S754 AGP board that allows 2 double sided sticks of RAM?

[Ike0069]

It seems that most of the S754 boards only allow 3 sides of RAM, but I already have 2x512 sticks of PC3200 and a 9700 pro that I will be using. What is a good S754 board that allows >3 sides of RAM, and is also AGP?

 

[DAPUNISHER]

I have tried 2 different BIOS and 3 different brands of ram with my kids DFI 250GB, but wasn't able to pass memtest86+ with 2x512mb DS, so I'd look elsewhere. I think I read Avalon said he was able to use 2 DS with a Epox? Look for his posts.

 

[Peter]

They all do - as physical reality applies here too, you'll see that timings and/or operating frequency of the RAM degrade as load count on the bus increases.

With a board whose BIOS is being too optimistic here, you'll have to choose an appropriate setting yourself.

 

[DAPUNISHER]

Originally posted by: Peter
They all do - as physical reality applies here too, you'll see that timings and/or operating frequency of the RAM degrade as load count on the bus increases.

With a board whose BIOS is being too optimistic here, you'll have to choose an appropriate setting yourself.

I did try looser timings, different slots, and more voltage, but not reduced frequency, at least not after overclocking was taken into acount I'll give that a go since it'd be well worth the trade off for more memory.

 

[Peter]

You are supposed to go back to 166 MHz and 2T command rate with two double-sided DIMMs. That's what the DDR standard essentially demands. With newer processor steppings, you might be able to eek out a little more speed.

See the table in AMD's BIOS writer's guide, here:

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

Page 173 ff.

 

[Ike0069]

Originally posted by: Peter
You are supposed to go back to 166 MHz and 2T command rate with two double-sided DIMMs. That's what the DDR standard essentially demands. With newer processor steppings, you might be able to eek out a little more speed.

See the table in AMD's BIOS writer's guide, here:

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

Page 173 ff.

Well that's not really acceptabe to me. Are you saying that ther are no S754 boards that allow 2 double-sided sticks at 200 MHz?

 

[Peter]

The DDR standard itself does not allow four ranks at 200 MHz.

If that's not acceptable to you, please go invent a better bus.

 

[Ike0069]

Originally posted by: Peter
The DDR standard itself does not allow four ranks at 200 MHz.

If that's not acceptable to you, please go invent a better bus.

Uhh, it works fine for S939.

 

[Peter]

No it doesn't. Look at the AMD document I linked to.

People who get it to "work" for them have a different picture of what "work" is - if you want to be safe over all operating conditions that occur in a PC (voltage fluctuation, thermal changes et al), you'll have to stick to the laws of physics and what the JEDEC engineering committee of ALL makers of RAM chips found to be ACTUALLY and PROPERLY working - and that's ONE single DIMM at 200 MHz, single- or double-sided. It degrades from there - two DIMMs even if they're both single sided are worse than one double-sided due to the added signal trace length and parasitic capacity, and from there adding more RAM chips onto the common bus signals adds more parasitic capacity, making the signal edges flatten out more and more - and THAT is what makes high frequency operation and 1T commands impossible.

This is exactly why platforms where LOTS of RAM are to be used go with "Registered" DIMMs. The register chips isolate the RAM chips from the common bus, so that the DIMM presents one single load to the common bus rather than up to 18.

 

[Ike0069]

Originally posted by: Peter
No it doesn't. Look at the AMD document I linked to.

People who get it to "work" for them have a different picture of what "work" is - if you want to be safe over all operating conditions that occur in a PC (voltage fluctuation, thermal changes et al), you'll have to stick to the laws of physics and what the JEDEC engineering committee of ALL makers of RAM chips found to be ACTUALLY and PROPERLY working - and that's ONE single DIMM at 200 MHz, single- or double-sided. It degrades from there - two DIMMs even if they're both single sided are worse than one double-sided due to the added signal trace length and parasitic capacity, and from there adding more RAM chips onto the common bus signals adds more parasitic capacity, making the signal edges flatten out more and more - and THAT is what makes high frequency operation and 1T commands impossible.

This is exactly why platforms where LOTS of RAM are to be used go with "Registered" DIMMs. The register chips isolate the RAM chips from the common bus, so that the DIMM presents one single load to the common bus rather than up to 18.

I think maybe you're trying too hard here.

It works for me and here is my defintion of "works":
I currently have 2 double-sided sticks (2x512) of Corsair Value series RAM running at 210 MHz at stock timings (2.5-3-3-8).

I 'm just looking for a S754 board that will do this.

 

[gunblade]

I had a MSI neo platinum that does that just fine. Doubled sided 512MB x 2 for 1 gig total of ram runnning 200Mhz 2.7v with no overclocking.

JEDEC standard are the most conservative and are mostly outdated by the time they came out. If you are on the mailing list for JEDEC meeting, you will see that they spend more time on politics and where to held the meeting than to form working groups on pushing the technology envelope.

 

[Peter]

gunblade, thanks for proving my point. You upped the voltage to get there, that IS outside specification parameters, nothing that a serious manufacturer would ever promise. You're into YMMV territory.

Ike, your two sticks on the socket-939 board are on separate channels with separate signals. Technically you're running that one in DDR400 specification, one DIMM per bus. No news there. Try four DIMMs, maybe also read up on how people struggle to get four DIMMs to run at DDR400 on socket-939, and you'll see that it's no different there. The push to dual-channel chipsets is not just about performance, but also about overcoming the limitations of the ever faster clock speeds.

JEDEC have been sweating long and hard to make a well calculated and proven DDR400 definition. Remember that none of the DRAM makers did originally even WANT to go there, seeing how DDR333 already had to be limited to two DIMMs.

Besides, I'm not "trying too hard", I happen to be involved in mainboard engineering deeply enough to know the difference between "appears to work" and "actually works properly".

 

[Megatomic]

I understand what you are saying Peter and I know that you know your stuff, but I'd like to interject that I ran two double sided PC3200 DIMMs on my Soltek K8AN2E-GR (NF3 250Gb)at 200MHz. It ran like this for over a year and it passed 24 hours of Memtest86+ v1.55 with zero errors. I never had any stability issues, I was very pleased with that system.

 

[Peter]

Like I said, YMMV. The point is that no manufacturer can boldly claim "this works" - thus I've been lecturing here about what people should EXPECT. If your individual results happen to have headroom above that, be my guest. The next guy using the same mainboard just shouldn't EXPECT that he'll get the same results.

 

[Fern]

Well, FWIW I got a Chaintech VNF3 250 nForce3 paired with a a64 2800 (IIRC winny) and 2 x 512MB of Corsair VS to run perfectly fine @ 200mhz.

Seems others have also. Did see one guy with a problem but he was using a Sempron chip IIRC.

Fern

 

[Andres3605]

MSI neo platinum working rigth now, with no problems