[AMD 16h / ASUS AM1I-A] 1866MHz CL9 RAM runs only as 1333MHz CL9 - how to fix?

[Mike Banon]

      A pair of 1866MHz CL9 RAM modules* runs only as 1333MHz CL9 on 16h
AM1I-A with coreboot is installed, but worked faster when a
proprietary UEFI was installed. To fix this "turtle RAM" coreboot
problem I tried to play with buildOpts.c -
https://review.coreboot.org/c/coreboot/+/33920 , but the things like
"#define BLDCFG_MEMORY_CLOCK_SELECT DDR1866_FREQUENCY" sadly did not
help.

Any ideas how to improve the RAM speeds? How I can force this RAM to run faster?

Best regards,
Mike Banon

[*] Crucial Ballistix Tactical Series DDR3 1866MHz CL9 (PC3-14900
9-9-9-24) UDIMM 240-Pin modules, part number BLT8G3D1869DT1TX0

[Kyösti Mälkki]

      On Wed, Jul 3, 2019 at 8:54 PM Mike Banon <mikebdp2(a)gmail.com&gt; wrote:
 
 A pair of 1866MHz CL9 RAM modules* runs only as 1333MHz CL9 on 16h
 AM1I-A with coreboot is installed, but worked faster when a
 proprietary UEFI was installed. To fix this "turtle RAM" coreboot
 problem I tried to play with buildOpts.c -
 https://review.coreboot.org/c/coreboot/+/33920 , but the things like
 "#define BLDCFG_MEMORY_CLOCK_SELECT DDR1866_FREQUENCY" sadly did not
 help.

 Any ideas how to improve the RAM speeds? How I can force this RAM to run faster?
Maybe AGESA debugging / IDS tracing is of some assistance? I have not
tried it for a while, though.

https://review.coreboot.org/c/coreboot/+/15320/7

Kyösti

[Mike Banon]

      Thank you for advice. I followed the instructions of this change, and
after fixing a few compilation errors (had to replace a few %x with
%llu at printf's) - using the same .config - I got a ROM which is
unbootable! Maybe because I don't have AMD HDT Debugger, and it
should've been connected to some usually-not-soldered header for this
ROM to boot?

Perhaps I can manually redirect these IDS prints to a standard
coreboot log - if that will give some useful info. Or I could dive
into AGESA and replace all DDR1333 stuff with a DDR1866 one, to force
it running as 1866MHz CL9 - since that "#define
BLDCFG_MEMORY_CLOCK_SELECT" seemingly doesn't work for some reason.

Do you have any other ideas to fix this "turtle RAM" AMD 16h problem?

Best regards,
Mike Banon


On Fri, Jul 5, 2019 at 8:44 AM Kyösti Mälkki <kyosti.malkki(a)gmail.com&gt; wrote:
>
> On Wed, Jul 3, 2019 at 8:54 PM Mike Banon <mikebdp2(a)gmail.com&gt; wrote:
> >
> > A pair of 1866MHz CL9 RAM modules* runs only as 1333MHz CL9 on 16h
> > AM1I-A with coreboot is installed, but worked faster when a
> > proprietary UEFI was installed. To fix this "turtle RAM" coreboot
> > problem I tried to play with buildOpts.c -
> > https://review.coreboot.org/c/coreboot/+/33920 , but the things like
> > "#define BLDCFG_MEMORY_CLOCK_SELECT DDR1866_FREQUENCY" sadly did not
> > help.
> >
> > Any ideas how to improve the RAM speeds? How I can force this RAM to run
faster?
> >
>
> Maybe AGESA debugging / IDS tracing is of some assistance? I have not
> tried it for a while, though.
>
> https://review.coreboot.org/c/coreboot/+/15320/7
>
> Kyösti

[Kyösti Mälkki]

      On Fri, Jul 5, 2019 at 5:36 PM Mike Banon <mikebdp2(a)gmail.com&gt; wrote:
 
 Thank you for advice. I followed the instructions of this change, and
 after fixing a few compilation errors (had to replace a few %x with
 %llu at printf's) - using the same .config - I got a ROM which is
 unbootable! Maybe because I don't have AMD HDT Debugger, and it
 should've been connected to some usually-not-soldered header for this
 ROM to boot?
You don't need HDT debugger. At least I have not experienced IDS
troubles because of not having one connected. But go through the IDS
options...

  Perhaps I can manually redirect these IDS prints to a
standard
 coreboot log - if that will give some useful info. Or I could dive
 into AGESA and replace all DDR1333 stuff with a DDR1866 one, to force
 it running as 1866MHz CL9 - since that "#define
 BLDCFG_MEMORY_CLOCK_SELECT" seemingly doesn't work for some reason.
That's what is supposed to happen already, it should dump all IDS
debug on serial console. Or whatever console you have enabled, IDS is
routed to printk(). Boots will be slow due the amount of data, maybe
30 seconds or so to get past ram training. There's lots of filtering
inside AGESA you can adjust.

Kyösti

[Mike Banon]

      Sorry it took me so long: using FT232H and screen Ctrl+A h option,
finally I got a complete boot log on AM1I-A Fam16h with these IDS
options enabled. Please could you take a look, to see if it contains
any hints why this 1866MHz CL9 RAM has been initialized as 1333MHz CL9
only?

P.S. This build is actually bootable but takes about 25 minutes to reach SeaBIOS


On Sun, Jul 7, 2019 at 3:34 PM Kyösti Mälkki <kyosti.malkki(a)gmail.com&gt; wrote:
 
 On Fri, Jul 5, 2019 at 5:36 PM Mike Banon <mikebdp2(a)gmail.com&gt; wrote:
 
 Thank you for advice. I followed the instructions of this change, and
 after fixing a few compilation errors (had to replace a few %x with
 %llu at printf's) - using the same .config - I got a ROM which is
 unbootable! Maybe because I don't have AMD HDT Debugger, and it
 should've been connected to some usually-not-soldered header for this
 ROM to boot?
 You don't need HDT debugger. At least I have not experienced IDS
 troubles because of not having one connected. But go through the IDS
 options...

  Perhaps I can manually redirect these IDS prints
to a standard
 coreboot log - if that will give some useful info. Or I could dive
 into AGESA and replace all DDR1333 stuff with a DDR1866 one, to force
 it running as 1866MHz CL9 - since that "#define
 BLDCFG_MEMORY_CLOCK_SELECT" seemingly doesn't work for some reason.
 That's what is supposed to happen already, it should dump all IDS
 debug on serial console. Or whatever console you have enabled, IDS is
 routed to printk(). Boots will be slow due the amount of data, maybe
 30 seconds or so to get past ram training. There's lots of filtering
 inside AGESA you can adjust.

 Kyösti

[Kyösti Mälkki]

      On Sun, Nov 10, 2019 at 3:52 PM Mike Banon <mikebdp2(a)gmail.com&gt; wrote:
 
 Sorry it took me so long: using FT232H and screen Ctrl+A h option,
 finally I got a complete boot log on AM1I-A Fam16h with these IDS
 options enabled. Please could you take a look, to see if it contains
 any hints why this 1866MHz CL9 RAM has been initialized as 1333MHz CL9
 only?
Not really, I expect it to involve several hours of work with little or no gain.

Kyösti

[Mike Banon]

      No problem (this log is really huge for some reason). At the beginning
of it - the most interesting part - I've found that "MemBusFreqLimit :
800" and "MemoryClockSelect : 667". Perhaps I'd try again forcing
either 933 (for 1866, although don't know if possible on this
platform) or 800 (for 1600) at the vendorcode. Another interesting
thing is that memtest86+ is failing but all the OS are working
correctly.

On Sun, Nov 10, 2019 at 7:16 PM Kyösti Mälkki <kyosti.malkki(a)gmail.com&gt; wrote:
>
> On Sun, Nov 10, 2019 at 3:52 PM Mike Banon <mikebdp2(a)gmail.com&gt; wrote:
> >
> > Sorry it took me so long: using FT232H and screen Ctrl+A h option,
> > finally I got a complete boot log on AM1I-A Fam16h with these IDS
> > options enabled. Please could you take a look, to see if it contains
> > any hints why this 1866MHz CL9 RAM has been initialized as 1333MHz CL9
> > only?
>
> Not really, I expect it to involve several hours of work with little or no gain.
>
> Kyösti

[Sergej Ivanov]

      I have same issue, memtest86+ failed on all fam16 family for me(AM1ML, AMD
devboard and so on) (usually in first 10Mb and near a dedicated video
memory). BTW i'm choose to use last free memtest86 (without plus), it
work's.

вс, 10 нояб. 2019 г. в 21:07, Mike Banon <mikebdp2(a)gmail.com&gt;:

  No problem (this log is really huge for some reason).
At the beginning
 of it - the most interesting part - I've found that "MemBusFreqLimit :
 800" and "MemoryClockSelect : 667". Perhaps I'd try again forcing
 either 933 (for 1866, although don't know if possible on this
 platform) or 800 (for 1600) at the vendorcode. Another interesting
 thing is that memtest86+ is failing but all the OS are working
 correctly.

 On Sun, Nov 10, 2019 at 7:16 PM Kyösti Mälkki <kyosti.malkki(a)gmail.com&gt;
 wrote:
 
 On Sun, Nov 10, 2019 at 3:52 PM Mike Banon <mikebdp2(a)gmail.com&gt; wrote:
 
 Sorry it took me so long: using FT232H and screen Ctrl+A h option,
 finally I got a complete boot log on AM1I-A Fam16h with these IDS
 options enabled. Please could you take a look, to see if it contains
 any hints why this 1866MHz CL9 RAM has been initialized as 1333MHz CL9
 only?
 Not really, I expect it to involve several hours of work with little or
 no gain.
 
 Kyösti
 _______________________________________________
 coreboot mailing list -- coreboot(a)coreboot.org
 To unsubscribe send an email to coreboot-leave(a)coreboot.org

-- 
Мастерство Поклона 7 \\ Увеличения нападают на власть, используя поклоны.

[Mike Banon]

      I've fixed 45 format-related warnings/errors of IDS_HDT_CONSOLE at AMD
16h vendorcode. Now a coreboot for fam16 could be built successfully
with IDS_HDT_CONSOLE enabled (IDSOPT_IDS_ENABLED TRUE and
IDSOPT_TRACING_ENABLED TRUE at board/OptionsIds.h) even with config
WARNINGS_ARE_ERRORS. Joined errors log (before a fix) could be seen at
attachments and here - https://pastebin.com/RZQBvTEc , will upload a
patch to review.coreboot.org ASAP. If it's successful, I could try
doing the same for the other AGESA platforms.

P.S. IDS_HDT_CONSOLE could be really helpful for debugging, although
so much logging causes a board to boot very slowly - i.e. 25 minutes.
These logs could be captured using USB EHCI dongle like FT232H.

On Sun, Nov 10, 2019 at 6:16 PM Sergej Ivanov <getinaks(a)gmail.com&gt; wrote:
 
 I have same issue, memtest86+ failed on all fam16 family for me(AM1ML, AMD devboard and
so on) (usually in first 10Mb and near a dedicated video memory). BTW i'm choose to
use last free memtest86 (without plus), it work's.

 вс, 10 нояб. 2019 г. в 21:07, Mike Banon <mikebdp2(a)gmail.com&gt;:
 
 No problem (this log is really huge for some reason). At the beginning
 of it - the most interesting part - I've found that "MemBusFreqLimit :
 800" and "MemoryClockSelect : 667". Perhaps I'd try again forcing
 either 933 (for 1866, although don't know if possible on this
 platform) or 800 (for 1600) at the vendorcode. Another interesting
 thing is that memtest86+ is failing but all the OS are working
 correctly.

 On Sun, Nov 10, 2019 at 7:16 PM Kyösti Mälkki <kyosti.malkki(a)gmail.com&gt; wrote:
 
 On Sun, Nov 10, 2019 at 3:52 PM Mike Banon <mikebdp2(a)gmail.com&gt; wrote:
 
 Sorry it took me so long: using FT232H and screen Ctrl+A h option,
 finally I got a complete boot log on AM1I-A Fam16h with these IDS
 options enabled. Please could you take a look, to see if it contains
 any hints why this 1866MHz CL9 RAM has been initialized as 1333MHz CL9
 only?
 Not really, I expect it to involve several hours of work with little or no gain.

 Kyösti
 _______________________________________________
 coreboot mailing list -- coreboot(a)coreboot.org
 To unsubscribe send an email to coreboot-leave(a)coreboot.org


 --
 Мастерство Поклона 7 \\ Увеличения нападают на власть, используя поклоны.

[Mike Banon]

      I've retested and it seems that my errors with memtest86+ 5.01 floppy
happen only if there are any other floppies than this memtest. Errors
are at bfce2c80 ( 3068.1MB ). Although I don't have this error on f15h
board, perhaps need to somehow improve my unofficial multiple_floppies
SeaBIOS patch.

On Sun, Nov 10, 2019 at 9:16 PM Sergej Ivanov <getinaks(a)gmail.com&gt; wrote:
>
> I have same issue, memtest86+ failed on all fam16 family for me(AM1ML, AMD devboard
and so on) (usually in first 10Mb and near a dedicated video memory). BTW i'm choose
to use last free memtest86 (without plus), it work's.
>
> вс, 10 нояб. 2019 г. в 21:07, Mike Banon <mikebdp2(a)gmail.com&gt;:
>>
>> No problem (this log is really huge for some reason). At the beginning
>> of it - the most interesting part - I've found that "MemBusFreqLimit :
>> 800" and "MemoryClockSelect : 667". Perhaps I'd try again
forcing
>> either 933 (for 1866, although don't know if possible on this
>> platform) or 800 (for 1600) at the vendorcode. Another interesting
>> thing is that memtest86+ is failing but all the OS are working
>> correctly.
>>
>> On Sun, Nov 10, 2019 at 7:16 PM Kyösti Mälkki <kyosti.malkki(a)gmail.com&gt;
wrote:
>> >
>> > On Sun, Nov 10, 2019 at 3:52 PM Mike Banon <mikebdp2(a)gmail.com&gt;
wrote:
>> > >
>> > > Sorry it took me so long: using FT232H and screen Ctrl+A h option,
>> > > finally I got a complete boot log on AM1I-A Fam16h with these IDS
>> > > options enabled. Please could you take a look, to see if it contains
>> > > any hints why this 1866MHz CL9 RAM has been initialized as 1333MHz CL9
>> > > only?
>> >
>> > Not really, I expect it to involve several hours of work with little or no
gain.
>> >
>> > Kyösti
>> _______________________________________________
>> coreboot mailing list -- coreboot(a)coreboot.org
>> To unsubscribe send an email to coreboot-leave(a)coreboot.org
>
>
>
> --
> Мастерство Поклона 7 \\ Увеличения нападают на власть, используя поклоны.