thuban mem controller testing

[massman]

FYI, also tested the Gigabyte 890GX board, but memory couldn't even pass 1750CL9. There were some issues with the bios.

[FlanK3r]

Chew:
help me a bit please, which board? Crosshair IV or MA890FXA-UD7 ? have u some trouble with BIOS? No vdrop?We have in Czech Crosshair IV now, but UD7 not. So, i can still waiting for UD7 or can take Crosshair IV? Thx.

[Leeghoofd]

MSIMAX possible to take a screenie of the bios settings... my CH IV and Thuban ES only are willing to give me 1900mhz on the rams... tried 4 biosses already... 965BE goes up to 1920Mhz C7 pretty easily

[[XC] Oj101]

WOW, DDR2000 SMASHED

MrIcee, does this mean we can expect a return from you?

And Brian, do these boards do a "true" CAS6? I remember you telling me that MSI's GD70 was doing well with newer BIOSs and CAS6 because other latencies were increased.

[MrIcee]

WOW, DDR2000 SMASHED

MrIcee, does this mean we can expect a return from you?

And Brian, do these boards do a "true" CAS6? I remember you telling me that MSI's GD70 was doing well with newer BIOSs and CAS6 because other latencies were increased.

MrIcee has almost fully re-equipped his test bench with the latest and greatest

I'm hoping by this weekend to be doing R + D testing on a CH4 and a 1090T. I have a 965BE and 555BE also for those that may want results.The last few weeks I've set up a CH3 and 965BE to get re-aquainted with the current AMD platform. Much has changed since my days testing on Abit and DFI Thank yous to chew* for his insights and assistance. Getting my feet on the ground here and running, tho these fellas clocking with LN2 are beyond my scope right now. I'll be starting with my old custom watercooler and vid card chiller, and also have a 404A modded Prommy. So my results should reflect what the average guy on watercooling can expect. I'm gearing up for 3D too, with a pair of Asus 5870 V2's ready to go.

I'm ready to start having fun again after a very long abscence!

Best regards,

Randi

[[XC] Oj101]

I look forward to your return

What's next? OPPINTER? macci? JCViggen (4GHz eight or so years ago)? MickeyMouse (one of the pioneers of LN2 pots)? Tom Holck (Mr. Cascade on drugs)? Captain Cascade? Digital Jesus (did he ever return after being called out about his 01 results or whatever?)? cpulloverclock (he was awesome with 939)? The guy whose avatar was a jumping cat with lasers coming out of its eyes And finally you. You were a legend on 462

Really really good times

[chew*]

MSIMAX possible to take a screenie of the bios settings... my CH IV and Thuban ES only are willing to give me 1900mhz on the rams... tried 4 biosses already... 965BE goes up to 1920Mhz C7 pretty easily

Leeg those were my runs i'm busy had msi post them for me.

Sorry what you see is what you get for timings......Use memtest, best i can say is versus 7-7-6-19-27 my optimized runs are 3 secs faster in 32m no tweaks.

[chew*]

WOW, DDR2000 SMASHED

MrIcee, does this mean we can expect a return from you?

And Brian, do these boards do a "true" CAS6? I remember you telling me that MSI's GD70 was doing well with newer BIOSs and CAS6 because other latencies were increased.

from my testing the performance at 2000+ is not optimal or where i think it should be but it is faster than my previous stuff at cas 6 at same speed on previous boards so not what i would say artificial......If it was I would not have bothered posting it.

[Leeghoofd]

Leeg those were my runs i'm busy had msi post them for me.

Sorry what you see is what you get for timings......Use memtest, best i can say is versus 7-7-6-19-27 my optimized runs are 3 secs faster in 32m no tweaks.

Hehehe Brian ; testing 1980 now (quits around loop 20 each time ), I just switched the rams to the other dimm slots... was shocked that 1960Mhz went just like that...Really wondering how Gary got 4dimms working at them speeds... super tweaking !

[mike]

looks like you guys are warming up already! Can't wait to see and learn what you'll be doing this weekend with that new AMD puppy! Wonder if Fugger is going to wear a turban on Friday

[chew*]

FYI, also tested the Gigabyte 890GX board, but memory couldn't even pass 1750CL9. There were some issues with the bios.

good news is we ran some 32m on UD7 and iirc 1900 tight was fine.

Have one all prepped up for ln2

[[XC] Oj101]

I just saw DDR2100 and a crash at DDR2160

[chew*]

I just saw DDR2100 and a crash at DDR2160

I can 1m alot higher.....didn't waste my time or end users.

[[XC] Oj101]

How high are we talking? Because DDR2000 and above is unheard of on AMD (until now).

[chew*]

How high are we talking? Because DDR2000 and above is unheard of on AMD (until now).

well above 2100 at those tight timings.....32m is a much diff story even at 10-10-10 wall was present. Wall around 2050 very simlar to the previous wall i tore down at 1904.....when i have the time I will see if it's a hardwall or a wall that can be massaged past.

So demand 32m or it's useless and misleading

[FlanK3r]

whau, good, UD7 looks simillary as Crosshair IV for OC ?

[[XC] Oj101]

DDR2100 was benchable but at rubbish timings (non Elpida RAM, it was CAS9 I think). The better controller might be the biggest reason for me to move

[bingo13]

Hehehe Brian ; testing 1980 now (quits around loop 20 each time ), I just switched the rams to the other dimm slots... was shocked that 1960Mhz went just like that...Really wondering how Gary got 4dimms working at them speeds... super tweaking !

My head hurts and my MasterCard is begging for mercy. The IMC is going to make a huge difference. So far, my best CPU (ES or Retail) is a random 1055T from Amazon that I bought for my son's system and now he will never see it. I sent you settings to try and of course this requires some Corsair GTX2 kits.

Also, we have a couple of memory tables left to fine tune for efficiency but with BIOS 0707, be sure to set HT manually (as high as possible) to ensure best possible write speeds as the tables flip above 1800 and HT on auto will fall back to under 2000. In an early beta release, 8GB at 2000 is working but efficiency at 1600 is still much better at this point.

[Dumo]

Also, we have a couple of memory tables left to fine tune for efficiency but with BIOS 0707, be sure to set HT manually (as high as possible) to ensure best possible write speeds as the tables flip above 1800 and HT on auto will fall back to under 2000.

QFT

[BeepBeep2]

FYI, also tested the Gigabyte 890GX board, but memory couldn't even pass 1750CL9. There were some issues with the bios.

Hmm...

Can I ask what bios? I just booted into windows @ 1775 CL7 on my Gigabyte 890GX board and I'm currently at loop 6 in 32M

It's Micron GTS @ 2.0v

[massman]

Also, we have a couple of memory tables left to fine tune for efficiency but with BIOS 0707, be sure to set HT manually (as high as possible) to ensure best possible write speeds as the tables flip above 1800 and HT on auto will fall back to under 2000. In an early beta release, 8GB at 2000 is working but efficiency at 1600 is still much better at this point.

Any reason why the HT link has to be above 2G for better performance?

In regard to the efficiency: I suppose that Thuban still has the same issues as Deneb has when it comes to memory data transfer. AMD people confirmed that an elevated memory frequency is no use if you don't increase the IMC frequency accordingly. An appropriate ratio would be IMCRAM as 3:1, which would mean a 6GHz IMC if you want to make propper use of 2G memory.

[massman]

Hmm...

Can I ask what bios? I just booted into windows @ 1775 CL7 on my Gigabyte 890GX board and I'm currently at loop 6 in 32M

It's Micron GTS @ 2.0v

I don't recall the bios, it was sent to me a couple of weeks ago. At the time, I had a very small timeframe in which I could test the Thuban using extreme cooling, so I had to go for the best board then.

FYI, I was using GTX2 memory. It's very likely the cause of the low memory frequency was due to incompatibility issues.

[BeepBeep2]

Any reason why the HT link has to be above 2G for better performance?

In regard to the efficiency: I suppose that Thuban still has the same issues as Deneb has when it comes to memory data transfer. AMD people confirmed that an elevated memory frequency is no use if you don't increase the IMC frequency accordingly. An appropriate ratio would be IMCRAM as 3:1, which would mean a 6GHz IMC if you want to make propper use of 2G memory.

If I remember that ratio is more like 2:1 The stock IMC freq. was supposed to be good up to 1066...

And the wall on my 890GX board is 940/1880...I can do 1850 CAS 7-7-7 and then everything else above is just junk. Not even 1900 CL10.

[bingo13]

Any reason why the HT link has to be above 2G for better performance?

In regard to the efficiency: I suppose that Thuban still has the same issues as Deneb has when it comes to memory data transfer. AMD people confirmed that an elevated memory frequency is no use if you don't increase the IMC frequency accordingly. An appropriate ratio would be IMCRAM as 3:1, which would mean a 6GHz IMC if you want to make propper use of 2G memory.

It is an auto rule function in the BIOS, in fact raising HTT (Bclk) above 240 with both NB and HT set to auto will result in both of them falling back to 1920MHz (depends on ratios also), which is even worse, they should stay in sync at 2400MHz until the next auto rule limit is reached. So far, most Thuban processors we have tested will do NB/HTat 2600MHz on safe voltage settings so additional tuning is in order for the tables.

This will be corrected in the next BIOS but the main reason for maxing HT is to sync with NB speed (which you obviously want as high as possible on this platform) to keep the current memory table at optimum settings. Right now the rules are set to ensure stability/compatibility instead of maximum performance. Now that the AGESA code is stable for Thuban, we will set different rules when the BIOS "sees" performance settings being utilized.

Before anyone says anything, we went through four different AGESA code levels in the last two months or so as the platform and processors were finalized. This is the reason why we had around seven BIOS releases since the ES samples were delivered in early March.

Yes, to truly maximize memory performance, we truly need higher NB speeds to improve efficiencies as memory increases above 1066 on this platform. This is one of the main reasons I like to run NB/HT/Memory in high-sync where possible and why 1600C6 is still my favorite memory setting for everyday use, provided I have a IMC that allows that setting and 3200 on the NB/HT side. I have one right now that will do it on air so I am fortunate.

[PatRaceTin]

damn!!!! i want 960T