Llano OC & benchmark collection thread

[Barr3l Rid3r]

Still on the old one, write is way down O.o

Need to re run that.
Having real issues with hypers in 3D, just isn't happening.

what you couldn't get them to work even w/ newer stilt bios?

chew showed some nice results w/ them working fine even w/ interleaving on.

[El Gappo]

what you couldn't get them to work even w/ newer stilt bios?

chew showed some nice results w/ them working fine even w/ interleaving on.

I was on the gigabyte ud4h. Haven't tried again them since I think...F4 bios? I think Chew capped out at about 2000 and I'm sure it wasn't for lack of trying, considering what your average PSC does on Llano it's still not worth it. Compatability remains poo.

http://www.benchtec.co.uk/forums/att...8&d=1316711106

The hypers worked on the biostar and.. 'kind of' on the msi. Don't have the asus board so no stilt bios

Got some new ram now, good bit better than the Geil stuff I was on, should be fun when I get my apu back

[The Stilt]

Tried some real good Corsair PSC on Llano today...

The Asus board still acts quite funny, yet I was able to do pass atleast AIDA64 memory benchmark at DDR-2600
Unstable at that speed of course, but not bad for air cooling

I´m using the old 0802-based bios so the NB is running at full speed (not latched like in newer Asus bioses and on Biostar & Gigabyte boards). At ~ 1220MHz NCLK I need to force it to lower PState to drop the frequency.

Even the tCWL is set too loose on Asus (massive impact in write bandwidth) the board scores around 12.2GB/s (read), 15.1GB/s (copy) and 41ns latency at DDR-2500 (with CPU at 3220MHz) thanks to the high NCLK. And that ain´t tweaked yet.

Oh and this is Bank Interleaving enabled.. passes 3DMark06 GT-1 too, no problem

For PSC based modules & Asus boards I recommend following settings (leave the rest to auto):

MEMCLK Drive Strength: 2.0x (helps alot at DDR-2300+)
Addr/Cmd Setup Time: 1 MEMCLK (absolutely mandatory with PSC)
Addr/Cmd Fine Delay: 15/64 - 24/64 MEMCLKs (depending from the frequency, at DDR-2500 I use 21/64 MEMCLK)

Without these two Addr/Cmd timings adjusted correctly, the board won´t even post at anything above DDR-2200.
Dunno if it´s because of the board, modules or APU or all of them.

[Formula350]

Yea, no prob. With that heatsink I was running 48C in an open-air environment, with an ambient temperature of about 21C (70F if you're a Yank like me lol).

Again, keep in mind that is with a rather small heatsink, outfitted with a crazy ass 120+CFM 92mm fan I would like to toss my Spire Thermax Exclipse II (the high performance model, not the lackluster older Thermax II) on there and see how it does, which being easily twice the Katana's size with dual 120mm fans, I think that 48C would drop a LOT. My 1090T I don't think ever reaches 48C loaded with that heatsink, on the same test bench with nearly the same ambient temps. Just it has a very time consuming mounting method and isn't very ideal for swapping about test systems

(crap, apparently I didn't hit "Post" on this a few hours ago lol)

[Formula350]

Do we have an idea as to why memory Write performance will suddenly take a hhhheavy nose-dive with the system overclocked? I thought maybe it was my tRAS and tRC, but doesn't seem to be. Then I thought perhaps it was the tWR (Write Recovery), which was set at 16, so I dropped it to 14 but no change...

Is it perhaps the usage of the 9.33 divider at a higher FSB that is impacting performance? It is the same settings that net me those 20gb/s Sandra scores, but in AIDA or MaxxMEM, the Write is only 9000-9100mb/s I know Sandra does things rather differently, but my point is that I figure if the write was really being impaired that the Sandra scores would be lower than previous runs, runs which had Write scores up into the 11000mb/s range like normal.

Overclock Configuration (Discrete Graphics):
FSB: 123MHz (123.6 actual)
CPU: 3583MHz 1.4125V (actual: 1.452V)
CPU-NB: 2224MHz (1112 / 9x multi) 1.20V (actual: 1.216V)
DDR3: 2300MHz 1.60V (actual: 1.65V) 10-11-10-30 2T 35 (tWR 16 or 14, tRFC 90ns, tRRD 7, tWTR 8, tWRP 8, tFAW Auto [32])
APU PCIe / VDDP: 1.22-1.233V (actual: 1.23-1.245V
VDDR: 1.21V (actual: ??? Not found that test point yet)
Load Line: 1/2 (BIOS quirks, no otherwise logical reason)
SB Voltage: 1.1-1.2V (1.11-1.22V actual) (peace of mind but also SATA stability insurance)

Note: Both Bank and Channel Interleaving are Enabled


I'm starting to suspect perhaps it's one or a combination of the tRRD, tWTR and tWRP; otherwise tWR lowered further. At a lower memory clock, and thus lower CPU and CPU-NB, I set the above Trio to 6-7-7 and I believe tWR to 14. Below DDR3-2300 I can run 10-10-10 and I can see if it'll get into Windows at 2300MHz (not tried) and quickly get off a Write test. I'll see in a bit.

I've TRIED combing through my AIDA results to find some sort of pattern, but I can't seem to find anything, at least not with the data I collected which left out those 4 sub-timings :\ Just is annoying to have all the frequencies raised so much, improving basically every benchmark across the board, but having memory Write performance significantly low (and memory related tasks like AIDA's CPU-PhotoWorxx)

Any ideas

[The Stilt]

Which motherboard is this?

[chew*]

Tried some real good Corsair PSC on Llano today...

The Asus board still acts quite funny, yet I was able to do pass atleast AIDA64 memory benchmark at DDR-2600
Unstable at that speed of course, but not bad for air cooling

I´m using the old 0802-based bios so the NB is running at full speed (not latched like in newer Asus bioses and on Biostar & Gigabyte boards). At ~ 1220MHz NCLK I need to force it to lower PState to drop the frequency.

Even the tCWL is set too loose on Asus (massive impact in write bandwidth) the board scores around 12.2GB/s (read), 15.1GB/s (copy) and 41ns latency at DDR-2500 (with CPU at 3220MHz) thanks to the high NCLK. And that ain´t tweaked yet.

Oh and this is Bank Interleaving enabled.. passes 3DMark06 GT-1 too, no problem

For PSC based modules & Asus boards I recommend following settings (leave the rest to auto):

MEMCLK Drive Strength: 2.0x (helps alot at DDR-2300+)
Addr/Cmd Setup Time: 1 MEMCLK (absolutely mandatory with PSC)
Addr/Cmd Fine Delay: 15/64 - 24/64 MEMCLKs (depending from the frequency, at DDR-2500 I use 21/64 MEMCLK)

Without these two Addr/Cmd timings adjusted correctly, the board won´t even post at anything above DDR-2200.
Dunno if it´s because of the board, modules or APU or all of them.

Sounds like your loving high bin PSC

[Formula350]

Which motherboard is this?

ASRock's A55 Pro3. Peppy little board I think, but I'm missing some of the goodies the 1.00E BIOS had in it, which got flashed and lost :\ So I have no testing data to offer with it to see if the Writes were better at the same speed.

The Kingmax sticks are PSC btw.

[The Stilt]

If you have set the "NB multiplier" to /9 in bios, your NB is running at 494,5MHz...
Because it is not a multiplier, but a divider:

On A8-3850 APU: BCLK (123,6) * MainPLLCOF (36) = 4449,6MHz.
4449,6MHz /9 = 494,5MHz.

Ideally it should be at 1:1 with DRAM speed.

[Formula350]

If you have set the "NB multiplier" to /9 in bios, your NB is running at 494,5MHz...
Because it is not a multiplier, but a divider:

On A8-3850 APU: BCLK (123,6) * MainPLLCOF (36) = 4449,6MHz.
4449,6MHz /9 = 494,5MHz.

Ideally it should be at 1:1 with DRAM speed.

Well a few things, clarifications. It's not at that speed lol I meant 9 as a multiplier, as my BIOS shows 900MHz as the pre-overclock speed (900/100=9). After overclock it works out to be 1112.

Which brings me to: I can't do 1:1 as 900 is my max and the memory is at 2300 :\ That unfortunately doesn't appear to explain my write scores though, but I'd have to double check. Quite certain my memory has always been faster though, since it's pre-OC speed is 933MHz (9.33x).