How about some large complex application test for Conroe?

[thestain]

Most of us have seen those simple benchmarks like PCMark 2005 where even a Pentium D 840 looks good compared to AMD's finest. What I am hoping to see is results on the bigger and more complex and more cpu intensive uses of Conroe.

I have heard the larger Cache has helped on the smaller and simpler side of things ranging from a couple percent to double digit and the extra decoders are supposed to help to.. with more help on the simple and small side than on the somplex since they are in the ratio of 3:1 simple to complex for a total of four, compared to AMD's X2's that have three complex decoders.

In addition to seeing results of objective, truly similar test. namely both use the same chipset, such as nforce 4 or 5, same machines, etc.. with truly comparable memory's used.. it would be good to see more... and learn more.

I have read up all I can.. and while it appears to be a mixture of things that has led to Conroe's performance gains, I am trying to separate the part where the Conroe was bound to look better from factors in its design that lead to better results than AMD in the smaller, simpler, less CPU intensive test and games and that part that has allowed Conroe to shine in all applications.

Long winded and wordy.. I will stop here to read some hoped for replies

The Stain

 

[Acanthus]

IIRC the anandtech conroe preview actually did have some real world tests.

As for the "complex" thead testing, over 80% of cpu tasks are simple. This is why intel chose 3 simple and 1 complex over AMDs 3 complex.

 

[thestain]

Guess i am trying to get a better breakdown of the new conroe... what factors contribute most to its performance... I have heard that most of the cpus task are simple, my concern was with only one decoder capable of complex.. in a highly complex portion of an application or game, wouldn't Conroe take a hit or has allowance been made for this using other factors?

I have read the reviews and previews.. just not enough of the larger test to satisy me yet..

I plan on getting PhysX once it comes out on PCI-e.. after all the talk I am one dissapointed puppy they did not release it on PCI-e to begin with.. I hate waiting, but then again it will give some time to watch what looks like a great Salvo by Intel to get the GHZ wars going again.

 

[Duvie]

The biggest problem is the morons who have gotten early chips (cough* extreme *cough) are only interested in stroking their little e-penises for world records in superpi, 3dmock, pcmock, etc.....They dont even concentrate on something someone would actually run their computer for.....

Just wait for the real reviews and if in some great surprise it shows some sort of weakness in a complex app you use, you will have some A64 and AM2 benches to compare to....

NOthing hurts in waiting for competent reviews.....plus there is limited chipset support right now and it will only get better...

 

[n19htmare]

Well, FCG on Xtreme ran the Half Life: Lost Coast and got a FPS rate of something insane like 175 FPS
1600x1200 res
I think the settigns were pretty high. i wish i had em written down.
(Cross Fire setup), don't remember.

I'd post the link but Me, FCG and another member got in into a little fit and the thread was removed...

I guess you can run the demo and see how you pair up with it.

The thing is that you would have to set up MULTIPLE machines and run same tests on all of them in the same controlled environment... none of these guys have the resources to do that..
They did all the tests that have same standards and the Conroe outperforms in all of them.

Just wait till its released for real tests.

 

[zephyrprime]

Originally posted by: Duvie
The biggest problem is the morons who have gotten early chips (cough* extreme *cough) are only interested in stroking their little e-penises for world records in superpi, 3dmock, pcmock, etc.....They dont even concentrate on something someone would actually run their computer for....

It's funny how they do that. It's as though Super PI were the most important application in the world when maybe 0.1% of the population has ever even heard of that program. I guess none of those guys would never deign to run OfficeBench :roll:

 

[Hepafleb]

Originally posted by: Duvie
The biggest problem is the morons who have gotten early chips (cough* extreme *cough) are only interested in stroking their little e-penises for world records in superpi, 3dmock, pcmock, etc.....They dont even concentrate on something someone would actually run their computer for.....

Just wait for the real reviews and if in some great surprise it shows some sort of weakness in a complex app you use, you will have some A64 and AM2 benches to compare to....

NOthing hurts in waiting for competent reviews.....plus there is limited chipset support right now and it will only get better...

Ya i read their boards several times and it seems like a lot of their posts are something like "ofmg kikikiki i shaved .0001 ms of my suparpi time pwnt"

:roll eyes: Who cares?

 

[ZOXXO]

Why the animosity towards the xtreme overclockers? Overclocking is what they enjoy doing with their computers. They go as far as using exotic cooling methods, hard modding their mobos and custom writing bioses. The benchmark apps are the tools used to compare and contrast their systems.

It is no different from some here pouring over gaming screenshots and gushing over a few fps.

To each his own.

 

[dmens]

Originally posted by: thestain
Most of us have seen those simple benchmarks like PCMark 2005 where even a Pentium D 840 looks good compared to AMD's finest. What I am hoping to see is results on the bigger and more complex and more cpu intensive uses of Conroe.

I have heard the larger Cache has helped on the smaller and simpler side of things ranging from a couple percent to double digit and the extra decoders are supposed to help to.. with more help on the simple and small side than on the somplex since they are in the ratio of 3:1 simple to complex for a total of four, compared to AMD's X2's that have three complex decoders.

In addition to seeing results of objective, truly similar test. namely both use the same chipset, such as nforce 4 or 5, same machines, etc.. with truly comparable memory's used.. it would be good to see more... and learn more.

I have read up all I can.. and while it appears to be a mixture of things that has led to Conroe's performance gains, I am trying to separate the part where the Conroe was bound to look better from factors in its design that lead to better results than AMD in the smaller, simpler, less CPU intensive test and games and that part that has allowed Conroe to shine in all applications.

Long winded and wordy.. I will stop here to read some hoped for replies

The Stain

No such thing as "more complex" codestreams, complicated programs only contain more instructions, even the simplest codes contain ops which go through the complex decoder and require flow lookups.

In regards to merom decode width in relation to general performance, it is an enabler rather than the reason for merom's performance gains. As long as the backend can accomodate the increased instruction flow, the frontend should accomodate it. The chokepoints are all in the back, not the front... except at restart, I guess.

On a tangent, to all the people who whine about the lack benchmarks with "complex apps" because of some strange hatred of superpi, maybe they can explain exactly how superpi is less complex than whatever benchmark they prefer... on a similar vein, what the hell is a complex app anyways, LOL.

 

[Acanthus]

Originally posted by: ZOXXO
Why the animosity towards the xtreme overclockers? Overclocking is what they enjoy doing with their computers. They go as far as using exotic cooling methods, hard modding their mobos and custom writing bioses. The benchmark apps are the tools used to compare and contrast their systems.

It is no different from some here pouring over gaming screenshots and gushing over a few fps.

To each his own.

Because they are the ones getting the chips, and we arent seeing anything worth looking at but suicide benchmarks in sythetic apps.

 

[dexvx]

Because they don't give a crap about real world benchmarks/games because they will probably never use them. Yea, as hard is it for some of you people to believe, most of them probably don't do any PC gaming. They look for a standard for benchmarking, and they use SuperPI and 3dMark. The people on Anandtech would probably use WorldBench and HL2, something they wouldnt give a flying fvck over on Xtremesystems.


Oh and BTW, they posted Cinebench, HL2, FEAR, FarCry, and Quake4 for Meron/Conroe.

http://www.xtremesystems.org/forums/showthread.php?t=98466

 

[Howard]

Originally posted by: dexvx
Because they don't give a crap about real world benchmarks/games because they will probably never use them. Yea, as hard is it for some of you people to believe, most of them probably don't do any PC gaming. They look for a standard for benchmarking, and they use SuperPI and 3dMark. The people on Anandtech would probably use WorldBench and HL2, something they wouldnt give a flying fvck over on Xtremesystems.


Oh and BTW, they posted Cinebench, HL2, FEAR, FarCry, and Quake4 for Meron/Conroe.

http://www.xtremesystems.org/forums/showthread.php?t=98466

As far as I'm concerned, SuperPI is a useless benchmark unless you're building a massively-parallel system to crunch pi to a ridiculous decimal place. Maybe it does accurately show the math capabilities of a CPU; I'm not doubting that. But it's not useful in any other way... it doesn't [necessarily] reflect real-world performance, and that's what we're looking for.

 

[Duvie]

Originally posted by: Howard

Originally posted by: dexvx
Because they don't give a crap about real world benchmarks/games because they will probably never use them. Yea, as hard is it for some of you people to believe, most of them probably don't do any PC gaming. They look for a standard for benchmarking, and they use SuperPI and 3dMark. The people on Anandtech would probably use WorldBench and HL2, something they wouldnt give a flying fvck over on Xtremesystems.


Oh and BTW, they posted Cinebench, HL2, FEAR, FarCry, and Quake4 for Meron/Conroe.

http://www.xtremesystems.org/forums/showthread.php?t=98466

As far as I'm concerned, SuperPI is a useless benchmark unless you're building a massively-parallel system to crunch pi to a ridiculous decimal place. Maybe it does accurately show the math capabilities of a CPU; I'm not doubting that. But it's not useful in any other way... it doesn't [necessarily] reflect real-world performance, and that's what we're looking for.

QFT..

Nevermind...Dexvx has been defending the lame (extreme) for sometime...

Cinebench is a good starter bench but until you actually do some rendering in an app like 3dsmax with radiosity you haven't seen stress and computing power...

The games help a some, but I want to see encoding, dvd-shrink, CAD rendering, Folding times, etc...some may like to see some compiling scores, office benchmarks, etc...then you get into multitasking test as well...

Ofcourse we see none of that...

It os not about 30-40 seconds of power...It is about projects that take hours...

But you keep defending them Dexvx!!! I see what side you stand...

 

[dmens]

Originally posted by: Duvie
QFT..

Nevermind...Dexvx has been defending the lame (extreme) for sometime...

Cinebench is a good starter bench but until you actually do some rendering in an app like 3dsmax with radiosity you haven't seen stress and computing power...

The games help a some, but I want to see encoding, dvd-shrink, CAD rendering, Folding times, etc...some may like to see some compiling scores, office benchmarks, etc...then you get into multitasking test as well...

Ofcourse we see none of that...

It os not about 30-40 seconds of power...It is about projects that take hours...

But you keep defending them Dexvx!!! I see what side you stand...

LOL, I'd like to hear you explain how:

1. 3dsmax is any more stressful or requires more computing power than cinebench or superpi
2. Hours of computing is more indicative than 30-40 seconds of "power"
3. superpi is parallel, since you're QFT'ing

Hell, I'd even agree with you on the need for more benchmarks, but are those listed any more important than superpi or the other XS favorites? Hell no.

 

[Duvie]

Originally posted by: dmens

Originally posted by: Duvie
QFT..

Nevermind...Dexvx has been defending the lame (extreme) for sometime...

Cinebench is a good starter bench but until you actually do some rendering in an app like 3dsmax with radiosity you haven't seen stress and computing power...

The games help a some, but I want to see encoding, dvd-shrink, CAD rendering, Folding times, etc...some may like to see some compiling scores, office benchmarks, etc...then you get into multitasking test as well...

Ofcourse we see none of that...

It os not about 30-40 seconds of power...It is about projects that take hours...

But you keep defending them Dexvx!!! I see what side you stand...

LOL, I'd like to hear you explain how:

1. 3dsmax is any more stressful or requires more computing power than cinebench or superpi
2. Hours of computing is more indicative than 30-40 seconds of "power"
3. superpi is parallel, since you're QFT'ing

Hell, I'd even agree with you on the need for more benchmarks, but are those listed any more important than superpi or the other XS favorites? Hell no.

Well it is clear you have no clue....

Cinbench is unrealistic in that it scales at a much higher rate per core and per mhz then actual testing in real world apps shows....Therefore it does not tell the whole story...4 cores gave me a huge boost in cinebench yet in most all CAD apps I tested I did not come close to scaling that well....Without radiosity it was even worse....So you see my skepticism towards benches like this....

How many times have we seen one cpu dominant superpi yet loose majority of real world benches...IE last round of netburst P4's...

How many times have we seen encredible 1m and 2m times to see 16m and 32m times not scale as surprising fast?? Not to mention most of the apps they run show us little in terms of smp ability...

Enough said...now go away!!! "that is my fun for the day"...got that from some genius!!!


superpi is not a great tool for interpreting much of anything other then a comparison tool to similar patforms....

I guess we can agree to agree that more benches would be nice....

 

[dmens]

Good job not explaining yourself, HAHAHA. Clueless indeed.

You dismiss superpi simply because it isn't the type of parallel workload you favor. The CAD and 3D workloads you desire are so much more specialized than superpi. Never mind MP, even in UP those workloads are run and considered seperately from the rest of the bunch when we do perf assessment. Scaling per mhz? What's that supposed to mean? Frequency results in linear gains in performance. Or are you talking about architectural effects on performance... care to dive into that discussion?

By "scaling" with 16m/32m runs, I assume you meant the bench starts taking the memory subsys into account (or did you LOL). Well then it's a more useful benchmark isn't it? And XS had plenty of those runs. Not good enough apparently.

Well, since you obviously have no idea the different parts of the CPU the benchmarks exercise you have absolutely no ground to dismiss individual benchmarks. The least you can do is be honest and admit you only want to see *your* favored benchmarks. But please don't be so arrogant as to call your choices "real" and others "fake".

 

[BrownTown]

the people at XS have run bretty much every benchmark i've ever heard of on Conroe/Merom, the problem is that its all in 15 different threads, so if you haven't been keeping up then you won't be able to find it, but don't blame them for not doing enough tests, if you want to blame them for something then it should be for not keeping them all in one area. But why should they?, its jsut alot of posers showing what they got, they have to obligation to keep things tidy, its up to you people to find them if you want.

If you want to know about games, then its all there, just look. The summary is pretty much that Conroe owns at gaming, but if you want the numbers go to XS and find them for yourself.

 

[dexvx]

Originally posted by: Howard
As far as I'm concerned, SuperPI is a useless benchmark unless you're building a massively-parallel system to crunch pi to a ridiculous decimal place. Maybe it does accurately show the math capabilities of a CPU; I'm not doubting that. But it's not useful in any other way... it doesn't [necessarily] reflect real-world performance, and that's what we're looking for.

"Real world" performance is different for everyone because we use computers for various reasons. To me, 3dsMax and DVD-Shrink are as useless as SuperPi in terms of "real world" performance because I will probably never use them.

However, I for one, unlike some other people in this thread, aren't pushing for a unilateral definition of what compromises a good benchmark and a bad one. I think the more varied benchmarks you get, the better.

Originally posted by: Duvie
How many times have we seen one cpu dominant superpi yet loose majority of real world benches...IE last round of netburst P4's...

How many times have we seen encredible 1m and 2m times to see 16m and 32m times not scale as surprising fast??

Did I even link a SuperPI 1M Score? No. Its a SuperPi 32M score and it does scale. And FYI, P4's were never that great in SuperPi; only the ones overclocked to rediculous speeds. SuperPi had always been in favor of A64's and Banias/Dothan. But ok, lets ignore those facts.

 

[Vegitto]

Guys, guys! Stop bashing. We'll see what Conroe's made of when it actually hits the shelves. I know I'll upgrade if it's worth it (and if I have the money).

 

[DrMrLordX]

The only problem I have with any forum that has users benchmarking is that the tests may not be run with the best methodology and the results are usually poorly-organized.

 

[Acanthus]

Originally posted by: dmens
Good job not explaining yourself, HAHAHA. Clueless indeed.

You dismiss superpi simply because it isn't the type of parallel workload you favor. The CAD and 3D workloads you desire are so much more specialized than superpi. Never mind MP, even in UP those workloads are run and considered seperately from the rest of the bunch when we do perf assessment. Scaling per mhz? What's that supposed to mean? Frequency results in linear gains in performance. Or are you talking about architectural effects on performance... care to dive into that discussion?

By "scaling" with 16m/32m runs, I assume you meant the bench starts taking the memory subsys into account (or did you LOL). Well then it's a more useful benchmark isn't it? And XS had plenty of those runs. Not good enough apparently.

Well, since you obviously have no idea the different parts of the CPU the benchmarks exercise you have absolutely no ground to dismiss individual benchmarks. The least you can do is be honest and admit you only want to see *your* favored benchmarks. But please don't be so arrogant as to call your choices "real" and others "fake".

The problem is their typical bench is 1m, which runs SPi entirely in cache and gives a rediculously unrealistic performance level.

 

[Howard]

Originally posted by: dexvx

Originally posted by: Howard
As far as I'm concerned, SuperPI is a useless benchmark unless you're building a massively-parallel system to crunch pi to a ridiculous decimal place. Maybe it does accurately show the math capabilities of a CPU; I'm not doubting that. But it's not useful in any other way... it doesn't [necessarily] reflect real-world performance, and that's what we're looking for.

"Real world" performance is different for everyone because we use computers for various reasons. To me, 3dsMax and DVD-Shrink are as useless as SuperPi in terms of "real world" performance because I will probably never use them.

Sure, I don't expect you to care about the benchmarks of apps you don't use. But with your logic, SuperPi wouldn't matter to you either. Unless you actually do crunch pi for some reason.

 

[Fox5]

Originally posted by: Acanthus
IIRC the anandtech conroe preview actually did have some real world tests.

As for the "complex" thead testing, over 80% of cpu tasks are simple. This is why intel chose 3 simple and 1 complex over AMDs 3 complex.

I don't think you're talking about the same kind of complex and simple stuff.
He's talking about real world apps basically. Typically real world apps are inefficiently programmed though, and much more limited by memory access performance and cache misses than how quickly a cpu can compute an algorithm.
The core stuff only relates to x86 instructions, of which most are or can be broken down into simpler instructions anyway. I think it's only vector processing that really requires complex execution units, and Intel has historically been bad at that, but hey, that's what SSE is for.

 

[dmens]

Originally posted by: Acanthus
The problem is their typical bench is 1m, which runs SPi entirely in cache and gives a rediculously unrealistic performance level.

It's 8MB, and the dataset in cache argument is meaningless. The fact that banias, dothan and yonah (with smaller caches) all did so well in superpi is indicative of superpi being an positive architectural outlier with the P-M uarch. So naturally it is extremely misleading to compare two uarchs using superpi only, or any other individual benchmark for that matter. I simply find it amusing to see people call superpi useless. All isolated benchmarks are useless.

 

[wwswimming]

http://usa.autodesk.com/adsk/servlet/index?id=5659302&siteID=123112
"3D Studio Max Rev. 8"

the Max Rev. 7 download may have been better, it bundled tutorials.
& included finished files, in case you didn't build the scene yourself.

i suggest the Commander Lake lip-syncing animation as a benchmark.

real world test, CPU intensive.