AMD X2 better than Intel's C2D for floating point calc's

[fatchicken]

I am looking at getting a new computer. I was thinking of getting a Core 2 Duo (probably E6400). I do some software development, mainly engineering software which involves heavy duty number crunching.

It seemed to me that the X2 5000 and the E6400 were neck and neck on Tom's CPU charts. However, I noticed tonight that the X2 5000 handily beats the E6400 for the SiSoftware Sandra 2007 Memory Floating benchmark (8355 to 5745) and for the SiSoftware Sandra 2007 Arithmetic MFLOPS (15998 to 13688). I am referencing those two as it seems that they would represent my interests more than the gaming benchmarks.

Is it commonly known that the Athlon X2 chips are noticably better than the Core 2 Duo chips at number crunching? Or am I missing something here?

 

[kmjohnso]

The problem with buying computers for scientific computing is that no one software/program is representative of any group of processing tasks. I can tell you with a high level certainty that my 64-bit 4D image reconstruction algorithms run much faster on K8 than C2D (clock for clock). This doesn't really mean anything though, since your software/computations do something completely different.

So you have two options: (1) Test your program/programs on two similarly configured machines and/or (2) base your purchase on its performance across a wide array of benches which clearly show the C2D winning.

Unless you plan on doing one very specific thing and have time to test that thing, its better to go with the overall.

 

[endyen]

Is it commonly known that the Athlon X2 chips are noticably better than the Core 2 Duo chips at number crunching? Or am I missing something here?

Where the program used is set up for SSE, the core2 duo chips do very well.
In most cases, where the program has not been so optimized, the AMD's may still hold a slight advantage.
Having said that, it's not like we are talking about the difference between an A64 and a P4. The C2D have a much better IPC than the P4s did.

 

[Pippero]

Guys, he's talkin about engineering software.
Not everybody buys a computer just for gaming...
And yes, if some SW uses x87 FP instructions rather than SSE, the Athlon X2 comes up faster (Sciencemark, for example).
Intel probably (and rightfully) felt that x87 code is dying anyway.
This also happens in bandwidth measuring synthetic tests (not because these tests are "rigged").

 

[gOJDO]

What you have seen has nothing with FP performance. It is a synthetic benchmark for memory bandwidth and memory bandwidth does not translates into FP performance. What you've seen is the sAM2 X2 with DDR2 which has double the theoretical bandwidth of the s939 X2, but it is not faster.
C2D has much better FP performance than X2 at same freqfency. There is a sticky thread on this subforum about the C2D. Check out the benchmarks.

 

[pat]

yes you are missing almost everything
the sis tests are not real its a guesstimate
if you take an athalon and C2d and do real tests with them the athalon gets destroyed
in fact we should be sending the company for the sandra tests emails asking why this is messed up. my guess is they used specific procs and system configs to design the tests. and now that c2d is out the tests look bad.



just curious what "heavy duty number crunching" tests do you do that is not for gaming?

thanks
dave

Yes, it is really possible to have the Athlon 64 (and not Athalon...) to perform better at certain task than core2 processor. Just like old P4 were faster than Athlon64 in some task. Intel was loosing to AMD in the gaming segment, so they put effort to improve this part. Those who believe that core2 processor are the one to have, well, it is partly true. depending of the task, you may better get an X2.

I can use core2 and X2 computer so I know what I mean. Overall feeling are the same for both. neither one seem faster, unless I do specific task on the core2. which one is the faster? ..both.

Would I trust benchmark better than real life experience? Never.

Would I trust advice from someone thant can even spell the CPU part correctly? No

Would I trust advice from someone that have no clue about applications that use heavy number crunching beside gaming? never..

 

[Pippero]

This is a bit of a stretch.
Core 2 is considerably faster (clock for clock) in, like, 99% of benchmarks...

 

[pat]

Is it commonly known that the Athlon X2 chips are noticably better than the Core 2 Duo chips at number crunching? Or am I missing something here?

It can be. depending of the algorithm used, one could perform better than the other. Not so long ago, the Athlon64 was overall faster than the P4, but in some application, the P4 was noticably faster than the Athlon. in some video application, the P4 was sometime faster. But sometime, dependingof the application or the codec used, the AMD was better.

If you could just write a small application like algorithm, that could represent some of your work, to make like a benchmark application, you could go to a store, and try it on both. this way, you will see if your needs are toward core2 or Athlon64 performance.

But don't rely only on what you see on benchmark. benchmarking machine are not really configured the same than your work machine, that run multiple task in the backgroung and can have some hardware limitation (network storage) that would just level the performance of both machine.

The fastest computer is always as fast as its slowest component.

 

[pat]

This is a bit of a stretch.
Core 2 is considerably faster (clock for clock) in, like, 99% of benchmarks...

If you take a 3.0Ghz P4 against a 3.0Ghz OCed A64, then A64 is generaly faster clock for clock, but still could have problem to keep up with the P4 for some specific task..

Just like in this case, maybe the specific task he's doing is better for the A64.. and there's nothing wrong (or right..) woth that.. it just happen.

 

[Doughbuy]

Get a Quad and call it a day =P

Also, the X2 5000+ is more comparable to the E6600 pricewise, any reason why you were comparing it to the 6400?

 

[geralt]

Can you give an example of task where 3GHz A64 can not keep up with P4 with the same clock speed? 8O

If you take a 3.0Ghz P4 against a 3.0Ghz OCed A64, then A64 is generaly faster clock for clock, but still could have problem to keep up with the P4 for some specific task..

Just like in this case, maybe the specific task he's doing is better for the A64.. and there's nothing wrong (or right..) woth that.. it just happen.

 

[clairvoyant129]

I am looking at getting a new computer. I was thinking of getting a Core 2 Duo (probably E6400). I do some software development, mainly engineering software which involves heavy duty number crunching.

It seemed to me that the X2 5000 and the E6400 were neck and neck on Tom's CPU charts. However, I noticed tonight that the X2 5000 handily beats the E6400 for the SiSoftware Sandra 2007 Memory Floating benchmark (8355 to 5745) and for the SiSoftware Sandra 2007 Arithmetic MFLOPS (15998 to 13688). I am referencing those two as it seems that they would represent my interests more than the gaming benchmarks.

Is it commonly known that the Athlon X2 chips are noticably better than the Core 2 Duo chips at number crunching? Or am I missing something here?

You shouldn't even be comparing the X2 5000+ to the E6400. First of all, the X2 5000+ pricewise is comparable to the E6600 like the above person stated. Performance wise though, it's comparable to a E6300-E6400. As we all know, the E6600 heavily outperforms the X2 5000+.

 

[fatchicken]

I am just replying to the last one posting, but thanks to everyone for some very valuable discussion. Just a few comments on some of the posts:

- I was comparing the X2 5000 to the E6400 because they seemed to line up to each other in the overall benchmarks on Tom's CPU chart. The price difference is why I why was originally leaning to the E6400.

- I like the idea of writing a small program that does some benchmarking and running on some similarly equipped computers with those two CPU's.

- It also dawned on me after the post, that a lot of the time will be spent compiling. Therefore the comments about looking out for overall performance as opposed to one task are quite appropriate.

- Comments about synthetic benchmarks as opposed to real life are noted.

Thanks again everyone.

 

[geralt]

Compiling could be the matter on actual code.
Some could run faster and some seems just to trash C2D 8O

http://www.planetamd64.com/lofiversion/index.php/t26394.html

mmoy
Sep 18 2006, 09:41 AM
I think that the main reason for the Core 2 Duo drop in performance gain is that Macro Ops Fusion isn't supported in 64-bits. But that's just a guess.

As far as performance goes, I have a Core 2 6600 desktop, Athlon 64 3200+ (754) laptop and X2 4400 desktop. The Core 2 machine is very nice but there is one area where it performs very poorly. And that's on 64-bit builds of Mozilla. I clocked it over the weekend and my old Athlon 64 3200+ laptop (over two years old) took less than half the time than the 6600 desktop took to get through the first stage of the build. The desktop had the advantage of 7200 RPM disks as well (one 160 GB SATA and one 160 GB Deskstar). The laptop was using 5400 RPM disks.

One of the main reasons why I bought the Core 2 desktop was for Mozilla builds. It looks like I will continue to use the laptop for 64-bit builds. I haven't tried 32-bit builds on the Conroe yet. The kids have the X2 desktop and prying it away from them is hard to do.

- It also dawned on me after the post, that a lot of the time will be spent compiling. Therefore the comments about looking out for overall performance as opposed to one task are quite appropriate.

.

 

[Doughbuy]

I don't trust any comments about Intel/AMD from a website that blatantly states PlanetAMD... sorry

I would still go with the E6400 since even in the benchmarks there's not a huge difference in performance while the price savings is quite a bit more (33% price saving to what... 10-20% performance increase?) Don't forget the OC potential (if your interested at all). If you really want end-all be all performance, the e6600 will do just fine...

or like I also said... quad-core anyone?

 

[geralt]

Well, that comment is so shocking that it must be true

For more balanced point of view:

http://www.ecrypt.eu.org/ebats/computers.html

In any instance I would just try to compile the code on similarly configured machines with different processors in local CS.

I don't trust any comments about Intel/AMD from a website that blatantly states PlanetAMD... sorry

 

[pat]

Can you give an example of task where 3GHz A64 can not keep up with P4 with the same clock speed? 8O



If you take a 3.0Ghz P4 against a 3.0Ghz OCed A64, then A64 is generaly faster clock for clock, but still could have problem to keep up with the P4 for some specific task..

Just like in this case, maybe the specific task he's doing is better for the A64.. and there's nothing wrong (or right..) woth that.. it just happen.

I don't remember which apps it was, but the p4 was very strong at it.. it was related to video work, iirc.

 

[habanero]

If you are a codewriter and need serious number crunching ability for iterative calculations you may want to have a look at a product called the ClearSpeed Advance™ Accelerator Board and the ClearSpeed software development kit.

see the link

http://www.clearspeed.com/

 

[enforcer22]

Can you give an example of task where 3GHz A64 can not keep up with P4 with the same clock speed? 8O



If you take a 3.0Ghz P4 against a 3.0Ghz OCed A64, then A64 is generaly faster clock for clock, but still could have problem to keep up with the P4 for some specific task..

Just like in this case, maybe the specific task he's doing is better for the A64.. and there's nothing wrong (or right..) woth that.. it just happen.

I don't remember which apps it was, but the p4 was very strong at it.. it was related to video work, iirc.

There never were benchmarks were a p-4 and athlon were at the same clock. P-4 would have been wasted way to bad if it had to work againced a athlon at a clock per clock basis. They didnt ramp the clock speeds mearly for looks they had to so thier chips could keep up.

Would I trust advice from someone thant can even spell the CPU part correctly? No

Becuase that actualy means he has no clue what hes talking about. Seriously when will this ignorance die.

Really though find out what optimisations the software you want to use incorperates. Either CPU is going to do the job and youll most likely not really see the difference. Just get what works best for your software if it really matters to you. Research doesnt hurt.

 

[Lucky_Luke]

The first step in making a good decision is choosing the right benchmark. In your specific case (software development and numbercrunching) I would recommend looking at the widely accepted SPEC CPU2000 and the just released successor SPEC CPU2006 (www.spec.org) for the following reasons:

- SPEC CPU is a real world benchmark consisting of a set of real applications from different fields (science, engineering, software development, encoding ...)

- Whoever wants to submit SPEC CPU-results to www.spec.org has to compile the source code of the benchmarks by himself before running the benchmark, with the compiler and compiler options of his own choice. So the software can be adapted in the best possible way to the CPU at hand. For example, if a CPU is good in SSE, the benchmark will be compiled to make heavy use of SSE, if not, it won't. So the conditions are equally fair for all CPU families.

- Since a number of different applications is involved, the results are a good average of different program characteristics.

For an estimate of compilation speed, look at the SPECint results for integer workloads, or more specifically at the results for the subtest that involves compilation. For an estimate of the numbercrunching abilities, look at SPECfp.

To find out about C2D vs. Athlon 64, just look up the best results published for each processor. You can look at Xeon Woodcrest or Opteron results as well for models with comparable frequency with the C2D or Athlon 64, since the cores are very similar, although the cache size might differ.

In general, judging from the SPECint2000-results, the Athlon 64 will be behind the C2D by about 30 % clock for clock for integer workloads. The results of the compilation subtest 176.gcc give a similar picture.

For numbercrunching, however, the Athlon/Opteron is somewhat faster overall and clock for clock in SPECfp2000 than the C2D/Xeon Woodcrest. In SPECfp2000, a 2,8 GHz Opteron 1220SE wins against a 2,9 GHz Core 2E by 3073 to 3050, and in SPECfp_rate2000, a 2,6 GHz Opteron 1218 wins against a 2,9 GHz Core 2E by 56,5 to 53,1.

 

[ElMoIsEviL]

Guys, he's talkin about engineering software.
Not everybody buys a computer just for gaming...
And yes, if some SW uses x87 FP instructions rather than SSE, the Athlon X2 comes up faster (Sciencemark, for example).
Intel probably (and rightfully) felt that x87 code is dying anyway.
This also happens in bandwidth measuring synthetic tests (not because these tests are "rigged").

C2D is still faster clock for clock then A64 X2 under Sciencemark. Not sure where you're getting that the A64 X2 is superior in this benchmark. Could you link us?

 

[kamel5547]

yes you are missing almost everything
the sis tests are not real its a guesstimate
if you take an athalon and C2d and do real tests with them the athalon gets destroyed
in fact we should be sending the company for the sandra tests emails asking why this is messed up. my guess is they used specific procs and system configs to design the tests. and now that c2d is out the tests look bad.



just curious what "heavy duty number crunching" tests do you do that is not for gaming?

thanks
dave

There are TONS of engineering modeling programs out there that are much more demanding than any game on the CPU. FOr example earthquake modeling, FLAC2D (ground modeling). These programs are all floating point, and run for days before producing a result (FLAC2D for example has runs that take 3 days at 100% CPU utilization and uses only 100 MB of memory). Anyhow I could come up with more examples if I thought about it, all related to civil engineering.

Anyhow I find that synthetic benchmarks are a terrible thing to use to predict real world performance...

 

[happy_fanboy]

Careful on your thread bud it looks like the psycho intel pushers found ya.
Some of them dudes will faster cut off your head than admit their C2D is a piece of crap.
It boils down to your programs code, and if it is multithreaded or not. If not save your cash and buy a single core Athlon or perhaps Opteron.
I don't claim to be a software engineer but i know 90% of software can only use one core at this time.
Intel no longer makes a reasonable single core solution, and a single core will always be a superior core.
You can buy AMD single cores in an unbelieveable 35 WATT TDP now, which is a fantastic technological leap IMO.

 

[dasickninja]

Would you shut the f*ck up already? If you don't have any kind of proof to back up your erroneous claims, I suggest you refrain from posting anymore bullshit.
This is my last warning to you.

 

[fatchicken]

Habanero and Lucky Luke: Thanks for the info and links.

Beerandcandy: Our office is in a minute field of engineering and we use about a dozen (incuding the ones I write) programs in which all they do is number crunch all day long. Multiply that by the umpteen fields in engineering and there are a lot programs out there. I'll guarantee that you've never heard of about 99.9999% of them, but that doesn't mean they don't exist.