4x512MB WINBOND BH-UTT 520MHz 2-2-2-5 2T SPI32M!

[kakaroto]

I was wondering what 4 times 512MB in on my venice system will do, i was very suprised of the awesome results.
Offcourse those 4 modules can do seperately 250-260Mhz 3.3~3.4v.

Venice 3000 (week 17)
DFI nF4 SLI-D (BIOS 623) @5V line
4x512MB G.SKILL PC3200 GH (/w 80mm fan)

I tried 250Mhz 1.5-2-2-5 2T with some fast settings :p:

http://www.xs4all.nl/~famchow/Memory/Gskill%20PC3200GH/4x512/2048MB_GS_250MHz_15225_2T_SPI32M.JPG

Now at 260MHz 2-2-2-5 2T with also fast settings but now 256 cycle time and 8ns max asyn cause I want to use 3.4v not more.

http://www.xs4all.nl/~famchow/Memory/Gskill%20PC3200GH/4x512/2048MB_GS_260MHz_2225_2T_SPI32M.JPG

:woot: :woot:

[DeltZ]

very nice :)

too bad that you can't do anything about 2 T though.

[ZeroX]

Nice results m8, the problem of 4modules is the 2T :(

[muaddib]

520Mhz? You mean DDR520? 520Mhz is definitley jaw-dropping. :D
j/k, very nice results as it is, 2x1GB and 1T of this stuff could be uber nice!

[kakaroto]

Nice results m8, the problem of 4modules is the 2T :(

yeah indeed, but this is faster then 2x1GB @270-80 3-4-4-8 1T. In games there is little FPS drop and the bandwidth, but for gamers I think this is the best solution of 2G gaming with low latency. Let's wait for 1G 2-2-2-5 modules :D

@muaddib : you know what I mean right? :p: 520MHz (DDR1040), yeah :banana:

[guess2098]

kakarto this is great!
but i don't think 3.4v 24/7 is a good idea tho.....^^"
however the performance is unbeatable

[Onepagebook]

kakarto this is great!
but i don't think 3.4v 24/7 is a good idea tho.....^^"
however the performance is unbeatable

nono..that's not unbeatable..Timings too loose.
Here is the one can say Close to unbeatable but not quite:D

http://img399.imageshack.us/img399/3028/ddr520gh2dn.jpg

or take this as challenge:D, I don't think there are more than 3 people did with 14 multiplier with over 265htt before.

http://www.ocxtreme.org/onepagebook/DIinsulation/32mpirec.jpg

[kakaroto]

Kev, I need a FX-57 :D I will tighten some for ya !!

Trfc at 12 works too, forget that.
I try the 270MHz...wish me luck...SPI32M..pff you make it difficult for me :p:

[Onepagebook]

fx57 ? sigh, no more great batch. they 're all gone to history :D

[uwackme]

Why does A64 tweaker memory clock say 166Mhz not 200Mhz?

My 250Mhz x11 4x512M VX (CH) @ 3.2V does SuperPi32 in 27m09s.

I know my OCZ new BH5 (3200Gold) does 255Mhz 2x512, need to get 2 more to see what 4x512M will do.

Going to wait till EMC's Vtt fix is installed before bothering to rate my ram.

[guess2098]

i am talking about 2G smart one :stick:


nono..that's not unbeatable..Timings too loose.
Here is the one can say Close to unbeatable but not quite:D

http://img399.imageshack.us/img399/3028/ddr520gh2dn.jpg

or take this as challenge:D, I don't think there are more than 3 people did with 14 multiplier with over 265htt before.

http://www.ocxtreme.org/onepagebook/DIinsulation/32mpirec.jpg

[Onepagebook]

i am talking about 2G smart one :stick:


Even with 2g, timing can be adjust that well. apparently you have no idea what's the big deal Gskill GH is so different. Go figure it out. :slap:


Edit: Oh sorry I though you only know how to PLP :eek: :D

[guess2098]

Even with 2g, timing can be adjust that well. apparently you have no idea what's the big deal Gskill GH is so different. Go figure it out. :slap:


Edit: Oh sorry I though you only know how to PLP :eek: :D

haa! better than someone without LP ; open every FX and sent it back to friends :woot: :slapass:
anyways ^^" no more BX with kev :D

good 2G btw kakaroto, i think u can add a bit volt to 3.5v and pull up a bit ^^

[dnottis]

My results show that 2-2-2-5 4dimms x 512 2T is faster than 2 x 1GB 3-4-4-8 1T as well. Sandra shows 6 gb/s vs 7.2 gb/s but trust me in real world apps even with 6gb/s bandwidth it's faster. So what does this say about throughput.... even 6 GB/sec is all that is needed for amd64, not even close to saturating it... 1:1 with high HT is NOT needed if you can keep the timings tight!

[Major]

My results show that 2-2-2-5 4dimms x 512 2T is faster than 2 x 1GB 3-4-4-8 1T as well. Sandra shows 6 gb/s vs 7.2 gb/s but trust me in real world apps even with 6gb/s bandwidth it's faster. So what does this say about throughput.... even 6 GB/sec is all that is needed for amd64, not even close to saturating it... 1:1 with high HT is NOT needed if you can keep the timings tight!

I'm going to give 4x512mb in my gaming rig another try next week :D

[Aleman]

Although 4x260 is nice, it (IMHO) is not really impressive. I have had 4x270MHz with tight timings and they were SP32M stable, although with 3.6~3.7V so, that aint cool for 24/7 (nor for benching coz of the 2T).

Too bad :(, anyway i think that many users cant 4x512 @ high clocks @ tight timmings coz of the MC's. So its not really a memory issue.

[Onepagebook]

haa! better than someone without LP ; open every FX and sent it back to friends :woot: :slapass:
anyways ^^" no more BX with kev :D

good 2G btw kakaroto, i think u can add a bit volt to 3.5v and pull up a bit ^^


:woot: it had better I opened that up,
even though I opened it up then so what? better than someone took 30000NT(about 900USD) for a single hardware beta testing fee. why don't you just zip and go watch college football. that be more insteresting than listen your :banana: :banana: here.
Poor CBB PLP guy here@ XS, damn poor
Oh and you should learn more photshop from the old mand and dog before you come to here and BS every single day. Sad.. :D :woot:

[eva2000]

My results show that 2-2-2-5 4dimms x 512 2T is faster than 2 x 1GB 3-4-4-8 1T as well. Sandra shows 6 gb/s vs 7.2 gb/s but trust me in real world apps even with 6gb/s bandwidth it's faster. So what does this say about throughput.... even 6 GB/sec is all that is needed for amd64, not even close to saturating it... 1:1 with high HT is NOT needed if you can keep the timings tight!
how about against 2x 1GB 3-4-3-7/8 1T ?

[Major]

how about against 2x 1GB 3-4-3-7/8 1T ?

how about against 2x 1gb 3-3-3-7 1T @ 290mhz :D

[guess2098]

how about against 2x 1gb 3-3-3-7 1T @ 290mhz :D

lol
how's the cpu? tested yet?

i think 260 2-2-2-5 2T 2G still a bit faster than 290 2G 3-3-3-7 1T
but, not sure^^"
i had a pair of UCCC can do 290 3-3-3-7 2.6v
i am waiting for another pair of utt bh-5 so i can compare the apple and orange XD

edit----
sry is 290 3-3-3-"8" 2.6v

[xs64]

how about against 2x 1gb 3-3-3-7 1T @ 290mhz :D
I prefer tight timings than high bus

4 x 512Mb @ 250 2-2-2-5 is great for me than 2 x 1Gb @ 290 3-3-3-8

[kakaroto]

Even with 2g, timing can be adjust that well. apparently you have no idea what's the big deal Gskill GH is so different. Go figure it out. :slap:


Edit: Oh sorry I though you only know how to PLP :eek: :D

True , but this really depends on the memorycontroller.
Most controller, like Venice like 2x512 better then 4x512.. so there is a difference.
Espessialy when you test 4 times PC4800 FF, I cannot hit DDR600 with 4x512 on this Venice with 2T :)

[krampak]

Just got 4x512MB of mushkin Level II, will play with it later :D (by now they are running at 220Mhz 2-2-2 2.9v).

[funkflix]

Nice results! I can't get working mine with 254MHz, even not with 3,5V! Now they running 233MHz 3,0V. I must try it a little harder, play with timings and that. We will see! :)

2x512MB Mushkin Lvl2 + 2x512MB Kingston.

[Tony]

Am I missing something here.
kakaroto benched 260x9 on a 3000+ venice with 2gigs BH5 UTT 1.5-2-2-5 and recorded 34mins17s he benched again at 2-2-2-5 and got 33mins13s, I benched at 3-4-4-9 260x9 same 32m test and recorded 32mins 26s, thats nearly a full minute faster and you all say 2gigs of BH UTT at 2T is faster???

http://homepage.ntlworld.com/a.leach80/2gig%20testing/2gig260x9.JPG
http://homepage.ntlworld.com/a.leach80/2gig%20testing/2gig260x9un.JPG

Timings used, pretty damn slack.
http://www.xtremesystems.org/forums/attachment.php?attachmentid=36661&stc=1



Run your bench unbuffered and take the fsb out the picture, im running 2gigs over 2dimms 1T any day over 4x512 2t even with tighter timings, 2.8V is also a lot less strain on the motherboard and the memory controller.

[-=DouglasteR=-]

Can i have some sucess whit 4x512 ocz VX on DFI ultra D ?! :confused:

Im thinking to switch from my 2x512 Bh5 Corsair who do 262 1.5-2-2-5 on 3.4.

Worth the pennies !? :confused:

[kakaroto]

Bigtoe, I was doing all stuff while Super32M is running. I was not doing for the time. I will do a clean run again. Let you know.

[uwackme]

Why does BigToe's memclk say 200Mhz, and my memclk says 200Mhz, but Karato's is 166Mhz.... doesnt this mean he is on a divider and the memory isnt running at the speed he says? Isn't it 5/6's of 260Mhz or... 216Mhz?

Mine is 250Mhz x 11, with 4x (FOUR sticks of) 512M VX(CH) @ 3.20V, 8,2,2,2.0 2T and I get SuperPI 32M... 27m26.125s; when using only 2 sticks at 1T I get 26m.34.187s, that is all at 250Mhz x11 1:1 2750Mhz.

BTW, this shows the 1T vs 2T difference at approx. 4%... 1m out of 26m, or about 1/26th difference in speed.

[kakaroto]

Why does BigToe's memclk say 200Mhz, and my memclk says 200Mhz, but Karato's is 166Mhz.... doesnt this mean he is on a divider and the memory isnt running at the speed he says? Isn't it 5/6's of 260Mhz or... 216Mhz?

Mine is 250Mhz x 11, with 4x (FOUR sticks of) 512M VX(CH) @ 3.20V, 8,2,2,2.0 2T and I get SuperPI 32M... 27m26.125s; when using only 2 sticks at 1T I get 26m.34.187s, that is all at 250Mhz x11 1:1 2750Mhz.

BTW, this shows the 1T vs 2T difference at approx. 4%... 1m out of 26m, or about 1/26th difference in speed.

What are you talking about? You mean the REFRESH RATE = 166MHz 1.95us?
That is not the Memory devider, that is the memory refresh rate.
My CPU/MEM is running 1:1 offcourse..

[Tony]

Your looking at the TREF bud, easy mistake to make

[Tony]

I have been thru all this before with the guys on 2cpu.com who proved 1t is key, just imagine if i had run 3-3-3 or 3-3-2 at 260, oh man it would really eat 2t UTT for dinner.

Lots of ram is a huge load on a memory controller, high vdimm is not best in this situation really. I can run 4x512 here but I see no point when 2x1024 is just as good or even faster with slacker timings. At 260 I get 3400MB/s or so in memtest86 with 2x1024, what you all getting with 4x512 2t 260x9 with a venice?

[uwackme]

I thought Tref was based off whatever "memclk" was selected, so 200Mhz memclk, with 1560 Tref yeilds...3.9us. 3120 yeilds 7.8us. And it was the reason why when using "dividers" for the memory the # you choose for Tref had to change. So selecting 166Mhz or 180Mhz memclk (engaging a divider) would require a different Tref value to achieve the same 3.9us or 7.8us value.

Guess Im just confused.

[eva2000]

my experience on DFI NF4, TREF is just fixed to what you set it bios regadless of memclk

[-=DouglasteR=-]

So the only flaw to 4x512 or 4x1gb is the 2t !? :woot:

Can i go for 4x1gb Crucial without fear ?! :confused:

[eva2000]

4X1Gb still needs 2T as well

[-=DouglasteR=-]

4X1Gb still needs 2T as well

But the overclockbility remains the same !? :confused:

[eva2000]

depends on

1. modules
2. cpu mem controller
3. board/bios
4. luck :)

but there's no way 4x 1GB will oc same as 2x 1GB

[-=DouglasteR=-]

depends on

1. modules
2. cpu mem controller
3. board/bios
4. luck :)

but there's no way 4x 1GB will oc same as 2x 1GB

1 - 4x1gb Crucial Ballistix Z503
2 - Winnie 3000+ that handle 257 1.5-2-2-5
3 - Ultra-D Bios !!??!
4 - :(

Thanks for the help :D

[kakaroto]

I have been thru all this before with the guys on 2cpu.com who proved 1t is key, just imagine if i had run 3-3-3 or 3-3-2 at 260, oh man it would really eat 2t UTT for dinner.

Lots of ram is a huge load on a memory controller, high vdimm is not best in this situation really. I can run 4x512 here but I see no point when 2x1024 is just as good or even faster with slacker timings. At 260 I get 3400MB/s or so in memtest86 with 2x1024, what you all getting with 4x512 2t 260x9 with a venice?

You got a strong point. How about 3.3v with 4x512MB on a X2 processor, would that change the situation a little bit?
I am now running 4x UTT at 250x10 on a X2 3800+ E6.
Next week I will test 278x9 (2.5GHz) with 3-4-4-8 1T 2x1GB.

[dnottis]

how about against 2x 1gb 3-3-3-7 1T @ 290mhz :D

how about against 2x 1GB 3-4-3-7/8 1T ?


Unfortunately my PC4000 cannot do 3-4-3-8, 3-4-4-8 @ 253 Mhz is where it is right now. I cannot tighten the timings one bit or shes dumping hundreds of errors in memtest. So all I have is 2 x 1024 3-4-4-8 1T @ 253 vs 4 x 512 2-2-2-5 @ 253 3.1v to compare.

Once again, here is the comparison chart once our testing was complete. Yes 1T 2 x 1024 shows more bandwidth, but look at the real world tests.

http://www.3dxtreme.net/images/reviews/OCZ%20EL%20DDR%20PC-4000%202048MB%20Gold%20Edition%20Dual%20Channel/graphs/results.jpg

[drcrawn]

4X1Gb still needs 2T as well

s940 doesnt have this problem :)

I don't think s939 can ever do 4x512 at 1T. I'd like to see one screen shot of someone doing it with stability at any speed or with any timings they want.

[dnottis]

s940 doesnt have this problem :)

I don't think s939 can ever do 4x512 at 1T. I'd like to see one screen shot of someone doing it with stability at any speed or with any timings they want.

well since none of the 939 chips yet support 1T with more than 2 dimms, you'll be hard pressed to get any sceenshots.

[Tony]

You got a strong point. How about 3.3v with 4x512MB on a X2 processor, would that change the situation a little bit?
I am now running 4x UTT at 250x10 on a X2 3800+ E6.
Next week I will test 278x9 (2.5GHz) with 3-4-4-8 1T 2x1GB.

Point is as soon as you hit 2t you lose...period, simple as that.

Also the more vdimm you apply to low voltage CPU's the more chance you have of braking down the diodes on the cpu that block the high voltage from the CPU core, after a while you just end up pumping current into a cpu that doesn't need it.

Only work around is increase the vcore also to stengthen the diodes but the bi product is more heat. With 2 modules its bearable, when using 4 with high vdimm you are really pushing it...in bigtoe land its a big NONO ;)

[kakaroto]

Point is as soon as you hit 2t you lose...period, simple as that.

Also the more vdimm you apply to low voltage CPU's the more chance you have of braking down the diodes on the cpu that block the high voltage from the CPU core, after a while you just end up pumping current into a cpu that doesn't need it.

Only work around is increase the vcore also to stengthen the diodes but the bi product is more heat. With 2 modules its bearable, when using 4 with high vdimm you are really pushing it...in bigtoe land its a big NONO ;)

Yes 1T wins over 2T, but not equal 2-2-2-5 vs 3-3-3-3. If those UCCC do 280 3-3-3-3 1T that is indeed faster. Have you tested game performance? (1T vs 2T)
Overall, there is a lot of stress when using 4x512 modules. It is harder to get it 100% stable and sometimes strange freezes popup.

about the voltage trough the CPU etc. Thanks dude, respect!

[drcrawn]

Yes 1T wins over 2T, but not equal 2-2-2-5 vs 3-3-3-3. If those UCCC do 280 3-3-3-3 1T that is indeed faster. Have you tested game performance? (1T vs 2T)
Overall, there is a lot of stress when using 4x512 modules. It is harder to get it 100% stable and sometimes strange freezes popup.

about the voltage trough the CPU etc. Thanks dude, respect!

3-3-3-x @1T is still faster than 2-2-2-x 2T regardless of memory speed.

I can test the difference between 1T and 2T with 4x512 for anyone who wants to see the numbers. :p: It's a big hit IMO, not sure how you could prove that in gaming environment though.

edit: ran some 01 and 05, all settings identical only difference is 1T and 2T timing. x850 @594/614.

http://img366.imageshack.us/img366/4262/1t3d017iy.th.jpg (http://img366.imageshack.us/my.php?image=1t3d017iy.jpg)http://img366.imageshack.us/img366/6204/2t3d018zi.th.jpg (http://img366.imageshack.us/my.php?image=2t3d018zi.jpg)

http://img366.imageshack.us/img366/8935/1t3d058bj.th.jpg (http://img366.imageshack.us/my.php?image=1t3d058bj.jpg)http://img366.imageshack.us/img366/2748/2t3d057fl.th.jpg (http://img366.imageshack.us/my.php?image=2t3d057fl.jpg)

[Tony]

you have the ram there bud, I did all these tests ages ago, 2-2-2 at 250 2t 4x512 does not beat 3-3-3 at 250 1T 2x1024 in games. Plus your ramming voltage and stressing the board, ram and cpu. going by Pi times 3-4-4- at 260 beats 2-2-2 2t at 260 so its probably close to a winner again for the 1024dimms.

I helped the opty croud with some dimms that did 3-3-3 at 250 a while back, this is where all this testing came from as they would not settle for 2t at any cost, with LOTS of ram installed.

[drcrawn]

opterons (s940) don't use 2t timing at all, so that's not an issue

and yes, 2-2-2 at 250 2T does not beat 3-3-3 at 250 1T

[dnottis]

Unfortunately, my sticks wont go any tighter than 3-4-4-8....so in my case 2T 4x 253 Mhz 2-2-2-5 IS faster than 2 x 1024 1T 3-4-4-8. 3-3-3-8 may be faster, I wish I could test it...

[kakaroto]

Thanks dcrawn :) I will test it by myself too.

[Dumo]

How bout this :) ..270 is easy...

4X512 Gskill GH and Mushkin Redline @ 3.6V/2T and Ultra-D with bios 510-2... :D
http://img236.echo.cx/img236/7225/fx55bn51iw.jpg
http://img62.echo.cx/img62/3585/screenshot0792nq.jpg
http://img300.echo.cx/img300/5402/screenshot0626aa.jpg
http://img300.echo.cx/img300/3171/screenshot0605rl.jpg

32m...

http://img287.echo.cx/img287/5320/screenshot0677rc.jpg

[Tony]

How bout this :) ..270 is easy...

4X512 Gskill GH and Mushkin Redline @ 3.6V/2T and Ultra-D with bios 510-2... :D
http://img236.echo.cx/img236/7225/fx55bn51iw.jpg
http://img62.echo.cx/img62/3585/screenshot0792nq.jpg
http://img300.echo.cx/img300/5402/screenshot0626aa.jpg
http://img300.echo.cx/img300/3171/screenshot0605rl.jpg

32m...

http://img287.echo.cx/img287/5320/screenshot0677rc.jpg

Apples to apples m8, get yourself a 3000+ venice and run 270 or even 260x9. FX55 have the best memory controllers so overclcoking ram is a little bit easier. our tests back at OCZ have shown an FX can be anything upto 25fsb better with ram over an el cheapo venice or winchester.

[uwackme]

Whats with the 1T vs 2T nonsense?

Dcrawn, your own figures show a "clear" difference.... a 2-4% difference, just like my apples/apples 1T 2T SuperPi32M results... 4%.

I would hardly call 4% an earth shattering performance hit.

Given most people have sticks of 512M laying around, some really nice ones too, claiming that the be-all end-all of life should be RUN OUT and GET 1G sticks, is just BS.

No way the $$ of $350 2x1G Ballistix is justified by a 4% boost. Sure for people choosing memory from scratch maybe, but certainly NOT for most of us with BH5 and VX up the wazzoo already.

As for the 1T requirement. That is AMD being conservative. They are requiring it based on the expected performance of the "average" 512M stick of ram. Since when are WE running the average day stick :fact:

I'd like to find out what is getting triggered when I try to FORCE 1T with the 4x512M. You'd think you could "back into" it, say start at 166Mhz and creep up in 1T trim, and such. But it appears "something" magically FLIPS OUT at the 200Mhz point. To me it smacks of bios code trying to play NANNY instead of doing what I TELL IT TO DO :nono:

Wouldnt common sense dictate SOME sticks would do 1T higher than others, and SOME of us could get to 205...210...220... with 1T and 4x512M? But some kind of switch flips at the 200Mhz point. Sigh. AMD/bios/something is stopping us from even testing the waters on this range of settings.

It aint making no sense. The 939 pin is a 128bit bus, 2 64bit controllers. So all the load on either is simply 2 sticks of 512M....now if my NForce2 PoS can do 250Mhz 1T with 2 512M sticks on it, why cant each channel of the superduper A64? Something smells like a deliberate holdback to me.

Meantine, this thing despite all its "faults" feels like a FKing Ferrari :woot:

[drcrawn]

I'd be bummed too if I had to run 2T and I felt it was an artificial barrier. Since 2gigs is fast becoming the norm for gamers/benchers maybe it's time someone started asking AMD some questions about this.

I have a feeling the memory controller of a s940 chip acts very differently in conjunction with registered ram, than the controller of a s939 chip using unbuffered ram.

The first place to start looking is here :

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26094.PDF

BTW, your 4% is a bit misleading, it terms of pure memory bandwidth it is close to 18% drop by using 2T according to my Sandra tests.

[uwackme]

I keep trying to point out... SuperPI32M... 3Dmark's...real game play... those are REAL things, Sandra's benchmark means :banana::banana::banana::banana:.

It's 4%... nothing misleading about it, that is the impact 1T vs 2T has on REAL world use of the A64. Sandra is misleading, and frankly meaningless beyond a measure of how changes to the system affect purely the memory performance.

[TheMeatFrog]

I agree, Sandra is worthless when it comes to real world performance. SuperPI and the 3DMarks is all that matters to me. :)

[Major]

Well this whole 1T vs 2T argument is sorta a mute point, obviously there is no winner or the argument wouldn't continue. Sorta like the low latency vs high fsb debate :D It's like trying to get everyone to agree on a favorite color ! LOL

Everyone has their opinion and attitudes about running 2x1gig vs 4x512mb, here's mine:

If your upgrading to 2GB and already have 2x512mb sticks then adding 2 more for 2GB is an option if your on a budget or if you enjoy torturing your hardware with 3.5vdimm. But I would rather sell the 512mb sticks and invest in 2x1gig sticks so that this time next summer when everyone is talking about getting 4gb you will already have a head start :D LOL

[arctic-k20]

just out of curiosity - if we could see some gaming benchmarks of 1t vs 2t just to see how much this 4% actually affects games and programs we actually use day to day - that would be great!!

i am running 4x512 of ocz plat 4200 and just for kicks i am running them @ 200mhz 2-2-5-2 2t and all the secondary timings are tight and my gaming scores are better than running them @ 255 with loose timings (x2-3800)....

at this point i would be happy to run 4x512 of vx with tight timings in the 250~260 range - ya the synthetic benchmarks look crappy but real world looks great!!

[dnottis]

BTW, your 4% is a bit misleading, it terms of pure memory bandwidth it is close to 18% drop by using 2T according to my Sandra tests.


Sandra means nothing. It's THEORETICAL or "CALCULATED BANDWIDTH"!


Please, read the chart below... the 2T 4 sticks beats 1T in REAL WORLD apps regardless of the over 1GB/sec difference in performance shown in sandra.

http://www.3dxtreme.net/images/reviews/OCZ%20EL%20DDR%20PC-4000%202048MB%20Gold%20Edition%20Dual%20Channel/graphs/results.jpg

[EnJoY]

I'm torn really. 2x1GB Ballistix and so sexy, and 2x1GB Ballistix will be even more sexy once AMD gets their memory controllers straight. However, I've stumbled on some phenominal 512MB Mushkin sticks, so I'm debating whether or not to just go a 4x512MB setup or what.

Offtopic: Dnottis, congrats on your 1,000 post! :toast:

[drcrawn]

Sandra means nothing. It's THEORETICAL or "CALCULATED BANDWIDTH"!


Please, read the chart below... the 2T 4 sticks beats 1T in REAL WORLD apps regardless of the over 1GB/sec difference in performance shown in sandra.

http://www.3dxtreme.net/images/reviews/OCZ%20EL%20DDR%20PC-4000%202048MB%20Gold%20Edition%20Dual%20Channel/graphs/results.jpg

Sandra means what it means, like I said before, how the drop in bandwidth applies to real world scenerios will vary.

I am not arguing about anything here...

EDIT: Ok, I was misreading the chart, it makes sense.

Frankly the argument doesnt concern me at all, as I can run 4x512 1T on my platform. I was only pointing out that IMO, 1T is important enough to override the need for low latency timings. In other words for 2gigs, running 3-3-3 @ 1T is preferable over 2-2-2 @ 2T at any HTT.

[drcrawn]

what are the timings for the last two 2x1024 sets?

[drcrawn]

just out of curiosity - if we could see some gaming benchmarks of 1t vs 2t just to see how much this 4% actually affects games and programs we actually use day to day - that would be great!!

i am running 4x512 of ocz plat 4200 and just for kicks i am running them @ 200mhz 2-2-5-2 2t and all the secondary timings are tight and my gaming scores are better than running them @ 255 with loose timings (x2-3800)....

at this point i would be happy to run 4x512 of vx with tight timings in the 250~260 range - ya the synthetic benchmarks look crappy but real world looks great!!


Which benches do you want me to run? I did 3dmark01 and 05 in this thread somewhere already. It was over a 1000 points in 01 and about 60 points in 05 I think.

BTW, anyone with 2x512 can easily switch back and forth between 1 and 2T and do their own tests...

[EnJoY]

what are the timings for the last two 2x1024 sets?

3-4-4-8 1T

[dnottis]

Offtopic: Dnottis, congrats on your 1,000 post! :toast:

Thanks! I haden't noticed I guess I'm officially an XS nerd now! :toast:

:woot:

[dnottis]

Frankly the argument doesnt concern me at all, as I can run 4x512 1T on my platform. I was only pointing out that IMO, 1T is important enough to override the need for low latency timings. In other words for 2gigs, running 3-3-3 @ 1T is preferable over 2-2-2 @ 2T at any HTT.


The timings will make a huge diff. My OCZ pc-4000 will not do 250 at anything tighter than 3-4-4-8 1T. up to 274 is memtest stable, 3-4-4-8.

If you have 2 x 1024 that can do 3-3-3-8 Im sure the results would be different.

[Major]

Thanks! I haden't noticed I guess I'm officially an XS nerd now! :toast:

:woot:

Dean, you've always been an XS nerd :D

[dnottis]

Dean, you've always been an XS nerd :D

:D Coooooool!

[EnJoY]

If you had four sticks total of same brand BH5, two were 512MB double-sided, and two were 256MB single-sided, would you still have to run 2T?

[bachus_anonym]

If you had four sticks total of same brand BH5, two were 512MB double-sided, and two were 256MB single-sided, would you still have to run 2T?
Yes.

[EnJoY]

Awww...why?

[drcrawn]

Can someone do me a big favor and run Sandra @ 200mhz 1:1 @ 2-3-2-6 1T please with their s939 system...and share the results...thanks. :)

[bachus_anonym]

Awww...why?
As long as there are 4 DIMMs populated, with at least 2 double-sided modules (most of 512MB sticks), 2T kicks in. That's the beauty of A64 S939 mem-controller :(

[EnJoY]

I thought it had to be 8 banks? Awww, that's too bad.

[drcrawn]

please please please someone run Sandra at 200mhz 2-3-2-6 1T for me...please I beg of you...

edit: well I guess no one is going to do it, forget it I'll just have to start a thread over a simple question in the AMD section :rolleyes:

[EMC2]

I'd like to find out what is getting triggered when I try to FORCE 1T with the 4x512M. You'd think you could "back into" it, say start at 166Mhz and creep up in 1T trim, and such. But it appears "something" magically FLIPS OUT at the 200Mhz point. To me it smacks of bios code trying to play NANNY instead of doing what I TELL IT TO DO :nono:

You answered you're own question ;)

Peace :toast:

[uwackme]

Yes, I was being rhetorical....hoping someone would chime in with an idea around it.

A bit too involved for a bios "tweak", more of a disassemble/hack thing, and we are fresh out of A64 bios Jedi's.

Maybe Oskar one day would give us a NON-NANNY version to play with :woot:

Or heaven forbid an open-source bios project managed to enter the world and save us all from Intel/MS domination in the coming years.

[HokaHey]

Hi Guys I have 4 sticks of TwinMos Speed Premium BH-5 ( 9 x 300, DIB166 245MHz@3.3.V@1.5-2-2-5) and can't make it stable in 3D applications (games... Battlefield 2 :p: ) SuperPi 32MB, Prime95, it's stable I can do it all night long ;-) NO PROBLEM but when I launch BF2 after 30 - 60 minutes I see my windows pulpit again :-(. On DIV 150 which is 225MHz there is no problem at all - everything just runs great (incuding BF2), but I have this strong feeling hehe that it's possible to make DIV166 245 MHz stable... I think it;s matter of Alpha Timings - can You guys just have a look at my timings and tell me if I should try diffrent settings ? Cause I'm lack of ideas as for now :-(

http://img240.imageshack.us/img240/6085/4x512ultrax4qn.th.jpg (http://img240.imageshack.us/my.php?image=4x512ultrax4qn.jpg)

[Vandread]

Might the new s939 opterons solve the 2t problem @ 4x512?

atm i got 3x SP3500/512's and 1 KHX3000/512 bh5
Thinking of finding another SP3500 bh if the new opterons do a nice job ;)

[uOpt]

Might the new s939 opterons solve the 2t problem @ 4x512?


Unlikely, they will have the same memory controller than Athlon 64s.

My guess is they are identical to Athlon 64s anyway, just better tested and possibly higher binned (which might explain the overclocking results). The 140-144 might just be SanDiegos with lower dividers (SD only does 2.2 and 2.4 GHz), which would also explain why they overclock well.