Infineon CE-5 vs Samsung UCCC - 18 benchmarks compared!!!

[bachus_anonym]

Yo XS!

Have you ever wondered which 2GB kit would perform better in your setup? Or how much difference there is between them? This short write-up is an attempt to give one an idea how Infineon CE-5 based 2GB set stacks up against it counterpart built with Samsung UCCC.

I "locked" my Opteron @ 3200MHz and tested each set of memory (CE-5 and UCCC) at it's max possible frequency that particular divider allows - div183 for UCCC @ 291MHz 7-4-4-3.0-1T and div166 for CE-5 @ 267 5-2-3-3.0-1T.
I put as much effort as I could to run each type of memory "as tight" as I could. I decided to limit my test to CAS3.0 on CE-5 as well as TRCD=4 on UCCC only. Both kits have somewhat above average OC potential.

Tests were run one at a time, screenshots taken, then PC was rebooted and next test was performed.

Setup:
AMD Opteron 146 @3200MHz
DFI LP Ultra-D R.A02, BIOS 11/14 Official
NForce4 drivers 6.70, USB, Network and Audio enabled.
EVGA 7800GTX 256MB KO @ 500+40/1400 (low-power mode disabled thru BIOS-mod)
Forceware 84.21, default driver settings (Quality)
Windows XP SP2 fresh install, not tweaked
2GB Corsair TWINX2048-4400PRO (Infineon CE-5)
2GB G.Skill F1-4000USU2-2GBHZ (Samsung UCCC)

There you are, here are the results - 1st column for CE-5 and 2nd for UCCC...



3D Benchmarks

3DMark2001 ------> 35371 ------- 35055

3DMark03 ---------> 19687 ------- 19679

3DMark05 ---------> 9173 --------- 9206

3DMark06 ---------> 4681 --------- 4711

Aquamark3 -------> 105792 ------ 105123

X³: Reunion ------> 77.92 --------- 78.05


Productivity/Pi/Synthetic Benchmarks

PCMark04 --------> 6428 ---------- 6396

PCMark05 --------> 5098 ---------- 5054

SuperPi 32M -----> 22:57 --------- 23:08*

PiFast 10M -------> 37.55 --------- 37.72

WinRAR ----------> 787 ------------ 773

Sandra Band. ---> 7902/7814 --- 8187/8133

Everest Lat. -----> 38.8 ----------- 38.4

SM2.0 L1 Band. -> 36848.43 ---- 36824.17

SM2.0 L2 Band. -> 11678.01 ---- 11609.84

SM2.0 Mem Band. -> 7482.55 -- 7918.02


Game Benchmarks at 1600x1200

DOOM 3 ----------> 107.0 --------- 106.8

FarCry ------------> 134.47 ------- 133.10

F.E.A.R -----------> 66 ------------- 65

Quake 4 ----------> 112.3 -------- 112.5


Game Benchmarks at 1024x768

DOOM 3 ----------> 125.3 --------- 124.4

FarCry ------------> 158.57 ------- 159.87

F.E.A.R -----------> 125 ----------- 125

Quake 4 ----------> 153.4 -------- 153.9

I just thought this could be of use for some. I'm gonna leave conclusions to you guys...
Questions, comments or requests welcomed

[ozzimark]

curious that the infineon takes a slight lead in so many of those tests despite having a slightly higher latency and lower bandwidth. guess i won't be putting much stock into those two tests anymore

very nice work as always bachus, though just two questions.. well, one question and one request really:

1. why the * after the 32m time for the UCCC?
2. would you mind running the memory bench in sciencemark?
http://www.sciencemark.org/frame.php?frame=40

if it take more than ~15 min to set up stuff to run another test, i won't be offended if you choose not to, but i think it'll be quite interesting to see how the memory bench compares to everest/sandra, since it breaks up memory latency into various sizes and into L1, L2 and memory

[guess2098]

3DMark05 ---------> 9173 --------- 9206

3DMark05 ---------> 4681 --------- 4711

yo!!!

[MorueM]

3DMark06 ---------> 4681 --------- 4711

Working lnks..


Good idea, good job on the run. Could that ce-5 be run 3.3.2.5?

[bachus_anonym]

1. why the * after the 32m time for the UCCC?
2. would you mind running the memory bench in sciencemark?
http://www.sciencemark.org/frame.php?frame=40

1. 291-292Mhz is the edge of 32M stability for my UCCC. Unfortunately, I did not manage to complete full 32M run @ 291Mhz this time. That's why I was forced to post 32M @ 290Mhz instead. Otherwise, time would have been about 4s better...
2. 1st post updated with results Thanks for good word and your suggestion

yo!!!

Sup Denny! Fixed it

Good idea, good job on the run. Could that ce-5 be run 3.3.2.5?

Thanks You mean, with CAS2.0? If so then yes, last time I checked it was fully stable slightly above 220MHz 5-2-3-2.0-1T.

----------

I have also run Guru3D's timedemo in BF2 but I can't find the way to get reliable results with it. I'm gonna work on it a bit more, though.

[HKPolice]

Hmm, I think these tests would be more pertinent if they were ran on an dual core setup because Doom3 and Quake4 are multithreaded and the nVidia drivers are dual core optimized even if the game isn't. The question remains: Which is more beneficial to a Dual core setup, higher bandwidth or lower latency?

Also, the single Video card setup might start to pose as a slight bottleneck as well depending on what settings you used for the gaming benches. In the case of Quake4 and FarCry which ran at 1600x1200, Vid card is a definite bottleneck.

[high5]

You mean, with CAS2.0? If so then yes, last time I checked it was fully stable slightly above 220MHz 5-2-3-2.0-1T.

no, i'm pretty sure he meant Trp=2. EDIT: obviously a typo in the 1st post some of us fell for.

also, what about Trp=3 for Samsung?

BTW you did some serious benching there.

[Revv23]

wow very nice work!!

i thought my 265 infineon was weak next to uccc now i dont feel so bad

[Cooper]

Very nice work Bachus. But as HKPolice mentioned dual-core tests would great to see
Another proof that A64 just loves low timings

[MonkSP]

I dont know if I undestand this correctly but you testeing them with diferent OCs right?

why dont you test both at 265 htt? That will be a more fair benchmark.

[high5]

Another proof that A64 just loves low timings

hmm... having in mind that the differences are actually about ~1%, i wouldn't say that 'A64 just loves low timings'.

when it comes to _real-world_ performance, the influece of memory on K8 is IMHO overrated. there aren't any differences you'll see with your bare eyes.

I dont know if I undestand this correctly but you testeing them with diferent OCs right?

why dont you test both at 265 htt? That will be a more fair benchmark.

CPU was always at 320×10 MHz.

this is comparison between Samsung UCCC at its best & Infineon CE at its best.

[krampak]

why dont you test both at 265 htt? That will be a more fair benchmark.

We know for sure that with both at 265Mhz would win CE-5.
I find right this comparison

[nugzo]

Yo XS!

Have you ever wondered which 2GB kit would perform better in your setup? Or how much difference there is between them? This short write-up is an attempt to give one an idea how Infineon CE-5 based 2GB set stacks up against it counterpart built with Samsung UCCC.

I just thought this could be of use for some. I'm gonna leave conclusions to you guys...
Questions, comments or requests welcomed

Very nice.. Thanks for sharing your results!

[alexio]

thanks for this little comparation. I think as you know yourself it's much better to run the gametests at a 1024*768 resolution because the system becomes more of a bottleneck then, but as you posted you plan to do this

High speed RAM is overrated very much, latencies are more important as you see in this little comparation.

I'm actually fairly sure that if you can run 3-2-2-* or 2.5-2-2-* at about DDR460 on one of these sets it will be actually faster in many benchmarks (in these sets of timings the '3' and '2.5' are CAS).

Thanks again for this nice comparation, and if your have any time can you please include tighter latency results?

[stummerwinter]

Nice work!

Again you can see, bandwith is not the bottleneck at the A64...

[ahmad]

I like the comparisons man. Shows just how much timings do matter on A64s. The higher bandwidth shown in sandra and 3dmark05/06 show that timings are not that big of a deal with those apps (which is very true). But for the rest, its latencies and not bandwidth that matters.

A couple of things: you said CE-5s ran at 5-3-3-3 and I wondered why your Trp was at 3 then I checked the benchmarks and it is in fact 2. A typo

Why run at Tras of 5 and Trrd of 2 when you can do 1/0 with the CE-5? You get slightly better SPI times with Tras @ 1.

Its good I got myself the infineon.. those are nasty timings on that UCCC expecially Trcd.

[vanquish]

Great comparison !

You must have put in some solid hard work there

[DAK1640]

Nice analysis bachus_anonym, thanks for taking the time...

[shuRe]

anohter great compaison, well thought out and executed.

[perry_78]

Amazing work there bachus, but you're very much aware of that

Next point, can you get hold of an FX to try 1:1 action?

[ozzimark]

2. 1st post updated with results Thanks for good word and your

hmm, sciencemark says the same thing about latency!
thanks for indulging my curiosity

[Getttosmurf]

nice work

[coop]

I don't even have this type of memory, but enjoyed seeing the comparisons. If I was to get any this would be very helpful. Thanks for taking the time and sharing the results.

[xs64]

Awesome job bachus_anonym That took so much time and energy

I wonder can you make a topic , which Rams use CE-5 ?

[high5]

now someone should provide bachus with some Micron rev. F stuff... that way we would have a complete overview of all the options for high-performance 2GB kits.