• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Real world effect of timings?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Dexx

Member
Joined
Nov 11, 2002
Location
Perth, WA
Does anyone know of a review which shows stats on the effect of memory timing on performance for s939 A64? I have a winnie that does 280x9. My current memory is only capable of 233MHz 3-4-4-8. I'm interested to know what improvement memory at 233MHz 2-2-2-6 or 280MHz 2.5-3-3-8 would buy me.
 
Name the benchies & settings you'd like to see and I'll run them for you.
 
from what I hear, not much. go with speed instead. Fast RAM with higher timings is better than slow RAM with low timings, IMO (learned that after I already bought this stuff). You can increase the FSB from 200 to 300 without really affecting the 4400 stuff, but trying to do that on my 3200 requires VERY lax timings. plus, you can lower the timings independantly and directly, whereas increasing the RAM speed affects the whole system and cannot be done directly.
 
OK. Some results. First I wasn't able to conduct the tests I REALLY wanted to because it seems I have rotten luck of late. I ordered 2x512 sticks of twinmos TMSP which were supposed to be here yesterday, but were delayed. Well they got here today and unfortunately are NOT winbond UTT. Basically they stink and I'm reselling them. The results I was able to get have me readily whipping out the CC to try again though. All results are compared to the initial 200x10@2-2-2-6 1T. Observe:
 

Attachments

  • RESULTS_640x240.jpg
    RESULTS_640x240.jpg
    46.6 KB · Views: 244
Last edited:
I added the prime95 benchmark tests because I've always found it to be a quick and dirty, but accurate measure of performance. It didn't let me down. What these results say to me is OCZ VX is THE memory to get for the A64 and if you're going for the DFi nForce4 boards then get that or take your chances on getting winbond UTT with twinmos/mtec. The performance of 250-260 Mhz memory(DDR500-520)@2-2-2-x 1T timings should be spectacular and I'm really anxious to test some now. Here are the individual prime95 results which I averaged and included above. If you want to inspect these results more thoroughly I've hosted the spreadsheet here as well as a larger image.

http://webpages.charter.net/timonex/Results.xls
http://webpages.charter.net/timonex/RESULTS.JPG
 

Attachments

  • P95_640.JPG
    P95_640.JPG
    46 KB · Views: 245
Last edited:
Interesting. Thanks TimoneX. So given an A64 running at 2GHz, you have the memory running at 200, 223, and 251. In 3dmark 2001, 200:2-2-2-6 is about the same as 223:2.5-3-3-6. Suprisingly 251:2.5-3-3-6 is only slightly faster. Given that low latency ddr400 is usually cheaper than mid latency ddr500, it looks like the ddr400 is the best choice.
 
TimoneX said:
I added the prime95 benchmark tests because I've always found it to be a quick and dirty, but accurate measure of performance. It didn't let me down. What these results say to me is OCZ VX is THE memory to get for the A64 and if you're going for the DFi nForce4 boards then get that or take your chances on getting winbond UTT with twinmos/mtec. The performance of 250-260 Mhz memory(DDR500-520)@2-2-2-x 1T timings should be spectacular and I'm really anxious to test some now. Here are the individual prime95 results which I averaged and included above. If you want to inspect these results more thoroughly I've hosted the spreadsheet here as well as a larger image.

http://webpages.charter.net/timonex/Results.xls
http://webpages.charter.net/timonex/RESULTS.JPG

damnit, the pic is quite evident...

timings DO matter for athlon 64...
 
PhobMX said:
damnit, the pic is quite evident...

timings DO matter for athlon 64...

Yes and additionally just as I suspected Sandra is a lying beezotch. She hands out these huge beautiful scores to go along with rising memory speeds but they don't jive with real world results.

PhobMX said:
what about ocz vx???

Yes that was with TCCD. I think VX or Twinmos speed premium(if you're a gambler) is the fairly obvious choice. BH5 is also fine with 256x2 but that's just not enough memory for most people any more, particularly with XP64's release around the corner. In all fairness I did not get to finish the testing, but it looks pretty clear that 2-2-2-x 1T around 250-270Mhz should be superior or at least equal to TCCD at 300Mhz 2.5-4-3-x 1T. Of course VX and TMSP require a minimum of 3.2v to offer this performance, but that's where DFI comes in with their nF4 boards capable of supplying 4v vdimm stock.
 
TimoneX said:
it looks pretty clear that 2-2-2-x 1T around 250-270Mhz should be superior or at least equal to TCCD at 300Mhz 2.5-4-3-x 1T.
I'd go further to say that 2-2-2-x 1T at 233MHz would give 300MHz 2.5-4-4-x 1T a run for its money.
 
Dexx said:
I'd go further to say that 2-2-2-x 1T at 233MHz would give 300MHz 2.5-4-4-x 1T a run for its money.

I agree and disagree. The bandwidth of TCCD at 300mhz on a 939 setup is close to 8gb/sec. The 233mhz would probably be around 6gb/sec. That is a NOTICEABLE difference in anything that relies on high memory bandwidths. The access times are quicker on the 2-2-2-x-1t, but over all, I think it would be hard to beat.

Why cant we just have ram that does 350mhz at 2-2-2 with like 2 volts? lol.

-Collin-
 
I'm not so sure. I'd like to see more tests, and I intend to run them myself when I have the appropriate memory sticks. It seems to take about a 25% increase in frequency increase to overcome the difference between DDR400 2-2-2-6 & DDR400 2.5-3-3-6 which is substantial and more than I would've expected. At the same time there's something else at work here. If you study those results carefully you'll notice the BW difference between 200 2.5-3-3-6 and 223 2.5-3-3-6 is 263Mb/s. That's an improvement of 4.73% for a speed increase of 11.5%, not unreasonable. However if you look at the BW gained between 223 2.5-3-3-6 & 251 2.5-3-3-6 there's only a 3.26% improvement for a 12.56% speed increase. It seems the laws of diminishing returns are setting in already at this point. Even worse is the fact that Sandra's BW is clearly showing this speed increase in the best possible light but if you compare the improvements in "real world" performance the returns for increased memory speed look even lower. The same 11.5% improvement in memory speed from 200 to 223 in 2001se yielded a respectable 2.32% improvement. Going from 223 to 251(+12.56%) gained a paltry .70% in 2001se. It'll be interesting to see first hand how real world memory performance scales with 2-2-2-x 1T timings, but I'd be willing to bet they'll scale in a more linear fashion up to at least 250Mhz.
 
Well, then this is certainly good news for the UTT I have coming in the mail. Very nice comparison you did there....always good to see numbers.

-Collin-
 
Just got back from 3 different circuit cities. They have that deal on Kingston valueram that ends up being $45 for 512Mb with some rebates. Anyway I waded through about 100 different boxes and not one of them appeared to be winbond UTT unfortunately. Oh well it was a nice outting. :)
 
Back