• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

bigadv on Dual 4284?

F

FrankD400

Limp Gawd
Joined
Jan 31, 2013
Messages
154
Can a dual 4284 machine meet deadlines for bigadv work units? I was thinking about pulling the trigger on 4386s, but I got these cheap and I figured they might be a good upgrade from my 4184s. I have a bunch of GPUs so this machine will spit out a bunch of PPD anyway, but I'd really like to crunch some of the bigger CPU work units if possible.

Also, does anyone know if there's a way to pump the FSB up on a Tyan S8225 board? Are there any BIOS options I should be aware of? Node/bank/channel interleaving on/off?
 
Last edited:
If you are running Windows (given your GPU comment, I assume you are), you are crippling yourself for smp folding in general. Even with a native Linux install, I don't think you have enough horsepower in that machine to do it - 16 IL cores at 3.3 doesn't seem like enough to me. We could calculate frame times if someone has a 3.3GHz IL machine with more cores. I do not have one, unfortunately.
 
24 MC cores @ 2.2ghz has 6hrs to spare on deadline for 8101. MC outperforms Bulldozer/IL. The way it was explained to be is a 16 core cpu packs the same performance as a 12 core MC counterpart.(clock for clock)

On the intel front, from what I recall you need 16 cores at 3.8-4ghz to hit BA.

There is nothing you can do on the FSB front to kick it up.

If you are after BA, I wouldn't do it.
 
My single 6276 ES at stock can finish a BA in time but receives no bonus points. So you may be able to do bigadvs but if you don't get bonus points then there is no point.
 
Spazturtle said:
My single 6276 ES at stock can finish a BA in time but receives no bonus points. So you may be able to do bigadvs but if you don't get bonus points then there is no point.
Click to expand...

We're referring to preferred deadline, which is what you need to get bonus.

From the other thread, the single 6282 is right on the edge of getting bonus points, so it's not surprising your 6276 does not.
 
I wonder if 2x 4386 would make the deadline and give bonus points?
 
Quisarious said:
From the other thread, the single 6282 is right on the edge of getting bonus points, so it's not surprising your 6276 does not.
Click to expand...

Hmmmm.... a 6282 with all 16 cores engaged runs at 3.0GHz in turbo mode. I have a ES 62xx 16c chip at 3.0GHz and does not come close to getting BA bonus points. How is the 6282 any different?
 
402blownstroker said:
Hmmmm.... a 6282 with all 16 cores engaged runs at 3.0GHz in turbo mode. I have a ES 62xx 16c chip at 3.0GHz and does not come close to getting BA bonus points. How is the 6282 any different?
Click to expand...


No idea, in thread he said he was getting 35:13 tpf on 8101, just basing on that.
 
Quisarious said:
No idea, in thread he said he was getting 35:13 tpf on 8101, just basing on that.
Click to expand...

That sounds about what I was getting. No bonus points, just base. 34:30 with "zero" upload time is the min TPF to get bonus points.
 
These are a 10% faster clock speed IL's (3.0 -> 3.3), so a 10% lower frame time puts us at ~32 minutes and makes it easily. I don't think you will see quite all of the clock speed percentage increase as a tpf decrease, but it is close. So, you would probably make it. As I said originally, running Windows natively is your biggest issue.
 
So with native Linux and no GPUs, you make it by a couple hours. If you can't do Linux or can't re-locate GPUs, you are out of luck.
 
musky said:
These are a 10% faster clock speed IL's (3.0 -> 3.3), so a 10% lower frame time puts us at ~32 minutes and makes it easily. I don't think you will see quite all of the clock speed percentage increase as a tpf decrease, but it is close. So, you would probably make it. As I said originally, running Windows natively is your biggest issue.
Click to expand...

If you are using tears tpf bear in mind he has fast 1600MHz Cas 8 RAM, also as somebody else has said HT will slow it down a bit and also I don't think 10% faster clock speed = 10% lower tpf. Also Piledriver is 2% faster clock for clock then Buldozer.
 
Spazturtle said:
If you are using tears tpf bear in mind he has fast 1600MHz Cas 8 RAM, also as somebody else has said HT will slow it down a bit and also I don't think 10% faster clock speed = 10% lower tpf. Also Piledriver is 2% faster clock for clock then Buldozer.
Click to expand...

Yes, memory matters matters, but it is minimal (a few seconds/frame), especially if you are talking the difference between 1600 C8 and 1333 C9.

From Tobit's testing on C32 versus G34, the HT has little effect on folding performance. A single chip with 16 cores will perform almost identically clock for clock with two chips at 8 cores each.

A clock speed increase percentage will lead to frame reduction of the same percentage, with a few seconds. This has been tested and verified a bunch.

4284s are IL chips, as are 6282s. The OP mentioned an AD upgrade, but got a good deal on the IL chips.
 
I COULD move GPUs to other machines I have if I can run bigadv on this, running Linux isn't an issue either - Slackware is my normal desktop. PD doesn't get much of a boost over BD for Folding, eh? That's kind of nice to hear actually, although with a 2% IPC boost and a 3% clock boost it could still make the difference between meeting deadlines and not.

I was going 4284s regardless, I wasn't shooting for a Folding CPU monster. I just figured if I could hit good PPD I'd do it. My RAM is not especially fast, but atleast it's 800MHz CL9.

If I can't fold bigadv and churn points I'm fine with leaving Windows on and running 670/670/660/660 GPU units with SMP.
 
It looks like chances are very high providing GPUs are removed and Linux is used as the OS.

Removing GPUs and swapping drives+doing Linux install should (hopefully) be low effort
thing so I'd suggest you just try it :)

Make sure to go over G34 performance checklist from Guides section and use psmax 1
trick (http://hardforum.com/showthread.php?t=1708870) as well.

If you're using Slackware, Ubuntu installation guide will not apply but make sure to at least
install the Kraken.
 
Update:

8101: 34m00s TPF
6901: 19m00s TPF

They run maybe a few seconds faster per frame. No tweaking done, NUMA on/off didn't make a different on the 6901 WUs.. haven't tried anything else yet aside from psmax. Seems like it's running in boost a good amount of time, though I'm not sure how to check it. Just from sensors, TDP varies between 105w and 115w. If I leave it at default CPB0 max power state TDP doesn't break 95w.

CoolerMaster Hyper TX3s seem to work pretty good, CPU reported temp is 40c CPU0, 49c CPU1 (CPU0 blows into CPU1), fan speed is somewhere around 40% so they're not loud.

I'll have to put something else together when Steamroller is released :) So at least dual 4284s can just barely make the preferred deadline for 8101 WUs.
 
Also, if PD doesn't get any IPC boost for FAH and TPF is linear with clock speed, assuming both CPUs are at CPB1 100% of the time then dual 4386s would decrease my 8101 TPF to 33m00s without touching anything. 2% IPC boost would put it at 32m21s. I haven't tried Kraken to see if it would increase performance, but I'm not really expecting much else from this machine.
 
Install TheKraken ASAP! It will give you nice boost.
From my experience on Quad Barcelona system (-smp only) from 11k to 25k. :)
 
You must log in or register to reply here.
Back
Top