AM3 RAM OC?

[Bullant]

37 minutes ago, I.nfraR.ed said:

Nice! Wish I knew what are these adjustments ?

I have tuned timings as well as I could (bios + software) and don't know what else to do to get some seconds off.

Still no "new" boards, hopefully will get at least one in the next couple of days.

If I can find a Asus am3 cheap enough I buy to test and compare, I think 955-965 is also very close eff as turban if memory freq is all equal.......I think for turban PSC might still produce the fastest run, with a touch of cold on the memory

[I.nfraR.ed]

On 12/27/2018 at 9:05 AM, Bullant said:

If I can find a Asus am3 cheap enough I buy to test and compare, I think 955-965 is also very close eff as turban if memory freq is all equal.......I think for turban PSC might still produce the fastest run, with a touch of cold on the memory

Yes, PSC@1100 and tight timings should be faster than Hyper@970C6. Can't get Hypers much above 1000 with this platform.

Meanwhile, I got the "new" 890FXA-UD5, but it is rev2.1 and it's not different than the rev2.0 I have, apart from the slightly different layout (alternating mem slots). Still can't boot 2200 RAM with Thuban, still can't do high HTRef (for Sempron benching). Installed the beta bios for FX CPUs and HTRef is stuck at 209MHz  no matter what I set above that. Core control also does not work. Have to find a 3.x revision to try. Should have the 990FXA-UD5 rev 1.1 this week, though. Hopefully works much better.

Also noticed these 890FX Gigabyte boards can't clock NB (uncore) as high as the Crosshair IV and V. Maybe some problem with straps or too tight/wrong max read latency. If it's a bios problem, then it's valid for both 2.0 and 2.1 since they use the same bios versions.

Edited January 2, 2019 by I.nfraR.ed

[Bullant]

Yeah I've only tested couple kits of psc but my Gigabyte boards won't do 2200 air, I need cold memory to do it, well if I want any sort of efficiency 

 

The NB issue you said, is this just testing what your chip nb does on air comparing the different boards 

[I.nfraR.ed]

Yes, same settings, same temperature (chiller set at 18°C water temp). On Gigabyte 3600 1.45V CPU-NB doesn't post, while it is stable on the Crosshair V.

Real volts can't be that different.

PS: Can't find the chart I did when Thuban launched. I tested CPU-NB scaling with voltage on my sample. It's still the same CPU - a very good one for water, but doesn't scale too well with LN2. Has a great IMC and CPU-NB though. It was running 24/7 stable 4270MHz 1.44V/ 3230MHz 1.33V CPU-NB for a couple of years as my daily driver. I have another 1090T that's untested on LN2, but has weaker cores and cpu-nb on water.

PS2: Here's DDR3-2400 single channel validation with this CPU: 

https://valid.x86.fr/show_oc.php?id=2171728

https://d1ebmxcfh8bf9c.cloudfront.net/u36113/image_id_703725.jpeg

Edited January 3, 2019 by I.nfraR.ed

[Bullant]

Yeah the 2400 is nice...I've not tried that....I think I would need my mem to be cold on this board push up that high. The 1090t I have need 1.55v CPU-NB volts a least for 3600 water on these boards 

[Noxinite]

I forsee more people benching AM3 in Rev 8, so maybe this thread will get more busy.

Edited January 4, 2019 by Noxinite

[I.nfraR.ed]

Modded the F4 bios for 890FXA-UD5 rev2.x and 2200 now works, but it's unstable. Can't solve the CPU-NB issue.

Will try to fix the HtRef issue of F7e beta bios.

[Bullant]

Nice, how's that board efficiency wise compare to your Asus boards? 

[Bullant]

I also have the rev 2.1, testing it for a little while now, its quite efficient......low memory here along with low NB freq. Im on my SS as no LN2 yet, dram 1.89v, board and this bios seems quite good on dram volts.....yes NB freq is little harder at all same settings as my other boards.

 

 

[Bullant]

 

More 32m 

[Bullant]

So had a quick play PSC on cold, running 4GHz and 38xx NB...on the 1090t and 2200Mhz 6-9-6-21--24....getting about 6-12 loops then shutting down, times in the 15m-57s if get it to pass. Has so much more in it just a matter putting it all together

 

 

Edited January 26, 2019 by Bullant

[Bullant]

Have a Gigabyte 870A UD3 coming to try, interesting to see how efficient it is....I'll run it up when I have it 

[Bullant]

Tested a new 965 tonight, seems better core, NB and imc then my other 6.6Ghz chips....it runs close to 940-950 mem tight and 5350 core on my SS -35c.....will run it up on ln2 later

 

Edited February 13, 2019 by Bullant

[TaPaKaH]

Some weeks ago I decided to return to AM3 but wasn't sure which board to buy. In 2010-2011 I had good experience with ASUS M4A79T Deluxe, but following Infrared's advice, thought I'd give some Gigabyte 9x0 boards a try as well. So I ended up with the following selection:
- ASUS M4A79T Deluxe
- ASUS Crosshair III Formula
- ASUS Crosshair V Formula (added 27 April 2019)
- Gigabyte 970A-UD3 ver 1.0
- Gigabyte 970A-UD3 ver 1.2
All tests done using the same random Phenom 955 chip I got along with one of the boards, plus Corsair Hypers some of which got damaged along the way.

Test #1: CPU overclocking - minimal voltage. Used 250x18 with same NB/RAM settings for reference.
ASUS M4A79T => 1.48V
ASUS C3F =====> 1.49V
ASUS C5F =====> 1.51V
970A-UD3 v1.0 => 1.47V
970A-UD3 v1.2 => 1.47V

Test #2: Max CPU clocks. Used same preset as above, only with CPU multi upped to x18.5. Tried 1.5125, 1.5250, 1.5375 and 1.5500V set.
ASUS M4A79T => instaBSOD at 1.5125 and 1.5250, crash after few sec at 1.5375, passed 32M initial at 1.5500
ASUS C3F =====> 1.5125 crash before initial, 1.5250-1.5500 crash before first loop
970A-UD3 v1.0 => 1.5125-1.5500 crash before first loop
970A-UD3 v1.2 => 1.5125 crash after loop 1, 1.5250 crash after loop 2, 1.5500 crash after 14 loops

Test #3: Uncore. Used same preset as above with CPU clocks reduced to 250x16.
ASUS M4A79T => 2750 pass at 1.20V set, 3000 can't enter OS at 1.30-1.50V
ASUS C3F =====> 2750 pass at 1.24V read, 3000 can't enter OS at 1.30-1.50V
ASUS C5F =====> 2750 pass at 1.21V read, 3000 can't POST at 1.30-1.50V
970A-UD3 v1.0 => 2750 pass at 1.21V set (mind the +0.2V offset), 3000 can't enter OS at 1.30-1.50V
970A-UD3 v1.2 => 2750 pass at 1.23V set (mind the +0.2V offset), 3000 can't enter OS at 1.30-1.50V

Test #4: IMC using 2x2GB GTX2 or 2x2GB 2000C7 GT based on the kit that was most compatible / most alive at the time of testing
ASUS M4A79T => 950MHz
ASUS C3F =====> 953MHz
ASUS C5F =====> 949MHz
970A-UD3 v1.0 => 857MHz
970A-UD3 v1.2 => 863MHz
I've spent some time trying to get Gigabyte boards to get anywhere near 900MHz but that wasn't to happen. It could be down to compatibility on particular CPU/RAM but the deficiency in clocks still counts against GA as it potentially adds additional constraints when binning CPU or buying Hypers

Test #5: 32M efficiency. Used my CPU binning profile at 250x16 with memory settings as close as possible. Tested on two different OS (one with lots of stuff, one clean), straight run, no CW
ASUS M4A79T => 16:45.031 on dirty OS, 16:42.047 on clean OS
AUS C3F ======> 16:44.000 on dirty OS, 16:42.000 on clean OS
970A-UD3 v1.0 => 16:41.625 on dirty OS, 16:38.328 on clean OS
970A-UD3 v1.2 => 16:39.922 on dirty OS, 16:36.812 on clean OS
If you look at other tests done at same settings, you'll confirm that the Gigabyte boards are ahead of ASUS on efficiency.

Edited April 27, 2019 by TaPaKaH

[lanpartyut]

i think on Llano platform  elpdia hypers maybe run better，but don not have a test platform

[I.nfraR.ed]

On my 970A-UD3 I can't POST with Deneb with Hypers > 850 or so and have to work my way up in windows. No matter the settings, it always goes to fail-safe if I set memory clock at anything higher than 850. It seems Sam encountered the same. It's not an easy board to bench.

Recently I've bought 990FXA-UD3 rev 1.2 and it is very similar in efficiency, but setting mem frequency works as expected and can boot bench settings without the need of adjusting refclock within windows, which in fact doesn't always work on 970A - sometimes it hard-locks before reaching target frequency.

I've noticed 970A worked with 1090T and Hypers at 947C6 straight from bios, however the CPU died. Don't know the reason, but I suspect CPU-NB voltage. I was running 3550-3600 1.55V for extended period of time on chiller and positive temps. Perhaps the board overshoot it, or it just died due to the stress conducted from running at that voltage on other boards.

Edited March 8, 2019 by I.nfraR.ed

[subaruwrc]

epic how much time and effort you can still put into am3 systems  respect !

[TaPaKaH]

On 3/8/2019 at 9:41 AM, I.nfraR.ed said:

have to work my way up in windows.

How does one actually do that? SetFSB gives me a chipset error and EasyTune causes an instant system crash even if I try to change FSB by 1 MHz.

[I.nfraR.ed]

38 minutes ago, TaPaKaH said:

How does one actually do that? SetFSB gives me a chipset error and EasyTune causes an instant system crash even if I try to change FSB by 1 MHz.

I use AMD Overdrive and usually up 5 MHz at a time.

[Alpi]

Bought a CR4E a bit earlier. Spi32 needed for comp, it's good time to see how it performs. It was better when I don't know actually..

Did some runs with Asus, change the board to my 990FXA-UD5 and did the same thing yesterday. Same os and try to stay as close as just can.

 

[TaPaKaH]

Got a C5F and added some comparison results to this post:
https://community.hwbot.org/topic/185674-am3-ram-oc/page/5/?tab=comments#comment-526287

I wasn't unexpected for me to see that board being 7-10s behind peloton at Deneb 4GHz 32M. What was unexpected is that the board tends to clock the CPUs noticeably worse than other boards. And this isn't only limited to Denebs - I am having similar issues with three Visheras as well, with difference being around 50MHz on suicide clocks. Did anyone ever compare Gigabyte's 9x0 boards and ASUS's C5F-boards for raw clocks?

Edited April 27, 2019 by TaPaKaH

[Alpi]

Does the C5F has the huge clock oscillation like C4F ? I mean the base clock but even @ 5 ghz, You have to listen when You make a Ss if any clock limit. 30-50 mhz when multiplicated. Cpu-z shows mostly when idle but that's not mean anything. This fluctuation can cause differency up to 10 sec I guess, or is this bullshit ?

[TaPaKaH]

The clock readouts were pretty normal for me, cpu speed didn’t move by more than 1-2 mhz

[Bullant]

Any more testing @I.nfraR.ed ? I'll run some more next time have ln2 

[speed.fastest]

On 4/27/2019 at 11:12 PM, TaPaKaH said:

Got a C5F and added some comparison results to this post:
https://community.hwbot.org/topic/185674-am3-ram-oc/page/5/?tab=comments#comment-526287

I wasn't unexpected for me to see that board being 7-10s behind peloton at Deneb 4GHz 32M. What was unexpected is that the board tends to clock the CPUs noticeably worse than other boards. And this isn't only limited to Denebs - I am having similar issues with three Visheras as well, with difference being around 50MHz on suicide clocks. Did anyone ever compare Gigabyte's 9x0 boards and ASUS's C5F-boards for raw clocks?

Back then i got 990FXA-UD5 rev 1.0 and CVF, for cinebench i can do better on CVF because CVF had LLC while UD5 rev 1.0 dont have, but under same vcore load clock is same. Different is on memory, on 990FXA-UD5 on 2400 unstable and loosing channel, on CVF 2400 is stable, can even boot 2600.