AMD port microblog :-)

Moderator: BarsMonster

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

AMD port microblog :-)

Post by BarsMonster » Tue Oct 21, 2008 8:45 pm

Good new guys, today I've finally received new and shiny POWERCOLOR AX4870 :-) For the first time ever I am having some top of the line card, not middle or low end :-) (I am not counting x2 cards because of possible game lags, so this is not an option for me).

Why should I choose between nVidia & AMD? I choose both :crazy:

It feels so powerful :crazy: Unfortunately Thermaltake PurePower 460W PSU cannot work with it on full clocks (800 Core/1000 Memory), I can run heavy 3d apps only at 500/800 or something (provided that I had to downclock my CPU to 2.8Ghz@1.1v instead of 4Ghz@1.45v).
Damn, when I was buying this PSU like 3 years ago I was thinking that I am investing in PSU which could survive serious PC upgrades(actually it did), but it is not even 4870x2 :-)

AND IT IS HOT (even with non-stock cooler). I know GTX280 produces even more heat, I am scared to think about heat dissipation of GTX280X2 :-)

So I'll be starting AMD port very soon on downclocked card :crazy: But in the nearest 2 weeks I will be buying some 500-600W PSU :crazy:

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Tue Oct 21, 2008 9:02 pm

Here is it :-)
Image

User avatar
the_drag0n
Posts: 217
Joined: Thu Oct 02, 2008 6:48 am
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by the_drag0n » Wed Oct 22, 2008 12:49 pm

what a sweet shiny thingy ;)
hoping to get some updates soon ;)
btw what caused the downtime this morning ?

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Wed Oct 22, 2008 9:15 pm

He, I've just saved 150S on the new PSU :-)
It appeared that old-school PSUs with 2x 12v rails with 1 rail exclusevly for MB CPU power supply no longer works fine. As CPU uses just ~60W out of ~180-200W that can we drawn from that rail, PSU just does not let us use most of it's power. Not sure what is the reason to make PSUs with multiple rails (less ripple?), but I've just decided to make a single rail PSU our of my Thermaltake PurePower 460W :-)

Now it can handle even overclocked 4870(voltmodded to 1.35v) with overclocked CPU :-) Voltages are in the +-2% of the target values.
I was only worried if that small wire is ok to deliver up to 10-12A of current :-) Well, as it is short and Cu-based, it should be ok.

But that is not the end of engineering fun today :-)

Image

Sc00bz
Posts: 136
Joined: Fri Oct 03, 2008 8:28 am
Contact:

Re: AMD port microblog :-)

Post by Sc00bz » Wed Oct 22, 2008 9:31 pm

I think the reason for multiple rails is so that there is not too much current going through each wire. If there is one rail then you could technically have all the current going through one wire. The problem with that is it will be too much current for that wire which will cause problems if it continues for long periods of time.

User avatar
easteregg
Posts: 30
Joined: Mon Oct 13, 2008 2:39 pm
Contact:

Re: AMD port microblog :-)

Post by easteregg » Wed Oct 22, 2008 9:34 pm

freaky :D
i would never risk a brand new highend GPU with such selfmade solutions ;) respect dude!
<| 2 AMD Opterons 2384@ 8x3,2ghz | Tyan S2915 | 10GB 667REG ECC Ram | Nvidia 8800GT | Nvidia 8400GS | Maxtor Atlas II 15K |>

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Wed Oct 22, 2008 9:38 pm

the_drag0n wrote:btw what caused the downtime this morning ?
Most likely server problems here

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Wed Oct 22, 2008 9:42 pm

Sc00bz wrote:I think the reason for multiple rails is so that there is not too much current going through each wire. If there is one rail then you could technically have all the current going through one wire. The problem with that is it will be too much current for that wire which will cause problems if it continues for long periods of time.
Well, multiple rails means multiple independent 12v stabilizers and current limiters. Number of wires is the same.
I.e. before this thing PSU was triggering overcurrent shutdown when I was using 15A out of 1st 12v rail, now it should shutdown only when I am using more than 30A out of total 2 rails.

If you keep a single wire cool, you can pass even 100A through it :crazy: If that piece of copper wire has resistance of 0.01Ohm, and we pass 100A through it, it will dissipate just 1W of heat :-)

Well, It should not do any harm to GPU as voltages are quite precise - 12.11v idle, 11.98v at ultra heavy load (both CPU & GPU). The PSU itself was quite high quality.

Sc00bz
Posts: 136
Joined: Fri Oct 03, 2008 8:28 am
Contact:

Re: AMD port microblog :-)

Post by Sc00bz » Wed Oct 22, 2008 11:21 pm

This just seems like you are using "just stick a penny in it" to solve the circuit breaker from tripping. Did you try to pull evenly across both rails. I had this same problem with an old computer and just switching a few cables fixed it.

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Wed Oct 22, 2008 11:54 pm

Sc00bz wrote:This just seems like you are using "just stick a penny in it" to solve the circuit breaker from tripping. Did you try to pull evenly across both rails. I had this same problem with an old computer and just switching a few cables fixed it.
Well I would happy to do so :-) The problem is that first rail has the only one 4-pin connector which goes to MB only. I've discovered that most 2-rails not-so-new PSUs used the same design: 1 rail for MB, one rail for everything else - PCI-E, HDDs, e.t.c. That was reasonable 4 years ago when GFX cards were not using more than 50W, while overclocked Prescott processors were using up to 200W :crazy:

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Thu Oct 23, 2008 12:49 am

Ok, here it continues.

Cooler controller rules were terrible on this card - it was trying to run it at 2%, making terrible noise. Also, it was repeatedly spining up and slowing down. I had to reflash card bios with fixed cooler curve which never goes under 10%, it became much better.

Before my PSU upgrade my system was shutting down every time I run anything that uses 3d.
After upgrade things have changed: I can run for example TF2, and it works fine, but suddenly - BSOD or hang.
It was even faster in Stalker: Clear Sky - happened within several milliseconds :-)
And finally, it happens immediately when launching FurMark.

I was quite disapointed - I was almost sure that my PSU trick didn't helped...
Then I've found how to check temperatures of all components (including VRM chips in the power supply of 4870 card).
When I launched FurMark VRM#3 temperature immediately reached 127, and then ~ -80 (he he, they are using signed char here :crazy:), which means somewhere near 200C. Right after that system hang, and there was a red status LED saying that I have critical temperature. That was very funny - I've bought PCS card (stands for professional cooling system) and it cant survive even 1 second of stress testing due to overheating.

I've removed VRM chips radiator, and have replaced gum-like thermal grease with Zalman one on all chips under radiator. After that things get better, it was surviving FurMark for ~15 seconds and was hanging at temperature ~160C at VRM#3 (BTW max rated operating temperature for these chips is 150C)

I've added 2 coolers just for the VRMs and 12cm cooler blowing for the whole card, this extended stable work in FurMark to 2 minutes. Looks like it is impossible to improve here with PCS stock radiator.
Looks like here we need Cu based radiator with surface area at least twice more provided that active cooling will be still needed.

That rapid heating caused that FurMark uses absoloutely all resources of card by 100%, so that current consumption is up to 80-85A (85 for voltmodded to 1.35v), while games and "normal" benchmarks make it use just ~50A. As VRM chips heating, their resistance increases, they produce even more heat, e.t.c so if there were no thermal shutdown, every single 4870 card with similar cooling system would get burnt when launching FurMark :-)

Anyway, if BarsWF would not cause overheating (as it would not use texture modules) I am not going to replace radiator.

BTW, AMD aware of this problem, and to avoid massiva RMA's, they've decided to detect if FurMark is launched, and if it was, use just half of SPs. If you rename FurMark executable, it will not aware what is it, and will run at full speed. Nice solution : overheat? No problems, let's just half the speed :crazy:

If you want to check your card, make sure to rename FurMark executable and run WITHOUT MSAA, as it acually reduces load of all components.

Also, another interesting thing about power consumption: at 1.35v(default is 1.26v) I was able to increase core from 820 to 850 only.
But what if we would reduce voltage? 1.1v is minimal for 400/950 (i.e for stock memory controller speed), and you can go at 400/475 AT JUST 0.85v :-) That is amazing.
At 0.85v card consumes 0.5A at idle and just 10A at full load while working just twice slower (i.e. it consumed like 10-12W at full load in comparison to ~170W).
TWICE SLOWER but power consumption is about 15 times less, or 7.5 more performance per watt :-)
I would want that in my notebook (with an ability to run full clocks when working on external power :crazy:)

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Thu Oct 23, 2008 12:54 am

Image
"Even with 2 extra fans VRMs still overheats."

I bet they could use more powerful ones, or use 6 or 8 phases instead of 4, or bigger radiator :sad:
CPU power circuits use up to 16 phases while delivering comparable power (~120-150W)

hardfalcon
Posts: 32
Joined: Thu Oct 09, 2008 11:33 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by hardfalcon » Thu Oct 30, 2008 10:56 pm

You know perhaps the german c't magazine (AFAIK it's the most sold computer magazine in Europe, and IMHO they're more or less the only printed computer magazine with a scientific approach on the market).

They have a test of VGA coolers in their most recent edition, where they tested the following models:
- Arctic Cooling Accelero Twin Turbo
- Coolermaster Coolviva Pro SE
- Scythe Musashi
- Thermalright HR-03 GT
- Xigmatek Battle-Axe
- Zalman VF900-Cu LED

The Scythe one had by far the best cooling results, the Zalman one performed by far worst. All the other coolers got sufficient to good cooling results.

As far as your thermal grease is concerned: ATM, I'm using this incredibly expensive Zalman stuff, too (ZM-STG1). Believe me, it is TOTAL crap. I got an Intel Q6600 (2,4Ghz, FSB 1066) with the stock cooler, and after a swap of my defective mainboard (I built a completely new PC from scratch in January/February this year), I had to get some thermal grease. That Zalman stuff is just incredible bullshit, I cannot use more than 2 cores of my CPU, otherwise the core temperature jumps to >85°C within less than 20 minutes. Some weeks ago, I cracked an MD5 hash using your tool. I had to pause it every 10 seconds of bruteforcing for at least 1-2 minutes to prevent my computer from shutting down due to overheating. I read somewhere on the internet that the ZM-STG1 has only got a thermal conductance of 4.0 W/m°K, whereas a good grease like for example Arctic Silver Silver 5 features >9.0 W/m°K (and is cheaper in some shops).

For my part, I've ordered know some Silver 5 grease, a Scythe Musashi for my GeForce 8800GT (becomes too hot with the stock cooler in my opinion) and a Xigmatek S1283 for my CPU. If there are people interested in cooling results, I can post the cooling results as soon as the stuff has arrived (which will be soonest in a week as I ordered this at a webshop in Germany).

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Fri Oct 31, 2008 8:50 pm

hardfalcon wrote:You know perhaps the german c't magazine (AFAIK it's the most sold computer magazine in Europe, and IMHO they're more or less the only printed computer magazine with a scientific approach on the market).

They have a test of VGA coolers in their most recent edition, where they tested the following models:
- Arctic Cooling Accelero Twin Turbo
- Coolermaster Coolviva Pro SE
- Scythe Musashi
- Thermalright HR-03 GT
- Xigmatek Battle-Axe
- Zalman VF900-Cu LED

The Scythe one had by far the best cooling results, the Zalman one performed by far worst. All the other coolers got sufficient to good cooling results.
Here are results of Russian testing lab overclockers.ru :-)
Image

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Fri Oct 31, 2008 8:57 pm

BTW I finally resolved problems with overheating of my 4870.
I had to remove vrm radiator, peel it, at put good piece of thermal grease on all 6 chips under radiator. Looked like they had a little different height, so there was a bad contact between chips and radator.
After that in hardest possible mode with overclocking and extra voltage (1.26->1.4) temperature was only 130C(without extra coolers), in normal mode it became 95-100C(without extra coolers) instead of 127C(with 2 extra coolers) after previous improvements, Shame on Powercolor :-)

When overclocked & stressed by FurMark, it shows insane current consumption :crazy:
Image
Somewhere on the forums people were guessing that 4870 shuts down due to current limit 80A, here it survives 107A in a long term :-)

User avatar
the_drag0n
Posts: 217
Joined: Thu Oct 02, 2008 6:48 am
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by the_drag0n » Sat Nov 01, 2008 8:52 am

wow i think i would have simply send back the card if it had not work ;)
i mean why are they selling such crappy cards ?!

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Sat Nov 01, 2008 9:43 am

the_drag0n wrote:wow i think i would have simply send back the card if it had not work ;)
i mean why are they selling such crappy cards ?!
Yeah,I would probably send it back too, but it was boughts in USA, and sending it back would be a little problematic :-)
I guess they've hired cheap engineers :-)

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Sat Nov 01, 2008 9:44 am

Back to the topic:

One guy sent me his draft MD5 cracker for AMD cards. It was writted in CTM, not Brook, and it was very specialized (my version would be more flexible, and will show a bit lower speed)
Results are:
AMD 4870:
800 Mhz: 930 MHash/sec
863 Mhz: 1005 MHash/sec

User avatar
the_drag0n
Posts: 217
Joined: Thu Oct 02, 2008 6:48 am
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by the_drag0n » Sat Nov 01, 2008 10:00 am

wow now thats impressiv!

mrb
Posts: 51
Joined: Sat Oct 18, 2008 6:04 am
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by mrb » Sun Nov 02, 2008 12:06 am

BarsMonster wrote: One guy sent me his draft MD5 cracker for AMD cards. It was writted in CTM, not Brook, and it was very specialized (my version would be more flexible, and will show a bit lower speed) Results are:
AMD 4870:
800 Mhz: 930 MHash/sec
863 Mhz: 1005 MHash/sec
There is no such thing as CTM language, perhaps you mean written in CAL IL (compute abstraction layer - intermediate language: the kernels look like "il_ps_2_0 ... end") ?

Also, it seems to me an HD4870 could theoretically perform much better. It can execute 600e9 32-bit instruction/sec @750MHz (stock frequency), 640e9 inst/sec @800MHz, and 690e9 inst/sec @863MHz. This is significantly higher than the GTX280 at stock clock frequency (311e9 inst/sec). I estimate an HD4870 is capable of 1400-1500 Mhash/sec...

x4d3
Posts: 29
Joined: Tue Oct 14, 2008 4:10 am
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by x4d3 » Sun Nov 02, 2008 1:08 am


mrb
Posts: 51
Joined: Sat Oct 18, 2008 6:04 am
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by mrb » Sun Nov 02, 2008 1:25 am

I know what CTM is. But "written in CTM" is vague: it could refer to either CAL IL or R600/R700 ISA.

x4d3
Posts: 29
Joined: Tue Oct 14, 2008 4:10 am
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by x4d3 » Sun Nov 02, 2008 2:07 am

Did you read the link I posted at all?

"written in CTM" means exact that.

CAL is a natural evolution to CTM. If you know what CTM is you know that CTM was the name of the first version of the API, which was tied too close to hardware, but not close enough.

CTM has evolved into two pieces: HAL and CAL.

– HAL : Hardware Abstraction Layer
• Device specific, driver like interface

– CAL : Compute Abstraction Layer
• Core API device independent
• Optimized multi-core implementation as well as optimized GPU implementations
• Heterogeneous computing

mrb
Posts: 51
Joined: Sat Oct 18, 2008 6:04 am
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by mrb » Sun Nov 02, 2008 2:54 am

- BarsMonster kind of suggests it is CAL IL.
- The wikipedia page implies it is Rx00 ISA ("CTM gave developers direct access to the native instruction set").
- Your reply mentions generalities about CTM without answering my question

I had to research myself to come to the conclusion that it probably means the MD5 cracker was written using a language/API completly different from CAL IL and Rx00 ISA (which I both know). You could have cleared up the confusion in 6 words: "neither, it is a different thing" :wink:

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Re: AMD port microblog :-)

Post by BarsMonster » Sun Nov 02, 2008 6:06 am

Dudes, no more fighting :-)
It was writted in IL, no matter how it is called :-)

Post Reply
[phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1266: count(): Parameter must be an array or an object that implements Countable

Who is online

Users browsing this forum: No registered users and 1 guest