noid]

Cas 3 doesn't work on the P5B. If you run 4-4-3-4 will actually be faster then 3-3-3-4.

 

[The-Noid]

Also 1:1 is faster on 975x. On 965 4:5 is faster. Completly different for chipset performance.

 

[cmdrdredd]

Originally posted by: Yoxxy
Cas 3 doesn't work on the P5B. If you run 4-4-3-4 will actually be faster then 3-3-3-4.

Only at 1:1

If I run 400fsb and 3-3-3-8 at 1:1 it's slow but stock fsb at 4:5 3-3-3-8 is decent.

 

[Acanthus]

Uhh, any real world performance differences?

Im a lot more interested in performance at settings i would be actually using...

Id wager in most cases it will be less than 1% in real world apps that arent running at minimum settings.

 

[Idontcare]

For the record, are we arguing about whether trading off latency/timings versus bandwidth/MHz is beneficial one way or the other OR are we arguing about whether trading off linked 1:1 synchronization between FSB/MCH versus going asynchronous and unlinking the FSB:MCH ratios?

If this is a latency versus bandwidth argument then why go any farther? The end user will benefit from optimizing for either case depending on whether their application(s) benefit from latency (access) or bandwidth.

Same argument for stripe-size in a Raid-0 array, are your applications accessing huge swaths of data many KB's in size (graphics, video, etc) meaning bandwidth is needed or is your application executing on many many small bites (or bytes as it were) of data meaning latency will rule your day?

Personally I am more interested in knowing the more fundamental, non application specific, latency penalty associated with walking your memory bus off the synchronous plank and setting MCH speeds asynchronous to FSB speeds, as 4:5 ratios do. There is an overhead, what magntiude is the question in my mind.

Anyways, someone let me know what we are pissing on each other's shoes over in this thread so I can join the fun, thanks.

 

[n7]

Originally posted by: Yoxxy
Cas 3 doesn't work on the P5B. If you run 4-4-3-4 will actually be faster then 3-3-3-4.

For second i remembered those threads stating the same thing on XS...but then i decided to bench all the same stuff @ 4-4-3-4 liked you stated.

Guess what?

CAS 3 does actually work just fine on my P5B-D; you'll see in the results i am about to update the OP with.

That being said, i am gonna add 800ish 4-4-4 vs. 1000ish 5-5-5 results when i have time.

Originally posted by: Yoxxy
Also 1:1 is faster on 975x. On 965 4:5 is faster. Completly different for chipset performance.

Care to run some testing to back that up?
I think you're gonna be unpleasantly surprised as long as you keep things 2T vs. 2T

 

[n7]

Originally posted by: Idontcare
n7, can you hotlink your test cases to the benchmark software download site?

I want to contribute but want to make sure I use whatever you used. Just a request, if too much work then I will hunt them all down myself.

I haven't been in the benchmarking biz for some years now, still climbing back into the saddle.

SuperPi
http://www.xtremesystems.com/pi/
If link doesn't work, just look for SuperPi XS Mod 1.5

3DMark01 SE
http://www.gamershell.com/download_6088.shtml
I think i am running the pro version though, which allows me to select just the games benches rather than all the graphics card testing ones.

Q4 bench v1.3
http://www.hocbench.com/q4.html
I ran with 800x600 High Quality no AA/AF/SMP enabled
Used id netdemo

WinRAR
http://rarlabs.com/rar/wrar362.exe
I'm still using an old 3.60 beta 7 version though.

 

[n7]

Originally posted by: Acanthus
Uhh, any real world performance differences?

Im a lot more interested in performance at settings i would be actually using...

Id wager in most cases it will be less than 1% in real world apps that arent running at minimum settings.

Not much really.

But then again, for stuff like WinRAR (which i use alot), or for video encoding over the span of a few hours, little differences can add up.

That being said, for games, the main thing is always the GPU.

However, the point of this thread is to show that at least on P965, higher bandwidth will basically always beat tighter timings.

I am sure it's similar on 975X & 680i, but it appears everyone with those mobos is scared to prove me wrong

C'mon guys with those chipsets, talk means nothing; i wanna see results.

We already know that 1T pwns pretty much everything on 680i, so in that case, since you can't get very high with 1T, then yes, lower bandwidth + tighter timings + 1T > all but insanely high bandwidth.

I'd like to see some 975X results though

 

[n7]

I added some FEAR & UT2k4 results.

I'm not sure why i bothered with FEAR; it's always been massively GPU-limited.

I forgot to remove minimum desired framerate from my UT2004.ini when benching it, but i highly doubt that being left to 35 affected the results too much

 

[Acanthus]

Yeah those were the results i expected in more modern games, cranking the settings up more would further diminish the spread.

My advice has always been: Unless youre an artificial bench junkie, get enough ram at a decent clock rate, and ignore the timings. High end ram really is the lowest bang for buck you can toss in a gaming machine thats on a budget of any kind.

 

[The-Noid]

Actaully I have over 100 pictures of 975x. I will have my computer up tomorrow. 4:5 DDR2 540 mhz 4-4-4-4 averages 7,000 megs read under/write under sandra.
1:1 420 3-3-3-8 averages 8,000.

I would test right now but I have no 975x chipsets left around. I just want to be perfectly honest with you n7. The people over at XS do this for a living, if you really think they would be wrong on 1:1, you don't give them enough credit. They have tested 100's of possibilities. Everyone knows that 1:1 is faster on 975x, 4:5 is faster on 965p. 680i is a crapshoot. I have about 100 pictures of it. 3-3-3-8-1T is as fast as DDR2-1100 4-4-4-8. If you link fsb:ram on 680i it is also faster. You are on 965p, you results do not matter to any other motherboard as they all perform different.

Why wouldn't you run 1T on 680i as well that is the whole reason to have the board. I will show some nice 2-2-2-2-1T shots from 680i. They have pretty good speed, but it's kind of unstable. Once I get a motherboard in I will get back on my computer. The ram test blew up my evga (literally blew it up, heard a loud pop and then couple of capacitors caught fire). <--- Grumpy this morning, I apologize in advance.

 

[Idontcare]

Originally posted by: n7

Originally posted by: Idontcare
n7, can you hotlink your test cases to the benchmark software download site?

I want to contribute but want to make sure I use whatever you used. Just a request, if too much work then I will hunt them all down myself.

I haven't been in the benchmarking biz for some years now, still climbing back into the saddle.

SuperPi
http://www.xtremesystems.com/pi/
If link doesn't work, just look for SuperPi XS Mod 1.5

3DMark01 SE
http://www.gamershell.com/download_6088.shtml
I think i am running the pro version though, which allows me to select just the games benches rather than all the graphics card testing ones.

Q4 bench v1.3
http://www.hocbench.com/q4.html
I ran with 800x600 High Quality no AA/AF/SMP enabled
Used id netdemo

WinRAR
http://rarlabs.com/rar/wrar362.exe
I'm still using an old 3.60 beta 7 version though.

n7, thanks!

 

[n7]

Yoxxy, while i'm not disagreeing with you regarding 1:1 vs. 4:5 on 975X, i am not finding any real world testing in the threads i've seen thus far.

I don't really care too much about SuperPI or Everest bandwidth; i'd like to see some games benching & 3DMark01 testing like i've done in my OP.

If you have any links for this, it'd be appreciated

 

[cmdrdredd]

Originally posted by: n7
Yoxxy, while i'm not disagreeing with you regarding 1:1 vs. 4:5 on 975X, i am not finding any real world testing in the threads i've seen thus far.

I don't really care too much about SuperPI or Everest bandwidth; i'd like to see some games benching & 3DMark01 testing like i've done in my OP.

If you have any links for this, it'd be appreciated

This would be nice to see too. I'd also be interested to find how much difference 1T really is. Everyone talks big about it, but I find nothing that actually shows.

 

[The-Noid]

A picture is worth a thousand words.

http://img214.imageshack.us/img214/1603/3439yr8.jpg
http://img80.imageshack.us/img80/4497/4448kk0.jpg
680i definetly likes bandwidth. Untweaked OS, no LOD. Fresh install. 975x shots tomorrow. I will also do 1T to show how big the difference is.

 

[lopri]

Yoxxy, what do you do to run 1T on your 680i? I've tried everything possible but the most I can do is single Super PI 8M. Share your secrets.

 

[lopri]

Originally posted by: cmdrdredd
I'd also be interested to find how much difference 1T really is. Everyone talks big about it, but I find nothing that actually shows.

Here is one. There was another thread regarding this, and I managed to run Doom 3 timedemo with 1T later on. If I remember correctly the difference between 800MHz/2T and 1000MHz/2T was 1~2 FPS, and 800MHz/1T was faster than everything else by like 5 FPS.

http://img292.imageshack.us/img292/603/680isp8mddr8001tjx0.png
http://img149.imageshack.us/img149/5030/680isp8mddr11002txf3.png

But under 2T, 1000MHz/4-4-4 is faster than 800MHz/3-3-3 on 680i whether linked or unlinked.

 

[The-Noid]

Raise TRC to around 16-17 works best for me. You need a lot of voltage on the ram (2.40 works pretty well). I will post some shots of 3d01 here in the next couple of minuts at 1t.

 

[cmdrdredd]

Originally posted by: lopri

Originally posted by: cmdrdredd
I'd also be interested to find how much difference 1T really is. Everyone talks big about it, but I find nothing that actually shows.

Here is one. There was another thread regarding this, and I managed to run Doom 3 timedemo with 1T later on. If I remember correctly the difference between 800MHz/2T and 1000MHz/2T was 1~2 FPS, and 800MHz/1T was faster than everything else by like 5 FPS.

http://img292.imageshack.us/img292/603/680isp8mddr8001tjx0.png
http://img149.imageshack.us/img149/5030/680isp8mddr11002txf3.png

But under 2T, 1000MHz/4-4-4 is faster than 800MHz/3-3-3 on 680i whether linked or unlinked.

How about something other than a synthetic benchmark? Got any encoding times etc?

 

[cmdrdredd]

Specs in sig same settings as N7 for each game. 1:1 = DDR2-800 4-4-4-8 and 4:5 = DDR2-1000 5-4-4-8

Quake4
1:1- 112.3fps
4:5- 116.7Fps

F.E.A.R
1:1- 62Min 129Avg 296Max
4:5- 63Min 129Avg 300Max

UT2k4
1:1- 92.574fps
4:5- 100.749fps

3DMark01
1:1- 42335
4:5- 43524

Winrar
1:1- 1369KB/s
4:5- 1540KB/s

SemperPi 4M
1:1- 1m 36.469s
4:5- 1m 32.953s

 

[cmdrdredd]

updated with Ut2k4

 

[n7]

Originally posted by: cmdrdredd
Specs in sig same settings as N7 for each game. 1:1 = DDR2-800 4-4-4-8 and 4:5 = DDR2-1000 5-4-4-8

Quake4
1:1- 112.3fps
4:5- 116.7Fps

F.E.A.R
1:1- 62Min 129Avg 296Max
4:5- 63Min 129Avg 300Max

3DMark01
1:1- 42335
4:5- 43524

Winrar
1:1- 1369KB/s
4:5- 1540KB/s

SemperPi 4M
1:1- 1m 36.469s
4:5- 1m 32.953s

Thanx for adding your results

Pretty amazing just how big of a difference 400 Mhz makes even, since i benched @ 2.8 GHz vs. your 3.2 GHz.

I should try running @ 7x400 to see how big the benefit of staying on the 1066 strap is.

BTW for anyone with UT2k4, here's the timedemo i use for benching UT2k4:
http://www.3dcenter.org/downloads/ut2004-primeval.php