Dual opteron desktop benchmarks? $(O244*2) <$(FX-53)

[josmala]

There is plenty of benchmarks running ALMOST similar like single processor systems, running just a little better clock speed or more cache but NEVER on dual opterons ...
Lets make a comparison. Pair of 1.8Ghz Opteron costs as less than the high end single CPU processors like FX-53. That means that even with increased MB costs and RAM costs you only need 50% scaling from doubling the number of processors to beat FX, Since with 50% scaling it would beat the 2.7Ghz AMD CPU, some scales better some worse, but still it would be interesting set up. Plus you get more memory bandwith, and stabler system. With 30% scaling it would do as well or better than FX-53. Probably better if the benchmark won't scale linearly with clockspeed...

K8T Master2-FAR looks like comparatively priced MB, and looks like it gives us 4 channels instead of 2 like some workstation Opteron MB:s. The difference in dual cpu MB cost compared uniprocessor one is made up by the processor price difference. There is price vice NO reason to include FX-53 and not include dual opteron with 4 channels of ram... Especially on linux. And price comparison to P4EE is similar so...

 

[hytek369]

give me some software and i can benchmark my 244s on K8T

 

[DAPUNISHER]

I'm not sure what you are getting at josmala The dual Opteron 1.8ghz has a different market focus than FX&amp;EE. Most do not buy FX or EE for server and workstation use, they buy them for high-end gaming/overclocking desktop use.

There are very few games that MP will help, so most gaming benchies, particularly the very popular FPS variety, will still show the FX&amp;EE strengths vs dual Opterons.

I do feel FX&amp;EE are poor values vs less expensive A64/P4 for the same skt's. but I don't think comparing a dual Opteron to them is very relevant given the difference in software&amp;tasks the Opteron MP is intended to run.

I presume you have seen benchmarks that compare these platforms in 'nix and various tasks so could you link the review/s that inspired your post? TIA

 

[josmala]

Firstly there is NO benchmarks showing whether your assumption of lack of support for multiprocessing, at desktop is true. For instance powerpoint spends plenty of time in other than primary thread, in 90s paper.
Modern game architectures will have coarsegrain parallerism, like running network in on thread, graphics in other 3rd running AI and 4th physics, not that everybody supports that. Secondly linear scaling according to clock speed is not true for desktop use either, so typicly assume getting 50% of your clock speed difference these days. So 1.8 vs 2.4Ghz while it is 30% difference in clock speed could get only 15% difference in real world performance and that much you could save is OS and GFX driver utilizing in 2nd processor so that primary processor for game don't have to deal with context switches, and spend time on other processes.
The X grafix system runs in separate process compared to the application so does window manager compared to X then there is kernel time. I see that under linux there is plenty of parallerism available, my question is that how that shows in realworld. Comparison between lower clocked dual opteron and single FX-53 , is needed to determine which brings better value. From mathematical point of view I see dual opteron as a GREAT desktop system compared to single 2.4 Ghz chip. Remember there is only 2 difference between FX-53 and dual opteron firs ECC memory support, and clock speed. Dual opteron is other vice no different compared to FX-53. The opteron is for AMD a road for bigger margins, and new markets. Its in no way castrated in desktop application except price, on other way to consider desktop system is nothing more than castrated opterons, that are sold at different price points. But two low end server chips is still cheaper than one highend dekstop chip, and thats where I'd rather see comparison.
Compiling is multithreaded, so is rendering, actually last linux DESKTOP article has more multithreaded application benchmarks than non multithreaded benchmarks. On windows there are applications that are multithreaded and applications that are not. But for SMALL as % difference that clock speed that comes from getting top of line and bottom of line desktop processor there is little reason for going high end desktop. But at top range dual opteron should be far better. Opteron is intended to different market, and the difference is mostly in price and added features, but as lower end opterons are much cheaper than high end desktop chip then getting two of them tilts the performance difference to opterons advantage.
What I'd rather see is Anandtech article comparing highend desktop processors vs equally costing opteron solution, in a typical anandtech benchmark range.

 

[Acanthus]

So in all that big word mumbo jumbo you pretty much blew around a lot of hot air.

There are very few game engines and applications that are SMP-aware, period.

This number will increase in the future when dual core products are out and all applications can see gains from parrallel processing. But as it stands now, multitasking is the main purpose of dual cpus for the desktop.

 

[josmala]

Lets make a exact comment. The lack of SMP comparisons in desktop is barely tolarable, but when people have same ignorance while writing a LINUX desktop system stuff its something.
A) ANY tasks that have been graphics intensive takes over 30% of cpu cycles in X server that alone should give you multitasking in SINGLE application since about half of cycles is spend in other PROGRAM, that handles the graphics and IO called X server. More or less then there is system kernel threads and user level tasks. Its ALL there. I don't have methods to benchmark but when CPU is 100% used and the process that takes MOST of the CPU take under 60% of CPU time and even that it spends half of it in system time so I don't see WHY linux desktop system wouldn't run faster on SMP. Think newrly ANYTHING that deals lots with graphics benefits from SMP under linux, since it runs inside 4 different processes, that all take major time.
B) Compiling is SMP aware tasks, everything that was previously linux.
C) I personally cannot afford SMP box right now. Thats why I cannot give exact benchmarks. And I don't have windows either
D) I know that 78% of Powerpoint is in primary thread which means, that its faster in dual opteron than single non-opteron. [Even if the performance doesn't matter thesedays in this task.]
E) Almost ALL 3D modeling programs are multithreaded, which makes those benchmarks fly too...
The powerpoint was mentioned because it was surprice to me too...
F) While there are some games that are multithreaded, it would be nice to know how many really are benefitting from SMP.
G) Anything that looks better with hyperthreading enabled gives MUCH better improvement from SMP than what you get from hyperthreading. [Other way isn't true since hyperthreading does cache trashing and still it uses shared reasources instead of having twice the execution resources.
What I was saying that there is kind of a multitasking that person sees ONE application running but that aplication uses, A) GFX driver B) OS services that can be run in parallel to real application. That benefit ALONE is enough in linux for opteron to beat FX-53 at same price.
H) There is plenty of applications that are SMP aware, but most people think some that are multhreaded is not, and most people run multiple heave duty processes without even knowledge about running so many tasks in background crabbing the CPU, and those tasks could help over the issue of being 25% slower clocked than the high end desktop processor.
I) I use my linux computer other tasks than gaming mostly, and those other tasks should benefit from SMP...
Like webbrowsing, compiling, running simulations, reading pdf:s etc...

 

[Mik3y]

cliff notes?

 

[Acanthus]

Cliff notes: SMP benches faster and does obscure, non-typical-user tasks faster than a single FX-53.

We know that.

The typical user still wont benefit as much from SMP, especially in games.

 

[Markfw]

I would be glad to help if possible. I have an Athlon64 3000+ at 2.1 ghz, and dual Opteron 248's at 2.2. I have Windows 2000 on both, but Win64 could be put on both. I even have Gentoo, and thought about installig it, but haven;t had the chance yet. I think that if you multitask that duals work better than HT, and personally would find it hard to go back on my primary computer to a single CPU, but they aren;t cheap. (I'm into my setup over 6k including the SCSI stuff..., 3.3K with no drives at all !)

 

[nitromullet]

http://techreport.com/reviews/...teron-x48/index.x?pg=1

...may be of interest to you. From these results (under Windows) it doesn't really seem like a slower clocked dual Opteron system would be a viable alternative for most desktop users. This has customarily been the case with dual Xeons vs a single P4 as well.

http://www17.tomshardware.com/motherboard/20030827/

Edit:

E) Almost ALL 3D modeling programs are multithreaded, which makes those benchmarks fly too...
The powerpoint was mentioned because it was surprice to me too...

But... Will those apps "fly" if the costs for the dual rig and the single are equal? Basically, will two considerably slower CPU's be faster overall that one fast one in desktop applications? ...In most cases, no...

 

[Zebo]

?

All I get out that post is the dually 2xx opterons cost way more than 1xx and to get a serious dually mobo you won't be buying a cheap POS like MSI but a $400 Tyan.. leading me to believe it's a wash.

 

[nitromullet]

Originally posted by: Zebo
?

All I get out that post is the dually 2xx opterons cost way more than 1xx and to get a serious dually mobo you won't be buying a cheap POS like MSI but a $400 Tyan.. leading me to believe it's a wash.

I'd say that pretty much sums it up.

 

[Markfw]

Originally posted by: nitromullet

Originally posted by: Zebo
?

All I get out that post is the dually 2xx opterons cost way more than 1xx and to get a serious dually mobo you won't be buying a cheap POS like MSI but a $400 Tyan.. leading me to believe it's a wash.

I'd say that pretty much sums it up.

I own it, and I can;t argue. It's very good at many things, and I paid the price for it. Would I do it again ? Since I can afford it, probably, but my advice is, if you aren;t loaded, you will be disappointed. It is DEFINITELY NOT bang/buck, but it is nice.

 

[bunnyfubbles]

yeah, the links showed that a dual Opteron 248 is right up there with the fastest but it isn't THE fastest handsdown than any system with 1 CPU. Bang for buck still goes to midrange A64s and P4s for top performance without forking over an arm and a leg, but if you would actually use a dual Opteron system for professional work and can't afford more for an additional gaming system you could very well more than get away with a dual system for work and play.

 

[laxxy]

i think it all depends on the application. i use some statistical software and one of the routines there works about 3.5 times faster on a 3-yr old Palomino Athlon XP 1800+ than it does on a new P4 3.0C. so try to find a benchmark which you think is closest to what you're going to be doing. and even then it'll probably be off...
Dual opterons do look pretty good in some comparisons linked above...
btw -- what are the major issues with sub-$200 dual boards? arent there any decent ones?

 

[DAPUNISHER]

Originally posted by: Zebo
?

All I get out that post is the dually 2xx opterons cost way more than 1xx and to get a serious dually mobo you won't be buying a cheap POS like MSI but a $400 Tyan.. leading me to believe it's a wash.

Yep :beer: and let's not forget about the more expensive ram 939FX doesn't require.

 

[Zebo]

Originally posted by: DAPUNISHER

Originally posted by: Zebo
?

All I get out that post is the dually 2xx opterons cost way more than 1xx and to get a serious dually mobo you won't be buying a cheap POS like MSI but a $400 Tyan.. leading me to believe it's a wash.

Yep :beer: and let's not forget about the more expensive ram 939FX doesn't require.

well he already mentioned that so I felt no need to repeat it.

But i did forget to mention with ECC/REG ram.. not only is it more expensive, you won't find any running tight 2-2-2 timmings and by the same tolken you won't find any 250Mhz FSB capable...the later matters less since you don't sound like you're not going to overclock. But still, already taking a 5% performance hit using ECC/REG combined with another hit buy using 2-3-3 (if your really lucky with OCZ EL ECC)instead of 2-2-2 as you would with the FX....

again it's a wash price/performance wise.

 

[josmala]

The expensive ram is expensive for a reason, unfortunately most people won't ask reliability for their computers ;(
Anyway those benchmarks did show something...
A) Memory bandwith and latency is important. [The benchmarks where run in such a configuration that one of the opterons gave up its integrated memory controller.]
B) Windows world probably has far less parallerism available, but thats ok for most people On the otherhand in windows world people do have different kind of background processes like kazaa &amp; antivirus that could make the background tasks interrupting front application time to time annoying...
C) When others clockspeed is 1.75 times desktop, and you have worse memory latencies and bandwith then you loose on desktop badly, even if you have two processors. By lower clocked I didn't mean go for lowest end.

And for 200$ price you get single memory controller dual channel dual opteron board while with bigger money you get to use both CPU:s memory controllers and somewhat more brings you to integrated GFX and SCSI controllers which is not for desktop use anymore... Some one made a typo at pricewatch and there was a 4 channel opteron board available for 200$ price range, so with different boards the difference is not too good for opteron, but after clicking it was different board on the other side ;( And I wouldn't have this talk if the typo was not made, as dual channel on dual opteron is a not too good because of incresed memory latency.
Anyway always keep notice of price jumps in processor prices still, and RELATIVE difference in clockspeed... Jumping from 2Ghz to 2.6ghz is different from jumping 1.6 to 2.2 ghz, for single processing. So I'm interested to see what 0.9 micron technology will do for AMD clockspeed / price distribution. The relative clockspeed difference between high end and midrange is dropping all the time, with both increasing same amount in absolute terms...

 

[jose]

josmala,

Can you put some blank lines in your post, they're hard to read...

Regards,
Jose