I7-920 vs 2x Phenom 955 vs 2x Q9400

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,364
Credit: 3,562,358,667
RAC: 109
Topic 194660

I'm planning to unload a few older PC's early next year* and anticipate getting enough funds to finance either a second 920 box or a pair of quad core systems. I'm wondering how they'd stack up in terms of credit earned? Figure modest overclocks for all options, probably stock cooling although the 1366 chip might get an upgrade.

As a side note, I realize an intel lga1156 would be somewhat cheaper than a 1366 box, but troubleshooting is much easier to do when you've got a second machine to swap parts with.

* if eBayers are willing to pay as much for a parted 939 system as for an entry level new quadcore setup why shouldn't I take their money.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,522
Credit: 690,809,773
RAC: 271,065

I7-920 vs 2x Phenom 955 vs 2x Q9400

I guess for optimizing credit/day/$, the strategy must include GPU consideration these days. Two (relatively) inexpensive systems each with one GPU card might be a better choice than a single one that might take two video cards but at a reduced bus speed. Or you might fit one quad with an NVIDIA card and the other with an ATI one to have a wider choice of projects.

As for E@H, both current searches, S5R6 and ABP1, really favor intel CPUs over AMD. Browsing thru the "Top Computers" list here at E@H is kind of depressing for every AMD fan.

CU
Bikeman

ML1
ML1
Joined: 20 Feb 05
Posts: 347
Credit: 86,314,215
RAC: 213

RE: ... As for E@H, both

Message 95848 in response to message 95847

Quote:
... As for E@H, both current searches, S5R6 and ABP1, really favor intel CPUs over AMD. Browsing thru the "Top Computers" list here at E@H is kind of depressing for every AMD fan.


Mmmm... Does this indeed suggest an unfair, artificial and wasteful 'certain preferential bias'...?

Can non-Intel CPU users sue Intel for deliberately wasted time and wasted electricity?

Happy fast crunchin',
Martin

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

transient
transient
Joined: 3 Jun 05
Posts: 62
Credit: 115,835,369
RAC: 0

This bias you're referring to

This bias you're referring to here, I thought that only went for the intel-compiler. If I remember correctly, GCC is used by Einstein.

ML1
ML1
Joined: 20 Feb 05
Posts: 347
Credit: 86,314,215
RAC: 213

RE: This bias you're

Message 95850 in response to message 95849

Quote:
This bias you're referring to here, I thought that only went for the intel-compiler.


Yes, indeed so.

Quote:
If I remember correctly, GCC is used by Einstein.


I didn't know that, thanks.

For GCC, I am very sure that anything like the "Naughty Intel" trick would never be coded in the first place let alone get past the open source peer review.

... So... We just need some kind soul to compile a 64-bit SSE2 version for the e@h "enthusiasts"... ;-)

Happy fast crunchin',
Martin

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,364
Credit: 3,562,358,667
RAC: 109

RE: I guess for optimizing

Message 95851 in response to message 95847

Quote:
I guess for optimizing credit/day/$, the strategy must include GPU consideration these days.

True, and if i was credit whoring I wouldn't've asked this question since an ATI 5xxx card would be a no brainer. OTOH as mum as E@H is with ongoing app development I've no idea if a GPU app that doesn't suck is 10 days away or 10 months away.

Donald A. Tevault
Donald A. Tevault
Joined: 17 Feb 06
Posts: 439
Credit: 73,516,529
RAC: 0

RE: RE: This bias you're

Message 95852 in response to message 95850

Quote:
Quote:
This bias you're referring to here, I thought that only went for the intel-compiler.

Yes, indeed so.

Quote:
If I remember correctly, GCC is used by Einstein.

I didn't know that, thanks.

For GCC, I am very sure that anything like the "Naughty Intel" trick would never be coded in the first place let alone get past the open source peer review.

... So... We just need some kind soul to compile a 64-bit SSE2 version for the e@h "enthusiasts"... ;-)

Happy fast crunchin',
Martin

Actually, the developers released a beta of 64-bit Einstein a while back. But, for some reason, it ran dog-slow, so they had to ask users to quit running it. I don't know if any progress has been made toward 64-bitness since then.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,849
Credit: 110,014,611,063
RAC: 23,318,230

RE: I'm planning to unload

Quote:
I'm planning to unload a few older PC's early next year* and anticipate getting enough funds to finance either a second 920 box or a pair of quad core systems. I'm wondering how they'd stack up in terms of credit earned?


I've done quite a few upgrades recently so you might be interested in some of my experiences. I had a big fleet of tualatin PIIIs - celeron 1300 overclocked to around 1500 - 1600. I acquired them very cheaply (around $10 each) and they were quite frugal on power (175W PSUs). They could outperform a low end P4.

The cases were nice and compact and stacked well to form a 12 high compute column. They had good ventilation with the covers off. The PSUs were the SFX form factor (3 screws instead of 4) that fit the flex-ATX style case.

By chance I found an ebay seller who was flogging a series of PSUs that were rebadged Seasonic 300SFDs that were rated at 300W total but could deliver 270W on the 12V rail. Perfect for a modern quad core budget crunching beast. I picked up 45 of these (new but surplus stock) at $11 each. I've converted pricing to $USD.

I have played around with 3 particular CPUs Q8400, Pentium dual core E6300 and Celeron dual core E3200. My aim was easy overclocking and best bang for the buck. I wanted to use stock air cooling and the cheapest possible motherboard. I wanted to reuse existing cases, HDDs and CDROMs.

After securing the PSUs, my next stroke of luck was to go for an ASUS mobo, the P5KPL-AM/PS. I started out paying about $55 each but I ended up getting the bulk of them for about $45. They had a wide FSB range and some primitive voltage options which turned out to be very useful. Also by using DDRII-800 RAM and by setting it to 667 in the BIOS, I could overclock the FSB without significantly overclocking the RAM. I ended up buying a big stock of 2GB sticks at around $26 each before the price rise so I was very lucky there as well.

I guess my favourite overclock was for the E6300 which at stock is 10.5x266.6 = 2.8GHz. I can run these all day at 10.5x344 = 3.612GHz and the memory is then at DDRII-860 (430MHz, 5:4 ratio with FSB), which is a small overclock only and gives no problems with a number of different value RAM brands that I've tried. I didn't even try going higher because 3.6GHz seemed like an overambitious target when I first started and I didn't want to push the RAM too far. I've got around 15 E6300s and every single one does 3.6 out of the box. For a chip that cost me around $90 (recently less than $80) they are great, delivering a stable RAC of close to 3K.

For the Q8400, stock is 10x266.6 = 2.66GHz. I can run these around 325 - 340MHz which gives a speed between 3.25 - 3.4GHz. These turned out to be a bit more variable in overclock ability, probably because of the inadequacy of the stock cooler in handling the heat output from 4 cores @ 100%. Once again, RAM is slightly overclocked using the 5:4 ratio. These cost me around $185 and deliver a RAC around 5K - 5.4K.

For the E3200, stock is 12x200 = 2.4GHz. I'm playing with one at the moment and it seems fine at 12x265 = 3.18GHz. It's only got half the L2 cache of the E6300 but it is doing tasks not that much slower. I reckon the RAC will get to close to 2.5K which is not too shabby for a $48 chip.

The total cost of upgrading a tualatin PII box to a 2.5K RAC machine was therefore PSU+Mobo+CPU+RAM = 11+45+48+26 = $130. All figures have been converted from local currency to $USD. For a E6300 there is an extra $40 approx and for a Q8400 it's an extra $140. So the E3200 delivers around half the RAC of a Q8400 for around half the upgrade price. All of the CPU only machines are running Linux - in fact exactly the same Linux as before the upgrade. I discovered that I could shut down a crunching machine, do the hardware upgrade and boot it straight back up again and simply resume crunching. The GUI worked without complaint so I guess the graphics driver in Linux must be able to handle quite a range of intel chipsets from i810 to G31. I was fully expecting to have to reinstall Linux and actually did on a few until I accidently discovered it would work 'as is'.

I've kitted out some of my upgraded machines with an ATI HD4850 graphics card. I bought 12 - most around $110 or so. I'm still using the same 300W PSUs as described above but I've also reused the previous 175W units (that I'd been taking out) as a power source for the graphics card. They can deliver 100W @ 12V and there is enough room in the case to mount both PSUs. I can trick the second PSU to fire up without its 20pin ATX connector being plugged in by shorting out one particular wire in the connector. The second PSU powers the graphics card and the CDROM. I thought it might be wise to have a small load on the 5V and 3.3V rails as well. The second PSU is powered from the same mains supply and the whole thing powers up nicely when the mains is activated and the ON switch is depressed.

The 12 machines that have the 4850 cards are both dual and quad cores and only E@H is crunched on the CPUs. There is probably a 5% or so loss in CPU production only. The graphics cards are crunching Milkyway and Collatz, with the latter simply a backup for times when MW is down. The GPU crunching at MW is amazingly fast. A task that might take 5 hours on a mid range P4 can be done in 205 seconds on an ATI 4850 GPU. That's about 3 secs CPU and 202 secs GPU. Each card is capable of delivering a RAC of around 80K. For anyone who likes credits, the numbers are quite insane :-).

PS: I started writing this more than 24 hours ago but got sidetracked on various other things. If anyone wants any further embellishment about my budget crunching and upgrading experiences, just ask away. I first got interested in MW during the major E@H outage back in February, I think it was. I was trying to find a physics/astronomy related project to be a E@H backup. I used MW for a while in this capacity but when the ATI aware BOINCs were developed, I shut down all MW CPU crunching and set up GPUs only for that task. In just a couple of months, I've clocked up close to 100M in credits there. I sure hope this 3D mapping of the MW galaxy is going to lead to something :-). And I'm still stuck without a reliable backup for E@H :-).

Cheers,
Gary.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,522
Credit: 690,809,773
RAC: 271,065

Hi! Catching up on several

Hi!

Catching up on several sub-threads :

The somewhat not-so-brilliant performance of most AMD CPUs (especially the consumer grade ones) compared to the rival Intel CPUs is NOT caused by some artificial slowing down, e.g. by the compiler. It's just that AMD has fallen behind Intel lately, and the fact that the E@H apps benefit from a lot of cache, where Intel tends to have more than AMD CPUs.

64 bit: I contributed some patches to the S5R5/6 source code that allows to compile for Intel/AMD 64 bit (I only tested under Linux, never tried Windows 64 bit, I don't even have one). But the speedup by 64 bit compilation alone for S5R5/6 is (depending on the machine) minimal to non-existing. We are talking single-digit % here. I'm not surprised the code wasn't put into production yet because it's just not worth the effort. The only thing that would justify it in my view is an increasing number of Linux distros that do not install the 32 bit compatibility apps by default.

Gary: A stunning report from "the farm" !!! Maybe it's time to retire my two Dual Tualatins as well, but OTOH I got emotionally attached to those old guys. Kind of the last of their breed and the bridge to the next big thing.

CU
Bikeman

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,364
Credit: 3,562,358,667
RAC: 109

On the 64 bit app front

On the 64 bit app front there're rumors that win8 is going to be 64bit only, although noone promulgating them has given a good reason why and I'm somewhat skeptical. "The cool kid (server 08) did it" isn't IMO a good reason, and the fact that Intel will be selling 32 but only chips (Current mobile atoms) until sometime next year argues against it.

If anyone's interested in buying cheap and noisy boxes someone's selling dual socket singlecore A64 blades on ebay for $13 a pop. You'll need to add a HD and DDR1 ECC ram to use them. They're reportedly loud enough to cause problems talking over the fan noise (40mm PSU fans).

http://cgi.ebay.com/Rioworks-HDAMA-DUAL-2Ghz-AMD-opteron-motherboard-CPU_W0QQitemZ230398060177QQcmdZViewItemQQptZLH_DefaultDomain_0?hash=item35a4cb6691

I suspect once you add power costs in most people'd probably be better of building a single i7 box than getting a half dozen of them and stacking them in the garage.

Ver Greeneyes
Ver Greeneyes
Joined: 26 Mar 09
Posts: 140
Credit: 9,562,235
RAC: 0

RE: On the 64 bit app front

Message 95856 in response to message 95855

Quote:
On the 64 bit app front there're rumors that win8 is going to be 64bit only, although noone promulgating them has given a good reason why and I'm somewhat skeptical. "The cool kid (server 08) did it" isn't IMO a good reason, and the fact that Intel will be selling 32 but only chips (Current mobile atoms) until sometime next year argues against it.


I do know they're writing a 128-bit kernel, which you could call forward-thinking but considering we're still in the transition from 32 to 64 bits, I think it's a bit silly (hell, modern processors can't even address all 64 bits of memory yet).

In my opinion they -should- make it 64-bit only, and not worry about building any kind of backward compatibility into it, except they should have a good virtual machine shipping with it that can emulate all past versions of Windows. I don't know if those rumors have any credibility though.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.