BIOS > DQS Training Question

[Groonx2]

When I have tRFC @ 195ns, with DQS training [PERFORM], and then I change the tRFC to 327.5ns, the first cold start requires ~11 minutes to POST. Presumably (but not for sure) this is the time the DQS training requires to make its adjustments.

When I do a subsequent cold start, the time to POST may require some 10 minutes less. Let's suppose that this time, the DQS option recognizes that the RAM is set to its liking, so it doesn't need those extra 10 minutes.

Now, according to what I have read, if I set the DQS training to [SKIP], it will return to "default". This would be very unfortunate if default really means default, because then the RAM would not be functioning optimally.

HOWEVER, if the DQS training were actually to return to my own system's now-established DQS "trainedUSING the settings that were established in previous DQS training, then I'd have the best of both worlds: speed and accuracy.

See what I mean? The first time, you recheck the telephone number to make sure it is right. (Accuracy.) Subsequently, you don't recheck it because you already know it is right. (Speed and accuracy.)

I'm concerned about the DQS "returning to the default setting". The default is [SKIP] DQS training. My mobo, RAM, and CPU are reluctant dance partners, and I need the most optimum settings I can get.

Anybody familiar with DQS training? Thanks.

[Lsdmeasap]

Re: BIOS > DQS Training Question

Sorry I am not familiar with this setting or option, and I have limited AMD Experience so I can't offer any advice but to say i think your assumptions should be correct.

Maybe someone else will chime in with AMD, like Artdriver. I will send him a PM and see if he knows about this.

Otherwise your best bet may be to ask Gigabyte Tech support directly
GIGABYTE - Support&Download - Technical Service Zone

I wll PM Artdriver now and see if he can help you out.

[artdrivers]

Re: BIOS > DQS Training Question

I don't know if this is helpful, but I'll offer it anyway

Do you get the error message: "DQS training failed on previous boot, Revert to slower DRAM speed Press F1 to resume." When you start your computer?

This error has been seen with some Gskill and other types of memory with DQS training enabled.
This is from AMD page 40----> http://www.amd.com/us-en/assets/cont...docs/41322.pdf

263 Incompatibility With Some DIMMs Due to DQS Duty Cycle
Distortion
Description
Some DIMMs exhibit a duty cycle distortion on the first DQS pulse of an incoming read request
which may cause the processor's DRAM interface to miss a beat of data in a read burst.
Potential Effect on System
Undefined system behavior due to incorrect read data.
Suggested Workaround
If the memory is DDR2-533 or DDR2-667 write 00000800h to F2x[1, 0]9C_xD040F30, else write
00000000h to F2x[1, 0]9C_xD040F30.
The write of 00000000h to F2x[1, 0]9C_xD040F30 is not necessary if BIOS can not change the memory clock speed without a cold reset or if BIOS does not support the above mentioned memory configurations.
When exiting from the S4 or S5 state, apply this workaround prior to setting DRAM Configuration
Low Register[InitDram] (F2x[1,0]90[0]). In addition, for the above mentioned memory
configurations, BIOS should set the DRAM read DQS timing control loop range to 32 during DQS
position training.
When exiting from the S3 state, apply this workaround prior to setting DRAM Configuration Low
Register[ExitSelfRef] (F2x[1,0]90[1]).
Fix Planned
No

With my desktop system when I click START/SHUTDOWN/STANDBY my system shuts off Then when I press the power switch the system comes back on where I left it when I went into standby. It does not matter if I have the DQS training enabled or not. I do not get a delay on boot in standby. When I shut down or restart it goes through the default P.O.S.T. but with the manual settings where I last set it. Not the manufacturers auto settings as it would when you clear CMOS. I believe that's what you were questioning.
I don't have any issue with the DQS training or standby function. I use Dual Channel, but I do not use Ganged mode either with my Phenom CPU because Phenom does not need it.
I am using 4 Kingston HyperX DDR2 1066 at Memory Clock to X5.33 and SPD timings. They are 2 different sets with the same exact Kingston HyperX numbers. (Which I don't know how that happened it's not what I was suppose to have purchased)

DDRII Timing Items SPD Auto
x CAS# latency 5T
x RAS to CAS R/W Delay 5T
x Row Precharge Time 5T
x Minimum RAS Active Time 18T
x 1T/2T Command Timing 2T
x TwTr Command Delay 3T
x Trfc0 for DIMM1 Auto 105ns
x Trfc2 for DIMM2 Auto 105ns
x Trfc1 for DIMM3 Auto 127ns
x Trfc3 for DIMM4 Auto 127ns
x Write Recovery Time 6T
x Precharge Time 4T
x Row Cycle Time 24T
x RAS to RAS Delay 4T

I may have shared more info then needed.
With my AMD 9600 black box edition I usually just change the CPU multiplier for mild OCing.

[Lsdmeasap]

Re: BIOS > DQS Training Question

Thanks for chiming in artdrivers, hope this helps him some

8GB of ram worries me as well. Are you running it at 1066? If so, don't. Use 800-900 or so

[Groonx2]

Re: BIOS > DQS Training Question

Thank you, both.

@Artdrivers, no, you didn't share more info than needed. Too often people are too brief in what they say. I can learn much more by studying a post like yours, where enough info is available to suggest a good perspective. So I appreciate your good work.

I've learned that "genuine" cold starts are very different from warm starts. The capacitors retain quite a bit of memory (well, that's what we pay them for!) AFTER all power is shut down. Some people turn off the power at the power bar and then press the case power button, or even keep it depressed for almost a minute, in attempts to fake a cold start. Even that doesn't diminish hardware temperatures, so a cold start after the system has been off for a number of hours is still a different animal.

On mine, "a number of hours" might be as few as 3 hours with no power at all, but I haven't proven that yet. Definitely overnight total power-offs of 7 hours or more would seem valid for data inferences; probably 3 hours is long enough.

As far as I know, any form of "restarting" via Windows or (likely) of maintaining an electrical connection to a system that is down but not out dramatically influences the POST data. I think the POST beep on a Windows restart occurs after much less information has been tested, but I am only just learning about this. On my system, I receive a distorted bd-BEEP! on the first POST following any cold start, but I get a perfect BEEP! on any subsequent POST (if the system recycles to try again) or on any Windows restart. So there are POSTs and there are POSTs; not all POSTs are created equal.

That is why (I think) you "don't have any issue with the DQS training or standby function". On standby, I don't think DQS training is taking place at all. The system isn't really off.

I haven't received any error messages regarding DQS. Not yet! But my question was a bit to the side of your interpretation of it. It is tricky to word it right. My question would be in regard to a person actually returning to the BIOS to reset DQS to the default [SKIP] option. So he has already had DQS [PERFORM] a number of times, supposedly establishing the right RAM settings for itself ("training"). Now he chooses to SKIP DQS training. I think, if he does this, the RAM settings won't stay with their "trained" results, but instead will default right back to the manufacturers' best guess based on standard hardware expectations.

But you are right. Once DQS training has been applied and it has run once, it SEEMS to take no time at all. So maybe it calculates the settings and then it can very quickly confirm they are OK in subsequent startups. In that case, all the theory becomes moot, because DQS training can be left at [PERFORM] and there will be no speed issue for the boot process, exactly as you say.

@Lsdmeasap, many thanks for introducing me to Artdrivers.

You think 8GB of RAM worries YOU! It REALLY worries me!! Well, my RAM is running as DDR2-800 and so far it always has been. This is DDR2-1000 RAM, and it would be nice to OC it to that, but I have not made any CPU adjustments so far. My AMD 5600+ theoretically has a max of DDR2-800, but I believe it could be OCd. The G.Skill RAM technician informs me that it would be the CPU that is holding my system back. He feels that the RAM itself is fine, and he bases that in part on the fact that Windows works without any glitches at all. Apparently a RAM issue wouldn't disappear as soon as the system was up into Windows.

[artdrivers]

Re: BIOS > DQS Training Question

I also performed this function when I have it set to where I want it. I will set DQS to skip, and then press F11 When cold boot everything should be as it was when DQS training was performed.
F11: Save CMOS to BIOS
Right? I had questions about DQS training. I googled to solve them and thought I would share---->

http://www.techarp.com/showfreebog.aspx?lang=0&bogno=433

[Groonx2]

Re: BIOS > DQS Training Question

@Artdrivers, thanks again! You and I have done a few "parallels" researching our systems, because I was there too! It is the very last sentence that made me think I might do well to run the DQS training:

The boot-time calibration will ensure proper alignment of the DQ-DQS signals for non-standard clock speeds, or memory modules.

http://download.micron.com/pdf/technotes/ddr2/TN4711.pdf

and

http://download.micron.com/pdf/technotes/ddr2/TN4723.pdf

[artdrivers]

Re: BIOS > DQS Training Question

I had this ram at 1100 without performing DQS Training. This motherboard is only rated to go to DDR2 1066. That's was the peak for me and it performed cooler with 2 1GB modules then it does with 4. I did have a slight wait at bootup when performing DQS training the first time with the cpu frequency OCed to 208Mhz & the dram multiplier at 5.33. It still locked up in games a couple of times so I set it back to 1066Mhz and set the perform DQS training back to Skip. I hope this helps. One more thing. Because of the Black Box edition. I still could set the cpu muliplier at X12.5 or even X13 but locked at X13 so it was not the cpu. I have had the Memory at 800Mhx with the bus at 240Mhz with the cpu multiplier at 10. But whats the use. It is much better for me to set the cpu multiplier to X12.5 and run the memory at 1066 for my OC or just the x11.5
at 1066 with cool and quiet on for browsing to save energy.

[Groonx2]

Re: BIOS > DQS Training Question

Well, a bit off-topic to begin with, I noticed your F10 BIOS revision, so I have downloaded it and I can easily do the flash. AMD has only the teensiest of descriptions for this revision, so I'm not sure whether I'm supposed to use it with my components or not. But I think I'll take the chance, as I'm certainly looking for improvements.

So you and I have the very same motherboard! I bought mine so I could upgrade to a Phenom eventually. If I can finally get the performance I want from this, I'll be happy. In theory my RAM should work, according to NCIX and G.Skill, but Gigabyte (when I contacted them about an RMA, which I could do) has no opinion at all.

[artdrivers]

Re: BIOS > DQS Training Question

I believe your boot time is due to drive configuration.
Lets check your drives? Do you have your SATA DVD drive on SATAII_4? This is the last internal SATA connector.

[Groonx2]

Re: BIOS > DQS Training Question

You are fortunate. I think it might be related to my Athlon 64x2 5600+ versus your Phenom. The G.Skill tech person said my CPU is the constraint. 8GB of RAM is tricky, but testing with just one stick didn't seem to help much; nor did testing without most of the hardware. My guess (just a guess) is that my system takes extra time at startup to get the mobo, RAM, and CPU working together just right.

I am also making enquiries about OCing the CPU to handle the RAM @ DDR2-1000. I have a hunch that that would MAYBE make things EASIER on the system. That sounds strange, I know, but I'd like to try it. So far I don't know enough about which settings should be changed to OC for that.

[artdrivers]

Re: BIOS > DQS Training Question

OCing your AMDX2 should not be difficult, but finding the drive that is causing your slow boot should be performed first.

[Groonx2]

Re: BIOS > DQS Training Question

So you think it is a drive, then?

Well, I've gone through that before, but I can do it again. Especially as I'm going to flash the BIOS with the new F10. Usually I remove the CMOS battery overnight with the the new BIOS all set to go on a floppy. In the morning I put the battery back in and do the BIOS upgrade, back-to-back. Then I set to Optimized Defaults, and after using F10 with that I return to the BIOS to put in my own settings and save via F11.

This is quite a lot of tweaking, so maybe I should try the system first with no drives, and work through that. I'm not expecting it's a drive, but I'd be happy to find the source of the problem, so I'm okay on looking in any direction.

[artdrivers]

Re: BIOS > DQS Training Question

I am only suggesting this because I have seen it previously here. Do you have everything set as IDE? Do you use AHCI? AHCI causes a boot performance degrade. That USB drive may be the culprit. You have alot of drives and memory which would take alot of power at spinup. Especially if the motherboard and parts have had time to fully discharge and cool down. A test of the drives should be performed from time to time as well as the PSU. Even with trouble shooting set aside, but you are trouble shooting.

[Groonx2]