Mhz Vs. Cas - Guess Whats Faster...

[Kunaak]

Super Pi 8M.
all settings done from bios.

I just chose one speed that I could use multiple fast mhz, and cas settings from. 7 x 367 worked out pretty good for me.

Hardware...

Asus P5B Deluxe. Bios 0711.
Mushkin PC 6400 2 gig kit, cas 4-4-3-10 set.
E6300 CPU.

from the looks of this, MHz is more important then cas alone, and 1:1 isn't better then other higher ratios...unless you can run 1:1 at over 1 ghz of course.

733 mhz at 1:1, cas 3-3-2-8 = 4M-10 Sec.

920 mhz at 4:5, cas 4-3-3-8 = 4M-3 Sec.

1100 mhz at 2:3, cas 5-4-4-8 = 3M-52 Sec.







this is just a simple test, and not meant to be conclusive.

[sierra_bound]

No real surprise. As I mentioned to someone else in another thread, if you have RAM capable of high clocks, take advantage of it.

[NickS]

Nice results, but welcome to last week.

I've been trying to explain to a lot of intel users that MHz > Timings. I guess this finally proves it clearly.

[sierra_bound]

Nice results, but welcome to last week.

I've been trying to explain to a lot of intel users that MHz > Timings. I guess this finally proves it clearly.

You mean welcome to two months ago. The people who weren't obsessed with FSB speed saw this early on.

Almost all my best Pi times were done running 4:5 or 2:3 with looser timings. You won't really see the benefit when you run 1M. But on the longer iterations, the value of higher memory speed really becomes clear.

[xlink]

you realize that the timings are measured in cycles and that if you increase the cycles the latency is staying effectively the same right?

[sierra_bound]

you realize that the timings are measured in cycles and that if you increase the cycles the latency is staying effectively the same right?

Timings do make a difference. It's just that RAM speed tends to be more important. Try running SuperPi 32M at 3-3-3 and again at 4-4-4 with same FSB and memory speeds. 3-3-3 will be faster, I guarantee you.

There were a couple of articles early on saying timings don't matter, period. Well, if you bench, timings do matter, especially if you're running 1:1. Writers tend to over-generalize.

[Revv23]

timings always matter, just mhz matter more muhahaha.

[guess2098]

nice test
but i done it like few months b4 hhehehhee
always good to remind all member!
higher mhz is better

[fhpchris]

Timings do make a difference. It's just that RAM speed tends to be more important. Try running SuperPi 32M at 3-3-3 and again at 4-4-4 with same FSB and memory speeds. 3-3-3 will be faster, I guarantee you.

There were a couple of articles early on saying timings don't matter, period. Well, if you bench, timings do matter, especially if you're running 1:1. Writers tend to over-generalize.

OPB said himself that 1:1 @ 3-3-3 was faster than 1200mhz 4-4-3, so I am sure this issue isnt as cut and dry as just mhz or just timings

[sierra_bound]

OPB said himself that 1:1 @ 3-3-3 was faster than 1200mhz 4-4-3, so I am sure this issue isnt as cut and dry as just mhz or just timings

Faster in what? iM? 8M? 32M? 3DMark?

I don't think anyone here said it's one or the other.

Let me quote myself.

"Timings do make a difference. It's just that RAM speed tends to be more important."

OPB uses mostly 975X boards. I can tell you they won't do 1200MHz 4-4-3, so that's kind of a moot point.

[fhpchris]

http://forum.coolaler.com/showpost.p...2&postcount=49

Congratulations less frenzy, the first eight seconds when the male vein lower than roro 19mhz, pi has a fast 0.2 seconds.fsb os and it is very important not know Vcore number, and C is his u is the same?Unfortunately, some feel that C is his shoulders.If the measurements above the 5.8 GHz band should be 1.73v 5526 pi 8.5 seconds on the strength of this proportion could not be considered;Before I go 5566mhz also try using 14x400 5600mhz.na run 1,200 2:3.4-4-3 ... the pi 1m 1m 9.125s 5566mhz worse there than in 1 :1 position, the higher the frequency, the greater the burden mch and ram, the relatively higher Vcore increases,cpu load greater, although the frequency of super elevated,But, it will reduce the efficiency of C. .. This is the great debate yesterday and today ..

1200 2:3 4-4-3 worse! than 1:1... at least that is how I think the translation is, my traditional chinese is rather bad...

[sierra_bound]

http://forum.coolaler.com/showpost.p...2&postcount=49



1200 2:3 4-4-3 worse! than 1:1... at least that is how I think the translation is, my traditional chinese is rather bad...

But look at Kunaak's results. Don't they disprove that theory. Kunaak ran 3-3-3, 4-3-3 and 5-4-4 at same CPU speed. Only memory speed changed.

[fhpchris]

I think we are agreeing here

I know 368ish 3-3-2 isnt that fast from experience, but that does not mean cas 3 cannot be fast at all, Im sure there is one point where running cas 3 is really fast, it would be great If I could get some good runs @ 1200 mhz in to compare.

I will see in ~a week or so

[exhausted mule]

you realize that the timings are measured in cycles and that if you increase the cycles the latency is staying effectively the same right?

i aggree with link here.

if your running at 100mhz with a 3cycle delay, you'll be able to do something else for the other 97cycles present in the test.

if your running at 200mhz with a 4cycle delay that'll give you 196 cycles to mess with.

its pretty simple if you think about it and can easily be represented with math. but of course there will be a tradeoff once the overall speed difference is smaller and you'd have to factor in the other timings as well and their relations but after a given overall speed the fsb speed will offer more bandwith.


i'm sure chipset latency has a play in this as well but you get the picture.

[sierra_bound]

I think we are agreeing here

I know 368ish 3-3-2 isnt that fast from experience, but that does not mean cas 3 cannot be fast at all, Im sure there is one point where running cas 3 is really fast, it would be great If I could get some good runs @ 1200 mhz in to compare.

I will see in ~a week or so

OMEGA3 and OPB run 1:1. They are the two best at 32M at the moment. But they also use X6800's and have lots of multipliers at their disposal. So they need good RAM speed, but not necessarily the fastest RAM speed. But I think for Joe Schmo, Kunaak's results are more relevant.

The people who say timings don't matter are smoking crack. The people who say RAM speed doesn't matter are also on drugs. Theories are neither here nor there. It's what the results show that counts.

[bachus_anonym]

Yep, pretty obvious but thanks for posting it, Eric CAS latency by itself does not make a whole lot of a difference (some - yes, but not a huge), even in majority of benchmarks we run around here, but especially 1M.

Same goes for your coolaler's post, fhpchris. SPi1M is really bad benchmark to run if you want to show or see differences timmings make. I can believe, that 460x12, 1:1, 5-3-3-3.0 vs 400x14, 2:3, 5-4-4-4.0 will not show a lot of variation for 1M, but I'm having a hard time believing same situation wll be for 32M, as that is the calculation where bandwidth matters.

I don't have 975x mobo, but e.g. sierra could run a quick 32M bench sometime and show us e.g. 433x9, 1:1, 5-3-3-3.0 vs 390x10, 2:3, 5-4-4-4.0 I'm pretty sure, bandwidth will win... Maybe not a huge one, but few good seconds for sure...

[sierra_bound]

I refuse to run 5-4-4.

I did do 393X10 2:3 @ 4-4-3 last night on my P5B Dlx. No tweaking.

http://img134.imageshack.us/img134/7...93x1023fk3.jpg

Now, compare that with someone else who was using a 975X board and running 1:1 at 3-3-3 at a higher CPU speed.

http://www.xtremesystems.org/forums/...4&postcount=29

[xlink]

again...

all the "minor" RAMs settings are probably staying the same and the latency increase is again almost negligeable.

going from 733mhz to 1000 is a 50% increase, the timings don't even increase by about 50% overall... technically its running at lower latency at 1100 in some scenarios(if you measure latency in nanoseconds as opposed to cycles)

[Tony]

You guys are not taking advantage of sub timings, asus losen timings in different modes and ratio's so you have to set like for like to see a true comparison.

Once we have fully opened up bios files you will soon see the power of 1:1 with lower latency over upclocks with relaxed timings im sure. At this time we are being crippled by bios files...trust me

[Revv23]

You guys are not taking advantage of sub timings, asus losen timings in different modes and ratio's so you have to set like for like to see a true comparison.

Once we have fully opened up bios files you will soon see the power of 1:1 with lower latency over upclocks with relaxed timings im sure. At this time we are being crippled by bios files...trust me

Thats why 975x xtreme benchers stay at 1:1

[sierra_bound]

Thats why 975x xtreme benchers stay at 1:1

I have the fastest 32M time with single-stage cooling and I was running 4:5 on a 975X board.

http://img434.imageshack.us/img434/6...m459845kz7.jpg

Denny was also using 4:5 and he has the fastest time with dry ice.

http://www.iamxtreme.net/cpu/x68/1041922.JPG

Those using LN2 don't need to run 4:5 or 2:3. They just raise the multiplier. The people with the top three times in the world all had RAM speed in the 440-445 range.

I run 32M a lot. So does bachus and a few other people here. Trust me, when you run a certain bench many times, you know what's best.

Theories are just totally meaningless in the world of benching. Don't tell me, show me. BTW, I'm not talking about you personally. I just can't stand those who aren't willing to put their theories to the test.

[hipro5]

NOTHING can beat 400MHz+ 1:1 at Cas 3 - 2 - 2 - 1 with FAST subtimmings.....Though you CAN'T find such rams today!....
Well.....It's also good the 4 - 3 - 2 - 1 with FAST subtimmings over 450MHz as well BUT 1:1.....

[hipro5]

Theories are just totally meaningless in the world of benching. Don't tell me, show me. BTW, I'm not talking about you personally. I just can't stand those who aren't willing to put their theories to the test.

Get me a bench like this EVEN at 1300MHz DDR frequency....Not theory.....Fact.....

[hipro5]

733 mhz at 1:1, cas 3-3-2-8 = 4M-10 Sec.

920 mhz at 4:5, cas 4-3-3-8 = 4M-3 Sec.

1100 mhz at 2:3, cas 5-4-4-8 = 3M-52 Sec.

How about testing at 733MHz at 1:1 with Cas 3-2-2-1-15us-8-20-5(or4)-8-8-6-3-4-3 .....You have Fatbodies....\

The people who say timings don't matter are smoking crack. The people who say RAM speed doesn't matter are also on drugs. Theories are neither here nor there. It's what the results show that counts.

100% agree but with 1:1 as well.....

I refuse to run 5-4-4.

I did do 393X10 2:3 @ 4-4-3 last night on my P5B Dlx. No tweaking.

http://img134.imageshack.us/img134/7...93x1023fk3.jpg

Now, compare that with someone else who was using a 975X board and running 1:1 at 3-3-3 at a higher CPU speed.

http://www.xtremesystems.org/forums/...4&postcount=29

Mmmmm......welllll NOT exactly the "same"......You're running tweked and he has "normal windows".....If he tweaks too, you're down....

Whould you like me to run one at 3930MHz at 1:1 tweaked with rams at Cas 3-2-2-1...with the rams at ONLY 393MHz....?..Your rams are at 589MHz.......I'll be at 393MHz.....thas 196MHz ram difference.....!.....

[sierra_bound]

Get me a bench like this EVEN at 1300MHz DDR frequency....Not theory.....Fact.....

Wow, you had to resort to Cas4.

Try doing 460MHz on your Cellshock memory with Cas3 (3-3-3) timings on an unmodded board.



I'm using new TeamGroup memory. By the way, I'm still benching. 32M at 461MHz 3-3-3 is almost done.