PDA

View Full Version : Here's a bit of weird RAM timings trivia for you...


Ninjaman09
11-11-04, 08:49 PM
Well, I got my OCZ PC3200EL Platinum today, and I couldn't be more pleased with it. It will do 2.5-3-3-7 2T @ 250MHz no problem, and of course runs fine at stock. Here's the weird thing, though - apparently, the 2T command rate makes a huge difference in the efficiency of the memory controller in Athlon 64s. My Sandra bandwidth benchmarks were showing that the RAM only had about as much bandwidth as it does at 200MHz with 1T, running at about 70% efficiency! So, I changed the ratio to 6:5 and set the timings to 2-2-2-5-1T @ 208MHz. Insanely enough, not only did this bring the bandwidth efficiency up to about 94%, it significantly improved performance in games! I guess they aren't kidding when they say Athlons love good timings. I'm going to try 3-4-4-10 1T @ 250 next to see how that goes, be back with results!

Update: Unstable. This RAM cannot do 1T at 250 MHz. Oh well, it's not that bad :)

Bill The Cat
11-11-04, 09:01 PM
Well, I got my OCZ PC3200EL Platinum today, and I couldn't be more pleased with it. It will do 2.5-3-3-7 2T @ 250MHz no problem, and of course runs fine at stock. Here's the weird thing, though - apparently, the 2T command rate makes a huge difference in the efficiency of the memory controller in Athlon 64s. My Sandra bandwidth benchmarks were showing that the RAM only had about as much bandwidth as it does at 200MHz with 1T, running at about 70% efficiency! So, I changed the ratio to 6:5 and set the timings to 2-2-2-5-1T @ 208MHz. Insanely enough, not only did this bring the bandwidth efficiency up to about 94%, it significantly improved performance in games! I guess they aren't kidding when they say Athlons love good timings. I'm going to try 3-4-4-10 1T @ 250 next to see how that goes, be back with results!

Update: Unstable. This RAM cannot do 1T at 250 MHz. Oh well, it's not that bad :)


Interesting.

For me, setting my timings from 3-4-4-8 to 2.5-3-3-7 doesn't do much on my A64 machine. The benchmarks note only a minute performance gain, and I can't tell any difference in gaming. I'll have to try going down to 2-3-3 and seeing if it makes a big difference.

Ninjaman09
11-11-04, 09:20 PM
Actually, upon looking at it further, the only thing that seems to have any effect on the RAM stability or efficiency is the command rate. My sticks are just not stable at 1T past about 215 or so, though they're rock-solid at 2T even at 250, and probably higher. 2.5-3-3-7 to 3-4-4-10 made no difference on the stability or speed.

The command rate seems to be the most important factor for the Athlon 64 memory controller. I've decided to continue running my RAM at 2-2-2-5 1T 208MHz, because it's simply faster and utilizes more bandwidth than 2.5-3-3-7 2T @ 250MHz. Before anyone asks, yes, I tried 2.8v. Didn't make a difference.

Riptide
11-11-04, 09:23 PM
Kinda odd that 1T isn't stable. Wonder 'sup with that. Could be your motherboard mebe?

Ninjaman09
11-11-04, 09:38 PM
Nope. I figured it out, finally. 1T command is totally stable - with single channel. Once dual channel comes into the picture, it reduces the stability significantly. I tried running 1T in single channel with 250 FSB and 2.5-3-3-7 @ 2.8v and it was rock solid. Are you sure you're running yours in 1T, Riptide? Some of those motherboard menus can be a little confusing. Mine, for example, says "2T Command Rate" with the options Auto/Enable/Disable. Auto, of course, enables it. Disable means 1T for me. Check yours out and let me know how that goes.

GlowStick
11-11-04, 09:50 PM
Hm, my timeings are 2-2-2-10 in dual channel mode at 1T and i have good stability, i also belive anandtech said they were stable at 1T also.

Good luck with it, sorry i dont have more info.

Ninjaman09
11-11-04, 09:54 PM
What speed is your RAM running at? My sticks run perfectly at up to about 210-215 @ 2-2-2-5 1T dual channel.

Riptide
11-11-04, 09:57 PM
I am absolutely certain it's 1T - luckily with my board you simply choose 1 or 2, not enable/disable. :)

What vDIMM are you using? You might also try testing both sticks separately and see what happens.

Ninjaman09
11-11-04, 09:59 PM
edit: Duhhh, I'm retarded.

VDIMM at 2.8v for all the high speed tests, though 2.7 works fine as well.

I did try both sticks seperately, that's what I meant when I was talking about single channel ;)

They worked like a charm.

Well, I got off my ass and did some research. Apparently, these modules are in fact supposed to run (and are tested at, in reviews using Socket 939) at 2T. They are certainly capable of doing 1T at 200 - 215 MHz at 2-2-2, but anything higher really requires 2T. This is a limitation of the dual-channel memory controller on Socket 939 Athlon 64s. I'm too lazy to post links, so just take my word for it ;)

Basically, if you're running your RAM at 250MHz @ 1T, you're lucky.

Ah, here's a great article (http://techreport.com/reviews/2004q3/abit-av8/index.x?pg=1) about the Abit AV8 board. It uses a 3500+ AND OCZ Platinum 3200EL Rev.2, along with an Asus A8V Deluxe for comparison! I was happy, to say the least. The overclocking section in particular is interesting, because those results parallel what I'm seeing in the Doom 3 timedemo. Here are my results:

Doom 3 Timedemo 1 @ 1280x1024 High Quality

2-2-2-5-1T @ 208 MHz, CPU @ 2.5GHz : 92.7fps
2.5-3-3-7-2T @ 250 MHz, CPU @ 2.5GHz : 91.9fps
2.5-3-3-10-2T @ 260 MHZ, CPU @ 2.47GHz : 92.4fps

Neat, huh? Obviously, there's no point in running 1:1 and overclocking the modules when they perform better at stock :D

Once again, it seems that the Socket 939 Athlon 64's memory controller much prefers lower timings / clock speeds for its performance. I'm sure not going to complain - with this new RAM, I get 88.3fps in the Doom 3 timedemo at the settings above with STOCK speeds on everything except the video card. Very nice indeed!

{Sniping}Waste
11-11-04, 10:52 PM
Looks like the older samsung mem on the dims. The newer TCCD is a lot better. Do what you can to keep T1 setting or deal with low effecency. The dim with Micron -5b c i have will do 250 2.5,2,2,5,T1 at 3.2V and a effecency of 98% but its a 3000+ s754 single channle. If your looking for better dims then take a look a Gskill memery with the TCCD Samsung. Most ppl can get 280 with 2.5,3,3,10,T1 2.8V in duel channle and its around $280 for a duel 1gig kit.

Ninjaman09
11-11-04, 11:08 PM
These sticks have the Samsung TCCDs. They are brand new and hand-picked, as I got them through an RMA from OCZ from some Hynix-based PC4000 Gold. Socket 939s have dual-channel memory controllers and are therefore less tolerant of 2T command rates. I'm not at all looking for better DIMMs, because frankly, I don't believe it's possible to get better DIMMs than these. I have yet to see a single person report a successful overclock at that range (280) at T1 on a Socket 939 platform.

{Sniping}Waste
11-11-04, 11:17 PM
These sticks have the Samsung TCCDs. They are brand new and hand-picked, as I got them through an RMA from OCZ from some Hynix-based PC4000 Gold. Socket 939s have dual-channel memory controllers and are therefore less tolerant of 2T command rates. I'm not at all looking for better DIMMs, because frankly, I don't believe it's possible to get better DIMMs than these. I have yet to see a single person report a successful overclock at that range (280) at T1 on a Socket 939 platform.
Here you go!
http://xtremesystems.org/forums/showthread.php?t=44200
http://xtremesystems.org/forums/showthread.php?t=45393

Ninjaman09
11-11-04, 11:33 PM
Thanks for the articles, but the first one does not specify what command rate he's running (not to mention there's no way that overclock was stable for long) and the second is a single channel socket 754. I can run my sticks at 250 2.5-3-3 at 1T in single channel mode, just not in dual channel.

circuitbreaker8
11-11-04, 11:46 PM
Mine are at 250mhz 1:1 1t 2.5-3-3-6 2.8v ( 2x512 ) 0 mem test errors...

Ninjaman09
11-11-04, 11:57 PM
Yes but yours is also single-channel.

I appreciate your input guys, but it looks like I'm out of luck as far as 250Mhz @ 1T goes in dual channel. Something on my system, whatever it is, just can't do it. Time to throw in the towel, I guess, and settle for a 6:5 ratio with 2-2-2 timings. Not too shabby, after all :D

Riptide
11-12-04, 10:45 AM
Humor me for a minute. Get the latest version of memtest86 and build a boot floppy. Set vDIMM to 2.8v, timings to 2.5-3-3-8-1T. Set FSB to 250 and your HT divider to 4x (800MHz).

Then run memtest and see what the results are.