960gm-U3S3: 667mhz Ram divider is wrong in BIOS 1.10&1.20

[Chozo4]

Greetings,

Just got working ram for my system so was able to start actually looking into overclocking these proper. My last sticks failed regardless of timings so left them on the lowest divider and as thus missed this issue initially. However, the timings in the BIOS for 667mhz is invalid to the actual settings applied. Running at full stock settings aside from the Ram Divider yield the following results:

Dividers (In-Bios) and Results (Out-of-Bios):
400mhz(800mhz) = 400mhz
533mhz(1066mhz) = 533mhz / 3:8
667mhz(1333mhz) = 533mhz / 3:8
800mhz(1600mhz) = 800mhz

Bios affected: 1.10 + 1.20

800mhz works perfectly fine as do 400/533 without issues and no fallback at all. However, selecting the 667mhz divider results in the same divider and timing as 533mhz. This does not match up with what the bios is showing in any form and is likely evident of a bug within the Bios GUI itself. It shows the divider fine visually in the bios but 667mhz outside the bios is a completely different result from within Memtest x86+ and CPU-Z (which show 533mhz). Is there any bios that has yet to be posted with this bug fixed at all? Cannot seem to find one beyond 1.20...

Relevant Specs:
CPU: AMD Opteron 1389 RevC2 @ 2.9ghz
RAM: 2x 4gb Samsung MV-3V4G3/US @ 1600mhz (800mhz)

[- wardog -]

Re: 960gm-U3S3: 667mhz Ram divider is wrong in BIOS 1.10&1.20

I'm betting what you're seeing is probably attributable to running an unsupported processor.

My 2cents FWIW anyways.

Have you contacted ASRock Tech Support about this? I'd be curious what they said.

[Chozo4]

Re: 960gm-U3S3: 667mhz Ram divider is wrong in BIOS 1.10&1.20

I'm betting what you're seeing is probably attributable to running an unsupported processor.

My 2cents FWIW anyways.

Have you contacted ASRock Tech Support about this? I'd be curious what they said.

I have yet to contact ASRock tech support as I'd thought I could get support through here (from both community and Emily perhaps) regarding if there was an unlisted beta bios update, workaround, or others experiencing the same problem. I'll look into contacting ASRock directly then and see what the outcome is. Will reply afterward when I get a response and/or to any further inquiries from others.

[rambling] Just figured that since all Deneb Cores are supported (and Opteron 1389 is simply a higher binned Rev-C2 Deneb/Phenom-II) that it would have inherited board support from the Phenoms. Granted, I've had zero issues whatsoever with the processor along with that it fully detects properly through bios. Would be rather.... odd for a mismatched divider from the bios having anything to do with the CPU Support List as if I'm correct it's dependent on the bios/board/BClk itself to be set regardless of the CPU. Had it been unsupported.. wouldn't it show as 'unknown processor' or wrong processor name during POST or even affected the other dividers as well rather than just 667mhz? All I use are opterons in consumer class boards so this would be a rather unique case of being unsupported regardless of the lists.[/rambling]

Thanks for the reply WarDog. ^^

Edit in: Query sent in to ASRock Support as suggested.

[- wardog -]

Re: 960gm-U3S3: 667mhz Ram divider is wrong in BIOS 1.10&1.20

Had it been unsupported.. wouldn't it show as 'unknown processor' or wrong processor name during POST or even affected ...........

Well .............<no answer on this other than maybe maybe not>

From what I can find Suzuka comes from/uses the same microcode as Deneb. But then I read on the official amd blog that Suzuka needed an update on his board.

Also, another FWIW while discussing the diff between Deneb and Suzuka(your 1389) the author of the blog in the comments posted
Quad-Core AMD Opteron

John Fruehe	September 8, 2009

The memory controller is very similar to the one the the Phenom processors as they share a common architecture.

I did site wide searches using Google for Suzuka on many MB manufacturers sites hoping to find board support and came up empty.

It doesn't look like they were widely distributed, or adopted. And being the first from amd to support DDR3 both the previous is surprising even if they are server silicon.

And another FWIW. You see what I see happening there in that CPU-z screenshot?
CPU-Z Validator 3.1

[Chozo4]

Re: 960gm-U3S3: 667mhz Ram divider is wrong in BIOS 1.10&amp;1.20

Well .............< no answer on this other than maybe maybe not >

From what I can find Suzuka comes from/uses the same microcode as Deneb. But then I read on the official amd blog that Suzuka needed an update on his board.

Also, another FWIW while discussing the diff between Deneb and Suzuka(your 1389) the author of the blog in the comments posted
Quad-Core AMD Opteron

John Fruehe	September 8, 2009

The memory controller is very similar to the one the the Phenom processors as they share a common architecture.

I did site wide searches using Google for Suzuka on many MB manufacturers sites hoping to find board support and came up empty.

It doesn't look like they were widely distributed, or adopted. And being the first from amd to support DDR3 both the previous is surprising even if they are server silicon.

And another FWIW. You see what I see happening there in that CPU-z screenshot?
CPU-Z Validator 3.1

Aye, I'd taken note of that CPU-Z result several times in the past prior to buying this Opteron (was either this or the Bulldozer-based Opteron). It's actually rather intriguing that it still identifies as such a high-model Phenom internally. Additionally I'd also come across that aritcle long ago during my initial research regarding it. Lack of results definately shows that Suzuka was in of itself a rather quiet release that wasn't talked of much which could explain why it never went through 'compatibility testing' by any manufacturers as you stated.

[parsec]

Re: 960gm-U3S3: 667mhz Ram divider is wrong in BIOS 1.10&amp;1.20

A small aside, if you are looking at the SPD tab of CPU-Z, with the 30nm Samsung memory I think you have, don't take CPU-Z as accurate information. CPU-Z can't read this memory's SPD data at all, and what it does show is quite humorous.

[Chozo4]

Re: 960gm-U3S3: 667mhz Ram divider is wrong in BIOS 1.10&amp;1.20

A small aside, if you are looking at the SPD tab of CPU-Z, with the 30nm Samsung memory I think you have, don't take CPU-Z as accurate information. CPU-Z can't read this memory's SPD data at all, and what it does show is quite humorous.

Of course - was basing purely off the [Memory] tab actually. I never bother with the SPD cept to retrieve the model/serial from the sticks as I'm rather used to seeing it show inaccurate frequencies and voltage. Under [SPD] the part number,serial,ManDate's are accurate along with the timings but not the frequencies or voltages. If you meant something far different from what I see - I'm curious to see what you meant!


My last sticks (which were bad the day I got them) would show 1.5v on 1.65v sticks under the [SPD] tab for instance while these Samsung show 1.28v with JEDEC frequencies not matching standards (609/685/761/838). Found thus far these Samsungs run fine at 800/1600 @ 8/8/8/16/1t - 90ns at only 1.35v.. 7/7/8/16/1t - 90ns at 1.45v.

[parsec]

Re: 960gm-U3S3: 667mhz Ram divider is wrong in BIOS 1.10&amp;1.20

No, that's what I meant about this memory, using CPU-Z. I'm a bit surprised at the timings you are able to run at 1600, but then I can run 9 9 9 24 1T at 2000, 16GB at 1.40V. This memory is crazy, but I like it!

[Chozo4]

Re: 960gm-U3S3: 667mhz Ram divider is wrong in BIOS 1.10&amp;1.20

Well - got a reply back from ASRock support with the following:

Please provides detail memory model that been using for the test.
Please try with another known good memory to eliminate the issue.
However, the CPU is not in the support list that also might be effect.

What gets me is that I DID provide the model#'s and so forth for the ram as "8g DDR3L Samsung Green MV3V4G3". However their point I did overlook was to try another set of ram sticks. I put back in my "known bad" Ramsticks instead as one was bad another good but since the test was only for BIOS/Memtest verification it wasn't neccessary for them to be stable. Needless to say the timings did infact come up proper and the issue is infact related to the ram itself. The sticks I used for testing were also DDR3 800mhz sticks (same stock speeds) but different voltage compared to the samsungs. One day I'll come across a processor I can bum for testing purposes later on to fully verify if it's a processor 'incompatibility' issue or not.

For what I wonder though the only real difference is the JEDEC frequencies specified between the two sets. Perhaps that's the reasoning behind the timings being rather... 'off'? Would figure since the frequencies on the sticks are in such a tight non-standard range it may be causing the bios to get confused as I've yet to see sticks with such strange factory frequencies stated. *shrugs* Perhaps this is more of an issue that should be taken up with Samsung regarding the sticks? It seems it isn't just CPU-Z reporting these JEDEC timings so they could be the actual stamped timings from the manufacturer afterall which are very much NOT standard by any means.

RamInfo-AIDA64 Extreme:


Also, as an aside it seems the my IMC (or board) might have issues with ram beyond 820mhz... higher by even 1mhz regardless of timings/voltages the bios shows both sticks but the second slot will not pass the POST Memory Test and silently continue with only 4096MB despite being 100% rock stable at 820mhz. Alas - running at 706mhz now due to upping the BClk to 265 & NB to 2650mhz for now so even then the 'skipped' divider is no longer valid for my needs anyway.

[Chozo4]

Re: 960gm-U3S3: 667mhz Ram divider is wrong in BIOS 1.10&amp;1.20

I know this is an old thread but I figured it best to provide an update as a closing statement for this topic. Issue is resolved with BIOS v1.30.

I had to replace the board for starters VIA the RMA process. Seems the (lack of) VRM's cannot handle the summer heat despite Enzotech copper mosfet sinks and effective airflow so it silently went dead during an unattended MemTest. Got the replacement in and noticed that there was apparently a BIOS update related to Windows 8 issues according to the 'notes' on the download page but it also included a silent fix for the Samsung memory modules. The divider that was at fault since version 1.10 and 1.20 has been resolved and now works as it is supposed to with the 1.30 BIOS release. No longer need to lose a fair bit of speed due to a borked divider dropping to the next one down.

As an aside, neither the memory modules nor the CPU have been added to either support lists as of yet and likely won't but just a nitpick more than anything so this issue being resolved as well down the line was/is a pleasant surprise.