G.Skill PC4800 FF not stable at 280, help plz

[jahjahbinks]

Here's A64 tweaker shot:

http://img471.imageshack.us/img471/5208/untitled4nk.gif

CPU is X2 3800+ at 2.8 gig. I tried Vcore up to 1.61 and got same result which is not stable in Pi 1M. BIOS is 623-3, Dram response time is set to normal (this doesn't show up in A64 tweaker), drive strength is 7, data strength is 1. It can pass test 8 in Memtest many times even at 285 at 2.5-3-3-6.

Any thoughts?

[m3csl2004]

says your mem voltage is running at 2.6v - have a look at the tccd information on here and check what kind of voltage these will run best at

[jahjahbinks]

Yes, I will try to raise Vram later. I had bad experience with last TCCD (another brand) I had, I suspect that the 2.9v probably cause damage to it, so I wanted to go easy this time. Plus I saw some screenshots here with people getting 300 at only 2.6-2.7.

Time for dinner.

I just dropped CPU multi to 9 (realized 2.8 on 3800+ is pretty high expectation), it passed Pi 1M. :clap:

[jahjahbinks]

I dropped it to 265, Vram at 2.7, still no go, any idea?

[NickS]

2.85-2.9v VDIMM, 256 clks idle cycle :).

[jahjahbinks]

swapped ram from orange slot to yellow slot, Pi 1M passed at 2.8 gig (280x10), Vram at 2.67

It seems this is a temporary thing, because I tried Pi 32M and it failed in less time than in 1M, then I tried 1M again, it failed within 5 seconds.

Tried 280 again, this time I dropped CPU multi to 6, Pi failed in a matter of seconds.

[Ozan Baseski]

You should try lowering your cpu speed. Your default cpu speed is 2 ghz and yet you are trying to overclock both the cpu to 2800mhz and the ram. Perhaps the memory controller in your cpu couldn't take that much load. Lower your cpu multiplier to 8 or 9 at least and try again

[Zeus]

Send 'm over to me. :D

Seriously, mess around with DRAM drive strength- data drive strength.

DRAM drive strength-7
Data drive strength 1 or 2
Usually gives the best results.

Also DRAM response time and Max Async Latency en Read Preamble will help you stabilize.

If nothing works try TRCD at4 instead of 3, if that fails you have got to send 'm my way. :)

[x3ro]

From my own experience i know that most tccd based modules run best at a voltage of 2.7v to 2.85v. Try to find the mem's sweetspot

Regards

[jahjahbinks]

Yes, I have been using DDS 7/1.

For MAL and RP, I tried 7/5, 7/5.5, 8/5, 8/5.5, 8/6, usually if u don't go too high or too low, A64 tweaker won't crash, so it's easy to play with them.

I am gonna try play it more because I refuse to believe it fails at cpu multi at 6, that's lower than the stock speed.

I just bought it some hours ago, but I don't believe it needs burn in.

[jahjahbinks]

Just finished Pi 32M at (285*6), but then my computer froze and when I reset, I got three red LEDs. Same problem as I described here: http://www.xtremesystems.org/forums/showthread.php?t=76944 I had to clear CMOS for like 10 minutes to POST again.

The reason I bought this RAM is because I have tried everything and I suspect the RAM I had before was bad. Well, the only thing left to change is my CPU, don't think that can cause three red LED, my motherboard, what's the chance my motherboard is killing my RAM, and my PSU.

That's very strange.

[Zeus]

Reading your other thread also, your mobo must have gone bad, or maybe even your CPU. :(

[jahjahbinks]

Well, I bought another power supply. This time computer did not die after stopping dual prime, but when it completed Pi 32M, it rebooted automatically.

[formyfaith]

Here's A64 tweaker shot:
CPU is X2 3800+ at 2.8 gig. I tried Vcore up to 1.61 and got same result which is not stable in Pi 1M. BIOS is 623-3, Dram response time is set to normal (this doesn't show up in A64 tweaker), drive strength is 7, data strength is 1. It can pass test 8 in Memtest many times even at 285 at 2.5-3-3-6.
Any thoughts?

FF's rated specs are as below.

PC4800(DDR600) 2.5-4-4-8 / PC4400(DDR550) 2.5-3-3-7 (Dual channel)

The reason I'm telling ya this is because, I've seen a lot of FFs that has no potential at all, and only does as shown in the rated spec. So don't be so surprised with your pair.

Just to remind you, I've recently had a chance to test 10 pairs of the new heat spreader version FF, and non of them were able to do 2.5-3-3-X at FSB 300. Also, surprisingly half of them only worked exactly as thier rated spec which is 2.5-3-3-7 at FSB 275, and rest of them around 280-290 ;) My guess is that price drop has made some effects to the new FF batches. You'll need some real luck to get a nice pair. :p:

[Zeus]

Say what?

What's going on at Gskill and their speedbinning proces?
Are you suer nothing else held back your RAM clocks?

These budget FX ram seems to do better than their topmodel FF.

It's pretty sad spending $250,- for FF to find it won't do 300 2,5-3-3-x while some $150,- FX ram can.

Thanx for warning, i was about to get some FF. ;-)

[jahjahbinks]

just finished Pi 32M at 300 @ 2.5-4-3-8, Vram is 2.65.

I see the potential in this one.

[TheMeatFrog]

I just got a new retail set of FFs w/ heat spreaders from the egg that do 307MHz 2.5-3-3-6 ;)

[mdzcpa]

My new FFs run 310 at 2.5-3-3-7 at only 2.83v. Hit n miss like anything else I thnk.

[formyfaith]

Say what?

What's going on at Gskill and their speedbinning proces?
Are you suer nothing else held back your RAM clocks?



Nope, nothing holding me back for sure. Take a look at the highest TCCD thread made by bachus, I was able to reach 2.5-3-3-5 @ 309 at 10 multi with a FF bought a long time ago at around 400$. Of course reaching means passing 32MB SPI.

just finished Pi 32M at 300 @ 2.5-4-3-8, Vram is 2.65.

I see the potential in this one.

Sadly, I can surely tell you 32MB SPI at 300 @ 2.5-4-3-8 doesn't really mean potential. Almost any FF can do that. At least all of the sets I've tested ;)

Also, Just to mention, tested 5 more sets, of course retail, and non can pass 32MB SPI 300Mhz @ 2.5-3-3-7. (actually a lot worse than people might be thinkin) Guess it's more than just Hit n miss. ;)

[jahjahbinks]

Yeah, i think G.skill labled their RAM according to how well they can pass memtest at rated timing at 300, however there can be still 10 miles of distance to go from that to passing Pi 32M.

If they use Pi standard, they will have a lot of problems with yield, and sooner or later they will be out of business.

512MB is 43 billion bits, assume each transistor is one cell which is one bit (not sure if any R&D has achieved anything less than one cell per bit yet, at least not when I got out of college), that's 43 billion transistors alone, not including the control circuitry and all other goodies. In comparison, a CPU has only several hundred million transistors. It's impossible to guarantee that every single one in that 43 billion works, so what RAM manufacturers do is to make say 44-46 billion (for example), then do a RAM test to find out which column/row has dead transistors, then disable those. Redundancy is the only way to guarantee reliability. Now, will a 512 stick with only 1000 dead transistors be able to run at tighter timing than one with 10 million dead transistors? I don't know, this is a hard question that has to do with how RAM operates at a lower level. :)

Now that was my engineering side of view. I can go deep there since that's what I get paid for. I am not a business person but from marketing point of view, you probably want to hand pick some really good RAM sticks when you first release the product, so when people buy it and find them to be really good, they tend to spread the news by posting reviews on various forums and such, then when the order number increases dramatically later on, they would not have enough time to sit there to hand pick them as before. So they would shorten their test procedures/times to guarantee they don't run them out (a way to stay in business). And what do consumers get? :D

see how the psychology works here? ;) I am no marketing, but I believe it doesn't take more than common sense to figure all out.

Btw, this applies to all semiconductor industry, my intention was simply to state the current trend in yield and such. :)

[formyfaith]

)

Wow, I don't really get what you're talking about regarding the transistor part :D, but I surely do agree a lot with you on the rest.