a64 and mobos for heavy duty floating point?-edited

Got a shopping cart of parts that you want opinions on? Get advice from members on your planned or existing system (or upgrade).

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
aqm consultant
Posts: 88
Joined: Mon Jun 20, 2005 1:11 pm
Location: California

a64 and mobos for heavy duty floating point?-edited

Post by aqm consultant » Mon Jun 20, 2005 1:29 pm

I've been monitoring SPCR forums and reviews for a while, and am thoroughly impressed with the spirit and thoughtfulness exhibited! Good stuff.

It's time to build a new system, and would appreciate guidance on CPU and mobo selection for intensive number crunching. I use existing, and write my own scientific fortran code that's heavy in array sizes and floating point arithmetic (mathematical simulations running occasionally in excess of 8 hours at 100% CPU on a 1.5 GHz Athlon T'bird). Under consideration are both Athlon 64 FX and new dual core Athlon 64s, as reviews (and my experience) shows them as strong on floating point, and reviews here seem to give them the edge on heat (if maybe not price).

Config likely to include a SATA RAID-1 array (OS and data) and a third drive for temp storage and backups. This sits on my desk, and the T'bird noise comes close to driving me nuts (although the noise is mostly the 4 HDDs).

[Edit] - A specific question or two: How much of a heat/noise penalty is associated with the 90 nm v. 130 nm cores? Will Cool and Quiet deal with this except when running max load?

Suggestions and experience would be welcome! Thanks.
Last edited by aqm consultant on Mon Jun 20, 2005 4:02 pm, edited 1 time in total.

lm
Friend of SPCR
Posts: 1251
Joined: Wed Dec 17, 2003 6:14 am
Location: Finland

Re: a64 and mobos for heavy duty floating point?

Post by lm » Mon Jun 20, 2005 1:52 pm

If it's possible to put the machine under the table, then the desk would block some of the noise.

aqm consultant
Posts: 88
Joined: Mon Jun 20, 2005 1:11 pm
Location: California

Post by aqm consultant » Mon Jun 20, 2005 2:01 pm

True, Im, and thanks. Problem is, I've got too much paper stacked under there . . . :oops: And it still won't solve my speed problem ;-)

alglove
Posts: 363
Joined: Fri Feb 06, 2004 11:21 am
Location: Houston, TX, USA

Post by alglove » Mon Jun 20, 2005 4:25 pm

If you have not seen these already, here are a couple of good articles on the Athlon 64 X2, inlcuding performance and power consumption measurements:

http://www.xbitlabs.com/articles/cpu/di ... 64-x2.html
http://www.anandtech.com/cpuchipsets/sh ... spx?i=2410

Sometimes the choice of benchmarks make me wonder if anybody actually uses a computer for programming anymore, but anyhoo. :wink: Since the X2 uses the newer 90nm process, it actually runs cooler than the older 130nm CPUs, which is good news.

The X2's dual cores, of course, have the greatest advantage when multitasking or using multithreaded applications. Even if your complied code is strictly single-threaded, the second core will allow you to use your computer without it feeling like a stick in the mud when your code is running full bore.

If you are wondering whether Native Command Queuing (NCQ) is a feature with holding out for in an SATA controller, my answer would be, "Probably not, at least not for single-user dsektop usage in a RAID-1 array." Native and Tagged Command Queuing have the greatest use in server-type and multiuser environments, especially when data is striped across different hard drives, like RAID-0, RAID-10, RAID-5. For a single-user RAID-1 setup, there is often little discernable difference, as evidenced by the graph at the bottom of this page:

http://www.storagereview.com/articles/2 ... TCQ_5.html

I am not saying that NCQ is something to avoid, just that I would not rate it as a "must have" for your situation. This should give you some extra flexibility when choosing the motherboard and hard drives.

To sum up, I would say that an Athlon 64 X2 would be somewhat warmer than its single-core baseline of the same clockspeed, but not vastly so. Perhaps 25% warmer, judging by the graphs, and certainly cooler than the 130nm CPUs. The hard drives can probably be any of the quiet ones recommended here on this site, as newer ones also have decent performance (as opposed to the 5400rpm vs. 7200rpm quandry of a few years ago).

A bit general, I know, but those are the types of things I would look for.

aqm consultant
Posts: 88
Joined: Mon Jun 20, 2005 1:11 pm
Location: California

Post by aqm consultant » Tue Jun 21, 2005 11:27 am

Thanks for these links. Very informative. The performance kick with a CPU-intensive task running in background is huge--especially considering how small the heat dissipation penalty is. Guess I should have anticipated that. I've done benchmarking on single core hyperthreading P4s, and found that I can run two instances of one of the simulation models simultaneously on the two virtual cpus in nearly the same time as a single instance, which is only capable of running at 50% (i.e., as a single thread). Interestingly, using "processor affinity" to assign each process to a different virtual cpu made no difference, although I guess it might have if I'd been trying to run other stuff at the same time.

alglove
Posts: 363
Joined: Fri Feb 06, 2004 11:21 am
Location: Houston, TX, USA

Re: a64 and mobos for heavy duty floating point?-edited

Post by alglove » Tue Jun 21, 2005 2:45 pm

aqm consultant wrote:[Edit] - A specific question or two: How much of a heat/noise penalty is associated with the 90 nm v. 130 nm cores? Will Cool and Quiet deal with this except when running max load?
Here is a graph where you can compare the idle power consumption of the Newcastle "CG" (130nm) and Venice "E3" (90nm) cores Athlon 64 3800+, as well as the Athlon 64 FX-55:

http://www.xbitlabs.com/articles/cpu/di ... 780_9.html

Mats
Posts: 3044
Joined: Fri Dec 26, 2003 6:54 am
Location: Sweden

Re: a64 and mobos for heavy duty floating point?-edited

Post by Mats » Tue Jun 21, 2005 4:17 pm

alglove wrote:Here is a graph where you can compare the idle power consumption of the Newcastle "CG" (130nm) and Venice "E3" (90nm) cores Athlon 64 3800+, as well as the Athlon 64 FX-55:

http://www.xbitlabs.com/articles/cpu/di ... 780_9.html
I'm not sure, but I think that those FX power numbers are for the 130 nm version, FX-55 have showed up with 90 nm technology.

A64E review.
A64 X2 review.

Power consumption, CPU only!

In my opinion, you have these CPU's to choose between:

3700+, 2200 MHz, 1 MB cache $310
4000+, 2400 MHz, 1 MB cache $478 Make sure you get 90 nm version!

4200+, 2200 MHz, 512+512 kB cache $558
4400+, 2200 MHz, 1 + 1 MB cache $604
4600+, 2400 MHz, 512+512 kB cache $822
4800+, 2400 MHz, 1 + 1 MB cache $999

FX-55, 2600 MHz, 1 MB cache $814 UNLOCKED Make sure you get 90 nm version!
FX-57, 2800 MHz, 1 MB cache $??? UNLOCKED The successor to FX-55, so the price will probably be a bit above $814, but still close to that.
[EDIT: I was wrong! 90 nm FX-55 seems to be the one to prefer, and it will stay in the market for a while.]

The three CPU's marked in bold text are the ones that is most intersting if you look at prices and features (Possibly the 4800+ too if you can spend $$$($?)).
I don't think you have answered to two important questions:
What's your budget?
Will you overclock?

Some useful links:
www.shopping.com - Prices from here.
www.amd.com/pricing - AMD's prices.
www.amdcompare.com - AMD's CPU search function.

If you want even more power you should go for a 2 x Dual Core Opteron together with some nice Tyan mobo.

http://www.amdcompare.com/us%2Den/opteron/

When I read your text I assumed that you wanted one of the better CPU's available so I excluded the ones with 512 kB level 2 cache: 3000+, 3200+, 3500+ (and 3800+, not worth it). If you want to use a small amount of money but are willing to overclock, you can get a fast and nice budget system. I don't know how important level 2 cache is for you though.

Post Reply