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

[Erssa]

Now I'd like to see the impact on running 4x512Mb 2T compared to 2x512Mb 1T with same timings...

 

[Avalon]

After turning off AF in my control Panel and using the in-game 8x HQ AF for Doom 3, and juicing my system back up to 2.44ghz (305x8) with a 140(7/10) memory divider, effectively giving me 213mhz @ 2-2-2-10 3.1v....Also underclocked my video card to 350/800...system is being weird, and I'm trying to single out what is wrong with it...anyway.

Doom 3 1T - 83.2fps (2.5% faster)
Doom 3 2T - 81.2fps

 

[TStep]

Originally posted by: Erssa
Now I'd like to see the impact on running 4x512Mb 2T compared to 2x512Mb 1T with same timings...

This is my primary interest, having the option of 4x256 mem w/out a massive hit. 256 is dirt cheap plus I have some 256 Mushy Level II stuff I want to put back in action in a 4x config when I move out of my 754 diapers and move up to the 939 big boy's club

 

[Fike]

I have run 512 MB 1T versus 1.5Gb 2T in timed Photoshop benchmarks and gained about 2 seconds in a 2 minute processing task. That is a 1.5 percent increase in performance. The extra memory more than makes up for any minor loss in performance running 2T. I can more easily make up that time by overclocking the processor and keeping 2T timings.

 

[saechaka]

Originally posted by: Erssa
Now I'd like to see the impact on running 4x512Mb 2T compared to 2x512Mb 1T with same timings...

this would be great if someone could add this to the bench.

 

[Insomniak]

Hey Z - my processor revision, current speed, and RAM speed are all in my signature along with everything else. Hope that helps.

I also benched Battlefield Vietnam and Far Cry and added them to my list. Nothing really unusual from the trend there though - just wanted some more real world data.

 

[xenolith]

updated sig

I haven't had time to bump up my TCCD's mem frequency yet to see if I get any measurable performance gain 2T may now afford me. Work can be such a nuisance. :clock::frown:

 

[hippotautamus]

While 1t does offer advantages over 2t, and it's of course preferable, it doesn't justify spending a lot of money on performance ram. The extra hundred dollars is the difference between a 6800GT and a 6800 Ultra

 

[Insomniak]

Originally posted by: hippotautamus
While 1t does offer advantages over 2t, and it's of course preferable, it doesn't justify spending a lot of money on performance ram. The extra hundred dollars is the difference between a 6800GT and a 6800 Ultra

Pretty much any RAM can do 1T. I think you're thinking of latency timings. Command rate is largely a function of the memory controller anyway.

 

[xenolith]

Plus, Winbond UTT chips can be gotten real darn cheap right now.

 

[Gerbil333]

I've been looking for this information for weeks! Thanks, guys.

Are you still considering making your own 1T/1T benchmark thread, Zebo?

 

[Zebo]

Bump for update, sorry been very occupied with external activites....not complete yet. I plan to form some narrative, add mine duvies etc, and excel them...give some averages. ...stay tuned.

 

[Insomniak]

Just a note Zebo: I had the most pronounced differences of everyone so far, but I noticed I also have the loosest memory timings. I would not be surprised at all if other people's tighter timings were making up that 1-2% difference between my 2T scores and their 2T scores.

If I get some time in the next few days, I'll tighten up the timings and re-run some of those tests, and see if the hunch pans out.

 

[Andres3605]

good job Zebo

 

[NINaudio]

Anyone have anything to add to this, I find it quite interesting. I will have to try this later when I get home and see what happens. I guess it would allow me one of two things, a higher OC on cpu, or a higher ram speed at the same timings......

 

[palouse]

If 1T vs 2T is a bandwidth issue, doesn't it make sense to do bandwidth related tests? Member fike mentioned a Photoshop benchmark, but included 512 vs 1536 MB at the same time. Overall, although a synopsis of his test, it makes it look like tripling the amount of RAM loses 1.5% performance.

Instead of games and 3D bmarks, how about burning a DVD, 1T vs 2T, at the same latency timings and MB size? (Not that I can help with that.) Is that a 'real' test of bandwidth? If not, what does test bandwidth? Perhaps the bmarks that A64 'loses' to P4?

 

[Insomniak]

Originally posted by: palouse
If 1T vs 2T is a bandwidth issue, doesn't it make sense to do bandwidth related tests? Member fike mentioned a Photoshop benchmark, but included 512 vs 1536 MB at the same time. Overall, although a synopsis of his test, it makes it look like tripling the amount of RAM loses 1.5% performance.

Instead of games and 3D bmarks, how about burning a DVD, 1T vs 2T, at the same latency timings and MB size? (Not that I can help with that.) Is that a 'real' test of bandwidth? If not, what does test bandwidth? Perhaps the bmarks that A64 'loses' to P4?

We're looking at real world performance, bandwidth or not. We're just interested in what the situation actually is.

Plus, with dual channel, processors generally have all the bandwidth they'll ever need anyway - at least for the time being.

 

[palouse]

Originally posted by: Insomniak

We're looking at real world performance, bandwidth or not. We're just interested in what the situation actually is.

Actually, that's my point. I'd like to know when 1T vs 2T is significant. I noticed that large-block data tests were lacking. I assume that your point is "from previous experience, bandwidth issues are not currently significant, regardless."

 

[Insomniak]

Originally posted by: palouse

Originally posted by: Insomniak

We're looking at real world performance, bandwidth or not. We're just interested in what the situation actually is.

Actually, that's my point. I'd like to know when 1T vs 2T is significant. I noticed that large-block data tests were lacking. I assume that your point is "from previous experience, bandwidth issues are not currently significant, regardless."

Pretty much. I have yet to see any kind of non-server load that can't be shoved through a 6.4GB pipe in a reasonable amount of time.

I suppose some video or 3d rendering could slow it a bit but still...

It seems that A64's on-die mem controller has made bandwidth mostly benign - this is why a higher clocked A64 with downclocked ram will score higher than an A64 at stock with RAM at stock. Performance scales with the processor on this platform, pretty much bar none.

 

[CMC79]

This is a really interesting thread...here's my thought, if cache size is playing a factor, might the performance hit be even less on the 1mb cache chips? I guess someone could disable the L2 cache in the BIOS, and then compare their results with cache on/1T, cache off/1T, and cache on/2T and cache off/2T. Or (with a lot of work), test more than 1 chip, like a 3500 and a 3700.

I would try this myself, but my Venice doesn't arrive till Thursday.

 

[NINaudio]

I have a question for you guys.... My OC is stable at 2860 with a 2T command rate and 2750 with a 1T command rate. I'm at work now and unable to test, but based on the test done here, Theoretically any loss from the 2T command rate should be made up with the extra clock cycles I gained on the CPU, right?

 

[GuitarDaddy]

I would certainly think the extra 110mhz would trump the affect of 2T

 

[blckgrffn]

I would think as little as 40-50mhz would be suffcient to overcome or very much minimize the effect of the 2T timings in most situations.

 

[Avalon]

Hey Zebo, I need you to make a correction in your main post on the front page. My sempy 2800+ is a 256KB L2 cache chip, not 128KB. Only the 2600+ and 3000+ are 128KB L2 cache chips. Thanks.

 

[blckgrffn]

More results from me:

A64 winnie 3000+ @ 1.8 ghz, DDR 400 3-3-3-8, 6800 @ 380/800, Doom 3 1024*768 HQ

1T Doom3:64.6
2T Doom3:64.0

A64 NC 2800+ @ 1.8 Ghz, DDR 333 2-2-2-5, 5700 @ 460/550, Doom3 640*480 MQ
1T Doom3:53.8
2T Doom3:53.6

3dMark2001SE -
1T:12442
2T:12239

Sorry I don't have 3D mark for the winnie, that is my dads comp and I ran out of time. I changed to the DDR400 and that was a big pain in the butt, and the top overclock dropped from 2.16 to 1.9ish. The 4 sticks of DDR333 @ 2T was probably a better set up. I will work with it some more when I get the time, I just hate that stupid MSI motherboard.

I think that Doom3 is VERY GPU bound on the 5700, so maybe that isn't the best benchmark for this test. The 3Dmark2001SE test is probably a better indicator. But obviously it doesn't look good for 1T vs 2T argument, even though I am going to let Zebo do the math

Cheers!
Nat