Has Anyone seen a 1T vs. 2T command rate thread for A64's?

[Zebo]

In just about every mem review Wesley makes qualitative statements like this:

The Importance of Command Rate
Socket 754 Single-Channel motherboards performed best with a memory Command Rate setting of 1T in BIOS, but that generally was a stable option with only one DIMM. 2 or more DIMMs normally required a 2T Command Rate setting for most stable performance. There was a performance increase at the 1T Command Rate setting, but the real performance increase was very small.

Socket 939 Dual-Channel motherboards were found to exhibit a very wide performance difference between a Command Rate setting of 1T and a setting of 2T. The impact on memory bandwidth is dramatic between these 2 settings. In SiSoft Sandra 2004 standard buffered Memory Benchmarks, a 1T command rate showed a Sandra bandwidth of 6000 Mb/sec, while a 2T rate with the same 2 DIMMs in Dual-Channel mode was only 4800 Mb/sec. This is a huge difference in memory bandwidth and the Command Rate setting definitely impacts performance test results on Socket 939 motherboards. All AnandTech benchmarks were run at a Command Rate setting of 1T. This includes all benchmarks that were run in the CPU tests, as all benchmarks were rerun in the CPU tests as soon as we had verified the performance impact of Command Rate settings.

Well, proof ie data besides erronous Sandra? Hav'nt seen it anywhere, have you?

I'm thinking about making a comparison if no one can point me the right direction for real proof.

-----------------------------------------
***********MAY 1 2005**************

Edit: Since we have not found anything around the net or at reviewers sites dispite thier 1T "massive" benefit claims our members have done some tests!!! Wooo woooo.. I am leaving out synthetics since we already know the crazy story they like to tell; Sandra a whopping 35%, everest a massive 25% etc etc etc. Screw that! We don't want to freighten people into running 1T But we rather give a realistic performance benefit they see moving from 2T to 1T all things being equal.

GuitarDaddy System: CPU - Winch@2548Mhz | Memory- 2-3-2-7 X @232Mhz

SuperPI 8m 1T- 6:03m
SuperPI 8m 2T- 6:09m
1T is 2.4% faster

Aguamark3 1T- 58156
Aguamark3 2T- 57985
1T is 0.3% faster

3dmark 03 1T- 6908
3dmark 03 2T- 6872
1T is 0.5% faster

HL2 1T- 97.5fps
HL2 2T- 95.5fps
1T is 2.2% faster

X2 the threat 1T- 87.4fps
X2 the threat 2T- 85.8fps
1T is 1.9% faster

Doom3 1T- 85.9fps
Doom3 2T- 84.3fps
1T is 1.9% faster

Not looking good for 1T supposed "have to have" (thanks G))
-------------------------------------------------------------------------

Sir Avalon System: CPU - Sempy128KB@1800Mhz | Memory- 2-2-2-10 X @225Mhz

SuperPi 1M 1T- 49s
SuperPi 1M 2T- 50s
1T is 2.0% faster

HL2 (coast 05, 1024x768, 2xAA/4xAF) 1T- 83.5fps
HL2 (coast 05, 1024x768, 2xAA/4xAF) 2T- 81.0fps
1T is 3.0% faster

Doom 3 (1024x768, 2xAA/4xAF) 1T- 74.8fps
Doom 3 (1024x768, 2xAA/4xAF) 2T- 73.5fps
1T is 1.2% faster

Avalon's using a 128KB LVL2 chip which probably accounts for the "massive" 3% improvment in one game
-------------------------------------------------------------

Insomniak System: CPU - Winch@2400Mhz | Memory- 2.5-3-3-8 X @200Mhz

Battlefield Vietnam 1T- 106.5FPS
Battlefield Vietnam 2T- 102.1FPS
1T is 4.3% faster*

Source VST 1T- 164.7
Source VST 2T- 158.3
1T is 4% faster*

Super Pi 8M 1T- 6:49
Super Pi 8M 2T- 6:58
1T is 1.4% faster

3dMark03 1T- 9258
3dMark03 2T- 9187
1T is 0.8% faster

Far Cry 1T- 105.2
Far Cry 2T- 103.4
1T is 1.7% faster

Insomniak definity showed the most pronounced differences so far, had a couple games in the 4% range. As and average though he's still less than 2%.
----------------------------------------------------------------------------

blckgrffn System: CPU - 754NC@2600Mhz | Memory- 2-3-3-5 X @195Mhz

Doom3 1T- 103.1
Doom3 2T- 99.2
1T is 3.7% faster

3DMark2001SE 1T- 22543
3DMark2001SE 2T- 21970
1T is 2.6% faster


DUVIE (from anotherthread) System: CPU - Winch@2640Mhz | Memory- u/k timings X @275Mhz

No data Duvie does the math for us: 1T vs. 2T

cinebench 2003 = 0%

superpi 2mb = <3%

prime95 benc 2048K = >1%

Povray 3.5 chess2 = 0%

3dsmax7 rendering = 1%

winrar (file compression) = 7%

the TMPGenc was 1%

"I dont see it ppl...tell me apps this will show up.... "

[TStep]

Insanetek article. Implies about 3% increase in cpu speed plus 10% faster mem clock needed for more or less equivalent performance in their testing.

That's the only real app testing I can recall off hand.

[Zebo]

Isaneteks is flawed as you implied. He uses different mem speeds and processor speeds and then shoots for similar performance. I'm not really even sure the point there??? It certainly does'nt tell you how much would be gained by using 1T or 2T all things being equal. Maybe if I think about it I can interpolate...


EDIT:
No his results are all fubard. To many variables thoughout the three tests. Differnt processor speeds, mem speeds and different mem timings is three variables that make any kind of guestimates impossible.

[TStep]

One thing about that article that I question is the LDT muliplier effect. Common presumption is that it has little to no effect between 3x and 5x, ie not a bottleneck. As shown, 3x to 4x is what I would have expected, but the move from 4x to 5x shows significant performance increases for a few apps. Baffled?

[Zebo]

Ya he's a moron. I think I've posted this before but http://www.xtremesystems.org/forums/...ad.php?t=39559

[TStep]

yup, remember that one.

[TStep]

Came across a Corsair Whitepaper that has some interesting info. They used Doom3 time demo on 3500+ 939 setup. Summary of conclusions pertaining to the topic at hand:

1) 2-2-2-5 1T vs. 2-2-2-5 2T, -3.29% performance loss in Doom3 timedemo, more so than any other timing change.

2) 2-2-2-5 1T vs. 3-3-3-8 2T, -17% in Sisuck Sandra, 90% attributable to CPC change.

Edit: spelling

[Duvie]

Zeb I did a bunch of test like a wekk ago and the performance was in the 1-2% range in some mUltimedia apps and what not......It is not the end all ppl whine about...it is clear they live too much for their beloved sissuck mem scores.....

In some instances if a guy needed to run 2gb and was doing video editing I would say even with 2t I would rather have it then 1gb of 1t...The penalty hit elsewhere will dwarf 1-2%......

I will try to find the link that had my quick test....I bet games as usual will show the biggest advantage since they were usually the one that showed the biggest liking of bandwidth numbers...

[Avalon]

Zebo, I had run a couple gaming benchmarks in my Sempron 2800+ 90nm thread, and found that dropping your memory command rate to 2T was the equivalent of a 400mhz drop in CPU speed at 10x7. At least I think it was about 400mhz. It's in my thread somewhere, but I remember it being definitely noticeable.

[Insomniak]

So basically the real world effect is about the same as slightly looser memory timings? i.e. 1 - 3%?

I'm sorry, but with A64 it just seems that the on-die memory controller has made memory's effect on performance negligable all around - clockspeed, latency, command rate....all of them can be dropped into the basement and the max you're looking at is a 5% performance hit.

[Zebo]

Hmm very conficting opinions even here.... avalon and corsair vs. duvie and insom

Wes makes it sound like 2T is the anti-christ for A64.

May have to do something about this stay tuned. It's a lot of work for something these reveiw sites should be doing when they do thier regular benches every week.

They can't fit one test suite in over all these years while making grandiose proclamations about 1T vs. 2T. :roll: I just took it on faith it was huge, as they say all the time, until I started looking for real numbers and could'nt find anything.

[Insomniak]

Yeah I've been hearing left and right about how 2T will murder A64 performance, but like you, I haven't seen much in the way of benches to back that up...

Of course, AT reviews usually tend to make much ado about memory timings as well, and again, we see a ~3% or less real world difference between 2-2-2-5 and 3-4-4-10, on A64 at least. In most apps.

(blanket statements are dangerous)


Anyway, yeah, this is ripe for a Zebo special - looks like another sticky thread will be coming soon with the eagle at the top.

[charloscarlies]

I seem to remember a pretty significant difference when I ran SuperPI w/ 1T vs. 2T...but I might have to check this again when I get home. I'm tempted to do some real world benches as well because my VNF seems to really hate 1T at any decent HTT speed.

[Frown66]

Zebo, I had run a couple gaming benchmarks in my Sempron 2800+ 90nm thread, and found that dropping your memory command rate to 2T was the equivalent of a 400mhz drop in CPU speed at 10x7. At least I think it was about 400mhz. It's in my thread somewhere, but I remember it being definitely noticeable.

Just some wild speculation here, but Avalon says his test showed a performance hit when changing from 1T to 2T. Other people claim very small % differences.

Avalon's test was on a Sempron, which has half the L2 cache of a Winnie. Maybe the command rate is somehow related (performance-wise) by the amount of L2 cache the cpu has? Less L2=greater the command rate affects performance, more L2=command rate affects performance marginally.

Just speculation on something that may support everyone's benchmarks.

[TStep]

I tend to agree w/ Frown on Avalons case. Less cache means tapping the slower memory subsystem more often. Thereby showing a greater relative performance hit than expected.

[Zebo]

Originally posted by: TStep
I tend to agree w/ Frown on Avalons case. Less cache means tapping the slower memory subsystem more often. Thereby showing a greater relative performance hit than expected.

Very astute guys I would'nt even thought of that.

So It apprears, while the question of how muchdifference between command rate is still open, it gets larger as Cache size get smaller.

[TStep]

You can always look at it this way: What could you possibly do to dual channel DDR that would make it perform theoretically less than a single channel A64 setup? We know from reviews how that compares clock-for-clock with the hinderance of a huge bandwith cut.

I guess we aren't going to be satisfied until one of us finds the time to perform test suite. I agree, I wish a website would perform a Holy Grail type review similar to Anands for memory and the 875 chipset focusing on the fine details.

[TStep]

I'm stuck on 754 (limited to SC), but if I had 939, I'd probably try the test matrix out with all setting stock, changing only:

2x512 SC 1T
2X512 SC 2T
2x512 DC 1T
2X512 DC 2T

That would help show the differing effects of SC cutting the bandwith vs 2T cutting the bandwith. The test matrix would get excessively large if too many tests were included however.

[Leper Messiah]

It doesn't seem to be all that bad, from what I've seen. The larger cache does help a little bit. I went from a CH to a winnie, and for some unknown (to me...) reason, my single 512 of KVR runs 2T. at the same CPU speed, and latencies, just different MB and CPU, 3dmk01 shows maybe 700pts difference. Same GPU too. Oddly, the KVR won't do SPD at 1T. Gonna have to get an OCZ DDR booster and some VX or something, although, for almost the same price, I could get a DFI ultra d and some muskin blue...

[GuitarDaddy]

I did some testing last night after reading this thread and I was shocked.

I changed only the command rate and left everything else the same


SuperPI 8m 2T 6:09 1T 6:03 2.4%
Aguamark3 2T 57985 1T 58156 0.3%
3dmark 03 2T 6872 1T 6908 0.5%
HL2 2T 95.5fps 1T 97.5fps 2.2%
X2 the threat 2T 85.8fps 1T 87.4fps 1.9%
Doom3 2T 84.3fps 1T 85.9fps 1.9%


2% difference in SuperPI and games even though Sandra bandwidth is 1000pts higher (20%), absolutely not noticeable. I like everybody else have been preaching that 2T is a performance killer Doesn't look like the numbers prove that.

*EDIT*

3200+ 0441 1.55v
283x9=2548mhz
Asus A8N-SLI 1004bios
XFX 6600GT
OCZ EL PC3200 platinum rev. 2
ram= 231.7mhz 2-3-2-7 2.85v
HTT multi 3x

[Leper Messiah]

wow. What kinda proc do you have?

So, if one is running a chip with 512k of cache or more, 1T/2T doesn't really matter....hum.

[GuitarDaddy]

Winchester 3200+, The tests were run with CPU @ 2548mhz, mem at 231.7mhz 2-3-2-7

[Zebo]

everybody else have been preaching that 2T is a performance killer

I guilty myself been parroting these reviews w/o proof.:thumbsdown: never again.


Excellent start BTW.

[Avalon]

I will run a few tests immediately. I'll need a few of you Winchester owners to do the same, just to make sure my 256KB L2 cache isn't the cause.

[Insomniak]

If I get some time later today I'll run some tests. Rig in signature - I can provide Data for the Winchy side of things.

Could this be the first Anandtech Forums Collaborative Technology Investigation?

We should form a club