Need help adjusting memory timing

[genetix]

Well, that anyway didn't go too good. Need to think a bit, if -1 CL gain was less than 100MB/s that means something there is now holding this thing back.

Edit #01:

I honestly am afraid that the problem is now the Memory controller speeds or FSB. Since on 1066Mhz your L3 Cache speeds just raised 1000MB/s on read speeds. CL6-6-5-18 did a lot better than CL5.

Edit #02:

Lets try something literally weird.. Try adding this there: CL5-7-7-24, tRC 25T and put tWR to 5 1066Mhz. Also, check your HTT / MCH volts wanna know what those are.

[retsamradassaT]

Okay, I forgot to about the "on 1066Mhz anything above and it should be minimum of 31T" part you mentioned on your edit, so here's the new results when my tRC is 31T:



Edited: In response to your edit#2, I will try that now.

[genetix]

Edit #02:

Lets try something literally weird.. Try adding this there: CL5-7-7-24, tRC 25T and put tWR to 5 1066Mhz. Also, check your HTT / MCH volts wanna know what those are.

Try that one still from my previous message. I've seen boards actually rise speeds when we lose up totally on low CL. So that might actually lose enough to free the bdw.

btw, that tRC did good on 31T. You can see the Lose on L3. Seems more you give it space less it drops stress on core and faster it goes.

[retsamradassaT]

Edited: How do I check HTT / MCH volts?

[genetix]

Didn't do to well. Could you copy me Everest page: Motherboard > Motherboard > Need specific Memory Bus bandwidths.

(There should be 2 'bandwidth' parts on that page 5 numbers. Copy those to me. don't need screen even.)

Edit #1:

We can worry on those volts actually later (sorry to say but these needs to be checked on BIOS since otherwise they would be on Everest Sensor page), but I think the bus bandwidth runs out on here, so, check that now first.

[retsamradassaT]

Field Value
Motherboard Properties
Motherboard ID 05/12/2009-RD780-SB750-7A66AG08C-00
Motherboard Name Gigabyte GA-MA790X-UD4P

Front Side Bus Properties
Bus Type AMD K10
Real Clock 200 MHz
Effective Clock 200 MHz
HyperTransport Clock [ TRIAL VERSION ]
North Bridge Clock [ TRIAL VERSION ]

Memory Bus Properties
Bus Type Unganged Dual DDR2 SDRAM
Bus Width 128-bit
DRAM:FSB Ratio 16:6
Real Clock 533 MHz (DDR)
Effective Clock 1067 MHz
Bandwidth [ TRIAL VERSION ] MB/s

Motherboard Physical Info
CPU Sockets/Slots 1 Socket AM2+
Expansion Slots [ TRIAL VERSION ]
RAM Slots 4 DDR2 DIMM
Integrated Devices Audio, Gigabit LAN, IEEE-1394
Form Factor ATX
Motherboard Size 230 mm x 300 mm
Motherboard Chipset AMD790X
Extra Features [ TRIAL VERSION ]

Motherboard Manufacturer
Company Name Gigabyte Technology Co., Ltd.
Product Information http://www.giga-byte.com/Products/Mo...d/Default.aspx
BIOS Download http://www.giga-byte.com/Support/Mot...BIOS_List.aspx
Driver Update http://driveragent.com?ref=59
BIOS Upgrades http://www.esupport.com/biosagent/in...m?refererid=40

[genetix]

damn, trial versions the needed value, lol, But hang on there not trying to waist your time just a bit tricky. Just downloaded your board manual so I can see what you're dealing with there on BIOS or what is changeable.

Edit #1:

Ok, got the manual.

All tRFC is minimum of 127.5ns (Best would be, if you use DIMM 1 & 3 as 127.5ns and DIMM 2 & 4 as 195ns.)

Need to know have you tryi'd to boot that up with 'Command Rate=1T' ? If not try to boot it up on CR 1T and drop the CL5-6-5-17 back there with tRC=26-31T.


(To note as far I can see on manual. This is all I can do memory based. I honestly think the main problem is that your memory controller bandwidth is simply not enough to get more power out. That is because you use fully the multipliers to clock the system or because of the HT Clocks.)

Edit #2:

Or I might simply be wrong on Memory controller not beeing enough. As on 1066Mhz there was correct jump on memory speeds from 800Mhz, but the damn weird thing in this is that on:

800Mhz: ~6900MB/s (this is 500MB/s too low and Latency is like no where near the ideal. This could do just fine with CL4 1T Command Rates, if possible)
1066Mhz (best): ~8170MB/s (This is near good and latency 52ns ain't too bad, but far from being proper, latencies should go even below what they were here.)

[retsamradassaT]

A little later after I joined this forum, like after I made my first post here in this topic , I tried booting my computer with Command Rate = 1T (with the settings listed in my first post), and the computer wouldn't even POST, like I can't even access the BIOS, so I had to clear the CMOS by sticking in that jumper thing into that clear cmos on my mobo.

With that said, should I try booting it with Command Rate = 1T?

[genetix]

A little later after I joined this forum, like after I made my first post here in this topic , I tried booting my computer with Command Rate = 1T (with the settings listed in my first post), and the computer wouldn't even POST, like I can't even access the BIOS, so I had to clear the CMOS by sticking in that jumper thing into that clear cmos on my mobo.

With that said, should I try booting it with Command Rate = 1T?

Well, the problem I was trying to solve with Command Rate 1T was that. It sets Secondary timings like tWTR etc etc all in lowest possible and it pushes the memory to it's limit on everything else, but main timings. Your board doesn't have these so called secondary timings and, if it doesn't POST it doesn't POST that's that. Don't really need to know, if you know it doesn't POST is enough for me.

[retsamradassaT]

Alright, thanks for the heads up.

Here's the benchmark results with the "All tRFC is minimum of 127.5ns (Best would be, if you use DIMM 1 & 3 as 127.5ns and DIMM 2 & 4 as 195ns.)":

[genetix]

Well, only thing I can think of to improve that setup now. Is to lower the latencies as low the go on 1066Mhz.

It's damn hard to tweak a system while not being able to quickly test stuff on screen, heh, if this would be standing on my table I'd test like 800Mhz drop everything where they go and as long it posts see what how tight latencies I could get. See, if that would lead this board to ~8000-9000MB/s with absolute maximum 48ns of latency on windows and still be flexible to L3. 1066Mhz acted a lot better on my opinion I think higher speeds is way to go on this board and I'd check how low that would go (this would probably get you 9000MB/s easier), but mainly, if the 1066Mhz raised that so damn well. I'd probably be eager to see where it ends to raise such speeds or even clock the FSB a bit to see if that helps raising.


The L3 is curious.. If 8305 MB/s = 8102MB/s memory read speeds. Does this mean 10479 MB/s Write would be around 10000 MB/s write speeds. That would be like 1000MB/s higher speeds than on my board already. Also, if Copy Speed is 11539 MB/s, if that has any reference that the memory Copying speed would be over 11Gb/s. That's like lightning fast inside the memory would mean all in all there there is some like intel boards 'Performance Level' or other words 'tRD(Read Delay)' which in 16:6(8:3) nor on your earlier 800Mhz(1:2) would be +3T over what it could be making writes and copying near 1/3 faster than read could be.

I simply cannot find this value on your board to change that read and write speeds could be closer together. Which would create better compination.

[retsamradassaT]

Well, I guess I shall thank you for all your help. I guess we can figure this out another time, as it is getting pretty late here. So of all of the settings that we have tested out so far, which one do you think it would be the best for me?

[genetix]

the 1066Mhz 5-6-5-17 with tRC 31 was looking good. Or at least it has some mind in it, If you could drop the write/copy speeds somehow I think that would be correct direction.

Perhaps also test other ratios than 8:3 or 2:1. loop them with FSB to right (in bareable levels like 225 x16).

[retsamradassaT]

Changing the FSB didn't worked out quite well, computer wouldn't even POST, so I had to clear cmos =(

and it looks like my Everest evaluation software is over.

[genetix]

heh, well changing fsb need more volts on MCH/NB/HT etc etc what ever was raised plus AMD doesn't exactly love this thing. So, it a bit tricky. Other thing which came to my mind was what other Ratios you got than 8:3 or 2:1.. hell there's 8 different ones on Intel most be more than 2 on AMD. Doesn't really matter what the DRAM Speed says (below 1100Mhz would be nice).

Anyway, different ratios acts totally different. Perhaps there's answer on there somewhere.

Free Legit Everest 4.60:
http://www.lavalys.com:8081/everestultimate460.exe
serial: RHGJE-54X47-HML9C-V498I-UUMVI

Source: MyDigitalLife Everest by: pro.de