AM3 RAM OC?

[Alpi]

d

Cr4 vs 990Xa-ud5  spi32

Edited November 4, 2018 by Alpi

[Noxinite]

But the Crosshair IV isn't UEFI (has no mouse control in bios?)? Or am I missing something.

[speed.fastest]

Probably it just settings on some bios by default that causing performance reduced, like Execute Disable Bit it must be disabled to improve SuperPi performance, if bios dont have that options you can disable it through boot.ini by add /noexecute=alwaysoff
And many other that we might not know, but generally its like this, not because its UEFI or its Classic BIOS.

[Noxinite]

Yeah, I manually set that in OS. Didn't test how big a difference it made though. I guess I'll return at some point to figure out my crap efficiency...

[I.nfraR.ed]

Well, it might be due to board layout, hidden latencies and timings that can't be set manually. I don't exactly why, but crosshair iv is just slower.

You might try some old bios, but I could never match Crosshair III even with higher memory clocks. It was a long time ago, so I might have missed something back then, however that's my observation. Haven't tested all the boards out there though.

I also have a 890G Asus board which is even slower and has less options in bios.

The problem with newer Gigabyte revisions is the microcode which is Orochi.PI and not the old AGESA anymore, so old AM3 CPUs run in "compatibility" mode. That's what I think.

The 990FXA-UD3 rev3 I have is a POS, looses dual channel above ~1700 mem even with loose timings, can't clock FSB high, some settings don't  stick and overall has a crappy bios. I hate that board.

Unfortunately I have dislocated my shoulder 2 weeks ago, otherwise I would test CHIV, CHV and some giga boards.

 

PS:  I set noexecute no matter if it is exposed in bios or not.

Edited November 4, 2018 by I.nfraR.ed

[Bullant]

Yeah, couple things Ive notice on both boards 890- ud7 and 990- ud3 is on all cpus if im pushing around 940 memory and big NB freq along with high cpu-NB volts I have CB -160 and sometimes loose channel "lost channel mainly on ud7". Just dropping cpu-NB volts down a little when this happens I can the go full pot

 

Some cpus are worse then others

Edited November 5, 2018 by Bullant

[Bullant]

Just testing the 890fx UD5 with PSC air and 1090t on SS ....I run this board up with cold PSC, should go nice

 

Edited November 20, 2018 by Bullant

[Bullant]

Ran a 5Ghz just once, so would need to test few more things to compare

 

[I.nfraR.ed]

10 hours ago, Bullant said:

Ran a 5Ghz just once, so would need to test few more things to compare

Wow, you beat me by whole 6 seconds. I should really play with my older boards.

[Bullant]

3 hours ago, I.nfraR.ed said:

Wow, you beat me by whole 6 seconds. I should really play with my older boards.

You should mate, I'll try  post some ln2 runs hopefully this weekend 

[I.nfraR.ed]

I can't even touch my old scores, something must be wrong with the OS.

Crosshair III is faster than 970A-UD3, but not much. Found 890FXA-UD5, hopefully not DOA, will see tomorrow.

Does anybody know what is the difference between board revisions?

rev2.x is rated for 1866+ DDR3, while 3.x is 2000+, some color difference and white vs black socket. Apart from that I can't see a difference.

Edited November 22, 2018 by I.nfraR.ed

[Bullant]

1 hour ago, I.nfraR.ed said:

I can't even touch my old scores, something must be wrong with the OS.

Crosshair III is faster than 970A-UD3, but not much. Found 890FXA-UD5, hopefully not DOA, will see tomorrow.

Does anybody know what is the difference between board revisions?

rev2.x is rated for 1866+ DDR3, while 3.x is 2000+, some color difference and white vs black socket. Apart from that I can't see a difference.

From what I've seen and tested....,the difference between rev2 vs rev3 is "rev2" uses 3-4 slots close together and "rev3 has the more modern 2-4 of set slot usage 

That being said 890fx that runs memory 2-4 slots and 990fx can run 4x dims at good memory freq....no boost in 32m but good boost in some benches I've eff tested 

I do have both ud5 890fx revs that use the 3-4 and 2-4 slot setups, Ill have to test both to see if memory and eff  difference 

My 890fx ud7 rev2 using 3-4 slots was doing a little better stable memory clock then the ud5 I've posted yesterday, that being said the ud5 I've not tried and tested other bios yet 

Edited November 22, 2018 by Bullant

[Alpi]

I bought a 960T. Yesterday I made some tests in my 990xa-ud5 and found that not the highest mem freq gives the best effs. Between 1750 - 1850 gives good bandwith and very low latencys too. I'm sure it's changing depend on cpu freq / benchmarks / tweaks even boards and / or bioses. 

http://hwbot.org/submission/3989851_

[Bullant]

Ran again on a different Gigabyte board, would need to look over bios more to compare with other board, just in case Ive done them different. It all takes time lol

 

 

[Bullant]

Will have ln2 on Friday, so I'll run up the CPU and memory "PSC" both ln2 Should see nice boost in eff both from cache and higher and tighter memory freq

Will gradually drop memory temperature while working out dram scaling 

will post results and findings 

[Bullant]

Just a quick one tonight, memory -5....Will push some more over the weekend

 

Edited November 30, 2018 by Bullant

[Bullant]

Ran a couple 4Ghz runs, memory -100

 

 

 

 

[Alpi]

Awesome things, awesome posts !! What was the problem with 2200 ram ? Imc or some timing ? tRas is very low compares to a usual psc timing but obviously i haven't any exp with this ic and Am3 cpus. 

[Bullant]

5 hours ago, Alpi said:

Awesome things, awesome posts !! What was the problem with 2200 ram ? Imc or some timing ? tRas is very low compares to a usual psc timing but obviously i haven't any exp with this ic and Am3 cpus. 

Hey, yeah the problem I have is for me dram volts not scale higher then say 1.87v, both psc and hypers. I've only tested one kit of hypers and two kits of PSC...

 

Yeah was only quick testing but yeah if I loosen tras to say 24 it would help a little, with maybe loosing some eff. That 2200 run was with -45 1.87v, it even ran a little at -100.... 2200mhz 6-8-6-21-18, very fast but dram volts limit 1.87v 

 

will play around little more more with some volt settings to see if any help ?

[Bullant]

Ran the 965 BE on the ud7....was only quick session, should drop few more sec with more memory and cache

[Bullant]

Went to test this chip for freq, not great...so ran some more 5G....UD7...Hypers 1.87v

The bottom fail 13m.11 xxx, I say and maybe 3 sec faster on the UD5, will run that up another time

 

 

 

[Bullant]

So it was cache freq holding it back, I adjusted some volts and lowered v core a little to get the pass

 

[I.nfraR.ed]

A whopping 14 seconds faster than my 5G run. I didn't spend much time on that, perhaps a single run, but still a big difference which don't think I can come up close at all.

Wonder what I'm missing. Also interesting that at loop #13 you're already half a second faster compared to the failed run, although frequencies are exactly the same.

What bugs me is the real memory vs available real memory and no way to figure out how you get that efficiency. Obviously still a lot to learn

PS: Waiting for 2 more boards, the 890FXA-UD5 rev 2.0 only likes one set of SuperT Hypers I have. The first board that didn't like the GTX2...

I always thought my eff is decent enough, especially compared to most of the guys at the top of the AM3 rankings. You can destroy the K10 32M if you find a good enough CPU.

Edited December 7, 2018 by I.nfraR.ed

[Bullant]

3 hours ago, I.nfraR.ed said:

A whopping 14 seconds faster than my 5G run. I didn't spend much time on that, perhaps a single run, but still a big difference which don't think I can come up close at all.

Wonder what I'm missing. Also interesting that at loop #13 you're already half a second faster compared to the failed run, although frequencies are exactly the same.

What bugs me is the real memory vs available real memory and no way to figure out how you get that efficiency. Obviously still a lot to learn

PS: Waiting for 2 more boards, the 890FXA-UD5 rev 2.0 only likes one set of SuperT Hypers I have. The first board that didn't like the GTX2...

I always thought my eff is decent enough, especially compared to most of the guys at the top of the AM3 rankings. You can destroy the K10 32M if you find a good enough CPU.

Yeah the faster second run I think was my waza, every run I tried after that failed run was faster then the failed run. Even tho start time is the same, my waza does vary

 

Yeah I have a ud5 too that not like my hypers but it likes the psc, I thought it might have been bios, I've yet to try another bios.

 

Have you tested the ud5 on eff or is the memory that works on it not great? 

 

 

 

 

[I.nfraR.ed]

4 minutes ago, Bullant said:

Have you tested the ud5 on eff or is the memory that works on it not great.

With the SuperTalent kit I can run same settings as on the 970A-UD3 and Crosshair III.

890FXA-UD5 is a little faster than the 970A, but nothing dramatic in terms of eff. Mem overclock is a little harder on both CH3 and UD5.

No way I can drop off 10 seconds with just a board/bios change.

Plus I can't even beat my old 4G score, where I didn't even tried hard: http://hwbot.org/submission/2990939_i.nfrar.ed_superpi___32m_phenom_ii_x4_965_be_16min_16sec_641ms