Is higher memory speed always better than lower timing?

S

Saint Andy

n00b
Joined
May 1, 2007
Messages
13
I've got an e4300 clocked at 3ghz at the moment, the memory, ddr2-667 is running at 3,3,3,12 at 880mhz. I've had the ram at over 1000mhz with relaxed 4,4,4,15 timings

Do I want to always aim for the highest speed I can get? Or is it better to have tighter timings sometimes?

Should I relax the timings even further? I've heard the ram will do 1200mhz at 5,5,5,18.

this is all at 2T as the board won't do 1T I don't think (ds3).
 
personally, tighter timings seems to be better once you're over 800mhz... You just don't use that bandwidth in most cases!

What ram is that btw?
 
the difference between 4-4-4 and 5-5-5 is hardly noticeable even in benchmarks, it might mean 50-100 3dmarks. but the difference between 1000 and 1200 is 17% which is very noticeable. you should try both and see what you like :p
 
ghost6303 said:
the difference between 4-4-4 and 5-5-5 is hardly noticeable even in benchmarks, it might mean 50-100 3dmarks. but the difference between 1000 and 1200 is 17% which is very noticeable. you should try both and see what you like :p
Click to expand...

Which benchmark?
 
Arcygenical said:
Which benchmark?
Click to expand...

2006. of course the screen shots i had are on the hard drive that just died and with the ram that i just RMA'd. i can try it again when i get the ram back. but i tried all sorts of timings and the difference was less then 200-300 points. the difference between 5-5-5-15 (stock for that ram) and 4-4-4-13 was around +130 something 3dmarks if i remember correctly. i ran the test three times and averaged the differences. though i can run 3dmark 5 times back to back, same settings and speeds and always get slightly different numbers. Sisoft Sandra doesnt like my computer for some reason, it wont run at all, so i didnt test actual memory bandwidth, but real world performance such as 3dmark showed little difference
 
It's crucial ballistix. Seems pretty good so far, doesn't get warm with a zalman 7700 above at mid speed at 2.2v.

update - i had some problems at 4,4,4,15 at anything over 1000, had to drop it to 5,5,5,18 for stability after further testing. still great for pc5300 ram I think.
 
Corsair FTW
PC8500_CPUZ.jpg
 
Saint Andy said:
thanks for the picture. but do you have any opinion on timing vs speed?
Click to expand...

Speed is better then timing IMO
Timing really doesn't impact performance. Timing can reduce latency. However, when you can match the speed of your ram with a good OC IMO you will see better gains then worry about tightening it up. Like I said, tighter ram shows well in synthetic benchmarks but in practical real world (such as games, etc) it's non existent. I know that if you run the winrar benchmark or Super PI you may see better results with tighter timing (for example) but for games, no. And, from my personal experience you can hurt performance if the timing is not jiving with your chipset. Although I am talking personal experience with limited make/models of ram. I found it more practical and noticeable to increase the speed and get the lowest timing possible: 4-4-4-12 with PC7100 or PC8500
VS
a low speed ram and a low timing: 3-3-3-10 or PC5400 or PC6400
 
IMO higher clocks are better than tighter timings. Its when you can get the tighter timings with the higher clocks you have found perfection. Just dont expect that to come cheap.
 
remember this is only true for core2s. older P4s and especialy AMD XF and XP series benefit much more from tighter timings then core2 chips do.
 
In real world performance, I think both will not be noticable....In benchmarks though, high mem speed with looser timings is faster than tight timings at lower speed..... But ofcourse nothing beats having best of both worlds...... Crucial Ballistix pc5300 :D

630.JPG
 
From the looks of things in the >800mhz zone, cas3 t ocas4 difference is a whole lot less than cas4 to cas5. That said, cas3 is pretty difficult, and you'd probably gain a lot more mhz on the ram, way offsetting the cas differences. Also, I know the p5b has the trcd issue when set to 3 is slow, but I'm not sure if that's just p5b or if it's general for p965.
 
I asked the same question here:
http://www.hardforum.com/showthread.php?t=1186580

Well, a bit of a different question and another slightly different question, but anyways, check the driver heaven review I linked. In real world performance (not pointless sandra graphs) it doesn't look like upping clock speed or tightening timings does a whole lot for anything.

Review:
http://www.driverheaven.net/reviews/PC2-6400 7-way memory round-up/index.htm

They experiment with tighter timings at 800MHz or clocking higher to 1100 MHz. While bandwidth goes up (yippee, who cares) real world performance increases by a very small margin. 1s on super pi etc. Unless you are a benchmark fiend I would say don't bother.

So I've resolved to stick with running 668MHz ram. Woohoo!
 
Im with you on that one..... never had a stick higher than ddr667 lol....
 
ghost6303 said:
remember this is only true for core2s. older P4s and especialy AMD XF and XP series benefit much more from tighter timings then core2 chips do.
Click to expand...

The Pentium 4's were bandwidth starved and benefited from high bandwidth more than they did timings.Core 2 Duo processors benefit from increased bandwidth over timings as well but it's a small percentage in comparison.
 
You should read this also as your FSB also comes into play with regards to the NB clock.

Secrets of the i965 based Asus P5B and chipset strap uncovered

http://www.thetechrepository.com/showthread.php?t=30

"From 467fsb onwards you will now see the highest performace from your dimms although you are relying on how good the NB overclocks under the 1333 strap."

This tends to be the consensus at XtremeSystems as well.
 
While theoretically that should be the way it works, it doesn't seem to be so in actuality, at least from the point of memory benchmarks. I'm sorry I don't have any figures to back up my claims at the moment since I'm currently without a bios chip on my main computer. But when I ran 4:5 at 400 for 500 ram, and also 1:1 500, both with the same timings, I was seeing gorgeous latencies at 400fsb, and meager at 500. Also, bandwidth at 500fsb seemed to average quite a bit lower, with the exception of sse tests, which it was a bit ahead.

That said, it's only with the mbench program. I haven't done much further testing on the subject since I wasn't all too comfortable with my p5b's ability to run 500fsb. Hopefully once I get my dfi running tomorrow, I'll do more testing, notably 533fsb vs 400fsb comparisons, albeit 400 will be a little memory handicapped by 66ddr2 mhz.
 
That DFI should do 500 fairly easy so we await your test results! :)

I ran a few tests myself. Ran each one 3 times and will post the average of the three. I ran these tests at orthos stable speeds. At above 500 I am only stable with high voltages that I don't want to go.

First at 500x7FSB (3500)
Mem 1000mhz 5-5-5-15 2T
1066 Strap, 1:1

Acquamark3

180, 378

Sandra Lite XI SP2

Bandwidth Int Buff: 8599 mb/s
Bandwidth Float: 8658 mb/s


Latency Random Access: 65 ns
Speed Factor: 72.6

440FSBx8 (3520)
1086 Strap
Mem 1100mhz 4:5 5-5-5-15 2T

Acquamark3

188, 768

Sandra

Bandwidth Int Buf 8389 Mb/s
Bandwidth Float: 8402 mb/s

Latency (RA) 62 ns
Speed factor 69.0

So at 500FSB I see a bandwidth gain and a latency increase. Of course there is a 100mhz memspeed difference too at 440x8 4:5. I can't get 500FSB to run mem any faster as the next choice is 1250mhz which I can't load into windows at
 
jzodda said:
First at 500x7FSB (3500)
Mem 1000mhz 5-5-5-15 2T
1066 Strap, 1:1

Latency Random Access: 65 ns

440FSBx8 (3520)
1086 Strap
Mem 1100mhz 4:5 5-5-5-15 2T

Latency (RA) 62 ns
Click to expand...

Hmm that's really interesting. Do you have the enhance memory options set to normal or fast?

It looks like the dfi is configuring internal latencies progressively or something. I've done testing on a p5b board, and here are some ballpark figures.

400fsb (seems to have some sort of tweak to internal latency / strap)
using 4:5 divider for 1000 4-4-4-4
I get latency of about 47.

When I go 500fsb
using 1:1 divider for 1000 4-4-4-4
I get latency of 60

Both is with static read control to fast (I'm assuming this is similar to dfi's memory enhance features)

When I run 401fsb with same settings otherwise as 400
using 4:5 divider for 1000 4-4-4-4
I get latency of about 72.

So clearly we can see a major change in internal latencies when going from 400 to 401 in the p5b. And even once we get back to 500fsb, 1:1, the latency still doesn't come close to running 400 4:5.

That said, the p5b seems more optimised for running 4:5 multiplier than the dfi. In my limited testing with the dfi, running 400 4:5 required a whole lot more vmch to be applied to have stability.

But I recall in my brief testing before I decided to flash the bios and brick it, that at 500fsb, latency was still 60 or so, so despite being set to 266strap, there's still some sort of internal latency change going on.
 
I've got PC8500 RAM with a C2D E6600. Would it be better to let the memory stay at 1066 or even OC it higher and stick with stock timings (4-4-4-12), or bring it down to the speed of the processor (around 400mhz after I OC it) and get really low timings? What difference does the linked/unlinked make?
 
what motherboard do you have? cas3 to 4 differences aren't as great as 4 to 5, so you'd probably be better off seeing how far you can go on cas 4.
 
ziddey said:
Hmm that's really interesting. Do you have the enhance memory options set to normal or fast?
Click to expand...

hmm I don't have that option with the Quad GT. This is the advanced chipset menu



DRAM Timing Selectable (manual or Spd)
X - CAS Latency Time (tCL)
X - RAS# to CAS# Delay (tRCD)
X - RAS# Precharge (tRP)
X - Precharge Delay (tRAS)
X - Command Rate
X - Refresh Cycle Time (tRFC)
PCI-E Compliancy Mode V1.0a
PEG Force X1
Init Display First
 
ohh my apologies. the way you started off the post, i misinterpreted it as you having a dfi board too. I should have checked your sig.
 
ziddey said:
ohh my apologies. the way you started off the post, i misinterpreted it as you having a dfi board too. I should have checked your sig.
Click to expand...


I wanted the DFI board badly! When I ordered all the parts for the rig it was out of stock everyplace but Atacom. So I bought it from them and 2 days later as the parts started to arrive they called me to tell me it was out of stock and would I not mind waiting till May 15th. :eek:

So I canceled it, made sure it was not in stock anyplace else (Some A-Hole on pricewatch had it and wanted $250 for it)

So I ordered the QuadGT-So far I am pleased with it. The release bios was a disaster but now things are pretty good. The hardware itself is very solid. I don't think the DFI would make much difference for me because 3500mhz seems to be the limit on my 6700 anyway, at least with Vista. When I go for 3600 at any FSB I have to run past 1.5Vcore to stay. Kind of funny that my sweetspot winds up exactly at 500FSB, although I think I can do up to 502 Orthos stable. When I have this all finished I think I will put a WindowsXP partition up so I can see how much more I could get with XP.

So thats my short lived adventure with the DFI P965-S :)
 
ziddey said:
what motherboard do you have? cas3 to 4 differences aren't as great as 4 to 5, so you'd probably be better off seeing how far you can go on cas 4.
Click to expand...

I have the DFI Infinity P965-S...it's coming FedEx today. :)
 
Update time. Hope these mbench results speak for themselves:

Test 1:

p5b 400fsb 4:5 500 4-4-4-4-4-25-3-9-4-10:

Access latency 48.7 ns (156 clocks)

Read datarate (INT) 7440 Mb/s
Write datarate (INT) 3141 Mb/s

Read datarate (MMX) 8194 Mb/s
Write datarate (MMX) 3140 Mb/s

Read datarate (SSE) 8579 Mb/s
Write datarate (SSE) 7267 Mb/s




Test 2:

p5b 500fsb 1:1 500 4-4-4-4-4-25-3-9-4-10:

Access latency 60.0 ns (240 clocks)

Read datarate (INT) 6797 Mb/s
Write datarate (INT) 2773 Mb/s

Read datarate (MMX) 7421 Mb/s
Write datarate (MMX) 2767 Mb/s

Read datarate (SSE) 7821 Mb/s
Write datarate (SSE) 9022 Mb/s




Test 3:

dfi p965-s 500 1:1 500 4-4-4-9-?-28-2-9-4-11:

Access latency 49.4 ns (198 clocks)

Read datarate (INT) 7952 Mb/s
Write datarate (INT) 3684 Mb/s

Read datarate (MMX) 8574 Mb/s
Write datarate (MMX) 3711 Mb/s

Read datarate (SSE) 9075 Mb/s
Write datarate (SSE) 9130 Mb/s





So. Latency checks out! :)! clocks still a bit higher, but probably because of the lack of the upscalar but instead a 1:1 ratio.

and bandwidth of dfi absolutely demolishes the p5b
 
I just did an mbench run also. Take into account that this is Vista 32 so I don't know how that impacts the results. I also can not do all those tight timings that you are doing at this FSB with my mem :)

Abit QuadGT 500 1:1 500 5-5-5-12-(no subtimigs)

Intel P6 processor (CPUID = 6f6) @ 5000.0MHz
Instruction set support : MMX SSE SSE2

Access latency 51.2 ns (256 clocks)

Read datarate (INT) 7614 Mb/s
Write datarate (INT) 2811 Mb/s

Read datarate (MMX) 8286 Mb/s
Write datarate (MMX) 2799 Mb/s

Read datarate (SSE) 8630 Mb/s
Write datarate (SSE) 9044 Mb/s
 
that's actually really impressive to be able to do that stably. I've had an issue with instability with memory enhancement to fast. Setting to normal with the same timings, I'm now getting pretty much the same result as I had on my p5b board, which is to say, way worse than your results. Hopefully I'll figure it out soon
 
You must log in or register to reply here.
Back
Top