The Conroe a Hoax?

[Alaa]

Originally posted by: n19htmare

Originally posted by: sandorski

Wow, u pwzd me!!

You forgot to quote this: "They are actual Apps that people use to work or play with." Evidently nether of you read it either.

What do you wanna know? How long it takes to load photoshop? How long it takes for start menu to pop up? how long it takes for a DVD to play? how long it takes for firefox to load? how long it takes for Microsoft outlook to open? how long it takes the calculator to add 2+2? How long it takes to encode a video for which the time has no meaning as it cannot be compared to anything esle? how long it takes to render an image in 3dmax for which again the time has no meaning as it has nothing to be compared with?

These are All real world applications. Yet of no relevance....

Name some ACTUAL APPS that you would like tested and how you would like them tested. We will be Glad to put it forward to the Conroe owners.

ENOUGH of this real world stuff, I want the names, and i want the process of testing it... nothing else.

these synthtic have no revelance..i remember tht peintium 4 used to outperform A64 in synthtic, yet it cudnt outperform it in (for example) games, becuase these tests are just crap! i have pentium4 proc so am no amd fan

 

[sandorski]

Originally posted by: n19htmare

Originally posted by: sandorski

Originally posted by: Absolute0
i totally agree with n1ghtmare 100%, those "real world" tests won't do you any good unless you can compare them to something else :S
If he takes some random image in photoshop and starts applying filters, i don't think many people are going to care.

I think we can all just rest assured that Conroe will do well in "real world" as well as in Synthetic. Why not? I mean you guys are at AT, you should have read the article explaining the Conroe architecture and the superior instruction processing.

You do compare them to something else, a similarily configured PC with either a comparable AMD processor or previous comparable Intel processor or preferably both. "A point of reference" for the sake of a point of reference means nothing. The only "point of reference" with any meaning is the experience of the end User.

EXACTLY my point. That would require multiple systems on both platforms. What do you do when you dont have multiple systems that are comparable.... YOU RELY ON SYNTHTIC MARKS to reflect real world performace. And to say that they dont reflect real worl performance is ridiculous.

You think a 15sec Sp1m computer will render an imager slower than a 30sec sp1m computer...
or a computer that scored 21000 in 3dm01 will do better in 3d than a computer that scored 59000

ComeOn people...its a good indication on how the systems will perform in real world.

Synthetic benchmarks *do not* indicate that at all. SiSoft, for eg, has always shown the P4 to be "superior" than practically any Athlon, but Real World showed that to simply not be true. Other Synthetics showed this or that specific "strength" or "weakness", but Real World applications don't do one thing repetitively, they use a mix of differing aspects of a cpu in order to perform their tasks. So while this or that processor has a theoretical advantage as shown by Synthetics, those advantages can quickly evaporate when a complex mixture of instructions is required.

Synthetics offer nothing for the End User. Maybe an Engineer or some IT person somewhere can use a Synthetic benchmark because the task needed to be done is very specific, but that's a very small niche to worry about.

 

[Absolute0]

Originally posted by: Njkid32
WOW I am just surprised with the ignorance here on this site!!! REALWORLD benchmarks like what opening a webpage or clicking on the start menu. You guys wouldnt know a REAL benchmark program is it smacked you in the face!!!!

At least someone isn't seeing everything through FanBoy Vision


Conroe naysayers, you want gaming?

http://www.anandtech.com/tradeshows/showdoc.aspx?i=2713&p=2

Strange, being AT readers you should have seen these before. Oh but of course, we're AMD guys so obviously a benchmark where Intel won must be rigged, even though a dozen web sites ran these gaming benchmarks a dozen times and checked both control systems out to be sure it was fair.......

When you've had the benchmarks for months, but choose to ignore then, and then say that all new synthetic tests are null, it implies some kind of obscene arrogance that i just can't comprehend.

 

[sandorski]

Originally posted by: Absolute0

Originally posted by: Njkid32
WOW I am just surprised with the ignorance here on this site!!! REALWORLD benchmarks like what opening a webpage or clicking on the start menu. You guys wouldnt know a REAL benchmark program is it smacked you in the face!!!!

At least someone isn't seeing everything through FanBoy Vision


Conroe naysayers, you want gaming?

http://www.anandtech.com/tradeshows/showdoc.aspx?i=2713&p=2

Strange, being AT readers you should have seen these before. Oh but of course, we're AMD guys so obviously a benchmark where Intel won must be rigged, even though a dozen web sites ran these gaming benchmarks a dozen times and checked both control systems out to be sure it was fair.......

When you've had the benchmarks for months, but choose to ignore then, and then say that all new synthetic tests are null, it implies some kind of obscene arrogance that i just can't comprehend.

This isn't about, for me, AMD vs Intel or even about Conroe, it's about Synthetics and their uselessness. I don't agree with the OP or his link for that very reason.

 

[akugami]

I'll wait for final benchmarks before saying Conroe is a bust or it's the next best thing. Early buzz is that it looks to be a powerhouse. Synthetic benchmarks are useless in real world applications. I can overclock my RAM and it could show a 20% boost in bandwith and other measurements but in real world usage, such as gaming, I probably will see a frame or two more per second at best. There are just too many variables used when running an actual program like games, video encoding and the like.

 

[BrownTown]

I think people here are nto quite giving "synthetic" benchmarks their due. Certainly things like Sandra numbers have only a moderate correlation to "real world" apps, but they deffinitely have a correlation. When it omes to other benchmarks like 3DMark etc, there is a very good correlation between 3DMArk score and in game FPS, no its not right 100% of the time, but it is a very good indicator, especially when you understand that even "real world" games show pretty large differenace in what hardware runs them best. SuperPi isn't that good, but if you understand what it is showing then it does provide some usefull information, the 1M time is more of a pure CPU and cache benchmark, while the 32M is a good memmory system benchmark. Also, alot of these "synthetic" benchmarks use algorithms from very popualt programs, like compression algorithms or FFT algorithms that are used widely in the real world. Also, please note that CPU benchmarks are designed only to show a CPUs strength, and therefore "real world" programs won't and SHOULDN'T be affected nearly as much since they are using the whole systemand not jsut the CPU. Also, and fool who uses a game as a benchamark to show CPU strength is foolish, games are almost always CPU bound unless you purposely set the settings to very low resolutions and quality.

In conclusion: there are already more benchamarks from Conroe than anyone can possibly keep up with on countless websites, the vast majority of people here probably havent even seen the half of them. Yes, MANY different games have been benchamrks by people at CS, and those who were lucky to get access at IDF. Also, "real world" application like WinRar have been tested. And so ahve 20 differnet CPU and GPU benchmarks. IF you think all that is out there is SuperPi scores then you are wrong, the SuperPi scores get all the attention becasue they are so insanely good, but the other scores are there too if you dig deep enough (actually alot of them have been deleted now since so many people were comming to XS and b|tching, but they are will on Victor blog)

 

[dexvx]

Originally posted by: Alaa
these synthtic have no revelance..i remember tht peintium 4 used to outperform A64 in synthtic, yet it cudnt outperform it in (for example) games, becuase these tests are just crap! i have pentium4 proc so am no amd fan

In what? Last I recall, 3dMark was dominated by AMD X2 systems.

PCMark, Sandra, SpecINT/FP etc are not indicative of gaming performance. They are indicative of peak CPU performance. And as shocking as it may seem, Pentium4's and Pentium-D's does tack up well to their A64/X2 counterparts in raw CPU power, if memory bandwidth is not an issue. It has been proven that XeonMP's with an IBM Hurricane setup will match or exceed Opteron 8xx setup in 8 to 32 way.

Originally posted by: sandorski
Synthetic benchmarks *do not* indicate that at all. SiSoft, for eg, has always shown the P4 to be "superior" than practically any Athlon, but Real World showed that to simply not be true. Other Synthetics showed this or that specific "strength" or "weakness", but Real World applications don't do one thing repetitively, they use a mix of differing aspects of a cpu in order to perform their tasks. So while this or that processor has a theoretical advantage as shown by Synthetics, those advantages can quickly evaporate when a complex mixture of instructions is required.

Its more to the fact that people like you don't understand what Sandra's CPU test means. It shows the peak output of a CPU that is not bottlenecked by anything (such as memory bandwidth, HDD speed, etc).

Originally posted by: sandorski
Synthetics offer nothing for the End User. Maybe an Engineer or some IT person somewhere can use a Synthetic benchmark because the task needed to be done is very specific, but that's a very small niche to worry about.

Synthetic testing is quite common. Things like SpecINT/FP or TPC is used as a common benchmark by IT's.

 

[TAGImperialMarch]

If this hasn't already been established, Sharikou is a biased AMD fanboy who censors his blog for anyone that says anything smart for Intel that he doesn't have an answer to.

 

[dmens]

Originally posted by: Duvie
Problem is most synthetic benches are not good comparisons when crossing platforms...some have biases towards one particular cpu yet do not translate into real world performance in real world benches...

LOL, it's the benchmark elistist again. Ever realize the fact that synthetic suites are designed for cross-platform comparisons? Single tests are vulnerable to architectural outliers that make overall comparisons meaningless... but you're obviously not interested in overall comparisons, only your very own suite of duvie's own benchies. HAHAHA.

Originally posted by: sandorski
Synthetics offer nothing for the End User. Maybe an Engineer or some IT person somewhere can use a Synthetic benchmark because the task needed to be done is very specific, but that's a very small niche to worry about.

On the contrary, good synthetics emulate averaged workloads that would interest a wider audience. Running a specific application (the so-called real life test demanded by some in this thread) is the very definition of a niche workload.

 

[firstdivision]

I personally am waiting for someone to bench a Conroe set up and an AM2 set up using Cubase SX 3.1 and various vst synths so I can see which one will be able to handle the most. But that is because I really only do audio production work with my newer computer.

 

[Amaroque]

I do believe that Conroe will beat the K8 in just about anything vectorized. Even though AMD was the first to release a CPU with vectorized FP (3D NOW), Intel has implemented it much better in their CPU's. Intel was also first to release vectorized Int (MMX).

The K8 x86 FPU may or may not be stronger, but that is not "real world" as most anything that can be vectorized, is vectorized in this day and age.

As far as boot times...(useless) HDD limited.

Clicking a link... (useless). That's contingent on your Internet pipe. A PII with a 4Mbps connection would kill even an FX60 on dialup.

I could go on, but I don't want to be too wordy. In short, "real world" would be the CPU bound software you run.

My 2¢

 

[dug777]

I don't rly get this thread. We've see countless benchmarks of Conroe beating FXs/Optys in all areas of application, surely an intelligent person would be capable of extrapolating the general conclusion that Conroe is faster clock for clock in pretty much any application?

Are the naysayers hopefully looking for one app that shows an Fx/Opty beating conroe?

I guess at the end of the day i think i'm all for just rolling over & saying 'sure, conroe will be slower than a p4 williamette, clock for clock' and waiting for AT to get a retail sample (which shouldn't be far away)...

 

[F1shF4t]

Conroe will probably be faster, but in the end who cares, 4 years down the road they will be both be slow, plus the speed advantage really wont make night and day.
I just dont get why people are trying to proove crazily either way. Well i can see one valid reason besides fanboyism, and that is buyers remorse, "why did i spend money on my athlon when something better is coming". Time is money think of how much u would have to pay if u had to rent ur quick comp while the new thing came out.

 

[Effect]

What's with the 'naysayer' and 'fanboy' comments coming up? I'm sure we've all established that conroe will, by all appearances, be better (clock for clock) than K8's. The point here is that while synthetic benchmarks have their place, they dont tell you much about how the performance we (as end-users) will be paying for. I can see by reviews that if i get BlaBla video card with YadaYada CPU, i'm likely to get somewhere near X fps in whatever game. Using synthetic benchmarks (like 3dmark and superPI), i can see that upgrading from a P4 2.8ghz w/ 512mb RAM to an X2 3800+ w/ 2gb RAM is a good idea (it's obvious anyway, but that's just an example). Not the best explanation, but i hope that give some insight.....(i'm tired, so that's the best i could do, haha).

 

[dexvx]

FYI, supposedly "Real World" benchmarks are also not indicative of application performance either. Since an application can vary on use, it does not necessarily accurately represent the performance that user will receive. Simple example:

Anandtech did the infamouse closed box Conroe vs X2 test. Using an Intel supplied timedemo of Quake4, the Conroe was about 20% faster in average FPS. Using the official timedemo of Quake4, the Conroe was about 30% faster in average FPS. The 10% offset could EASILY sway the performance crown if Conroe and X2 matched clock for clock.

One could argue that a timedemo of a game (or encoding a particular movie/rendering a specific file) is just a "synthetic" benchmark of that certain application.

-----

Well anyways, I'll just copy/paste what one of the authors of ScienceMark wrote (via Aceshardware forums):

Conroe's score is amazing.

Keep in mind that he tested both the Pentium-M optimized binary and the Pentium 4 binary. Out of both binaries, the Pentium 4 was faster.* This is compiled with Intel's latest *publicly* available compiler.

Conroe is more similar to the Pentium-M than to the Pentium 4, but the binary doesn't utilize Conroe's wider resources. In *scalar* code, x87 code is generated by the Pentium 4 optimized binary because the Pentium 4 can't execute pipelined scalar sse adds, but can in x87 mode. (I had thought this was not the case but Intel's optimization guide says this is the case, I'm not sure if I 100% believe that, but given that x87 code was generated with the most aggressive flags possible -- sse/sse2 code was used sparingly in with some of the x87 code).

I imagine that if you ran the 64-bit binary Conroe's lead would widen even more.

What I am perplexed about is why Conroe bombs on the encryption code -- the entire instruction mix consists of BSWAP, XOR, and various MOV instructions, none of which are micro-coded on other processors, and no jump instructions. With Conroe's wide integer resources, I don't see why this is the case. I'm guessing there is an address generation limitation, but since I don't have a Conroe I can't really test that theory. It's a blind guess.

In any case, I see this as a *strong* showing for Conroe, not a weak one as the blogger claims. In the benchmarks that matter (BLAS, MolDyn, Primordia) Conroe is at least equal to if not exceeding an Athlon at the same clock speed. The fact that in 32-bit mode, an Athlon64 clocked 400 mhz higher cannot exceed Conroe's performance gap is telling.

I can't wait for a 64-bit run....

* This should shut some people up that the Conroe is just a souped up Pentium-M, which some people claim is just a souped up Pentium3.

 

[Quinton McLeod]

You people must realize that entire programs can fit in the Conroe's 4MB cache. Programs simply run through that and it, quite literately, runs its process on the cache. Once you use up that 4MB cache, that's where the problems start. That is what is being said. No one dared to test it out. It may be due to the fact that everyone is too scared to question the Conroe because their feelings may get hurt. Put all this hope into one processor only to find kinks in its armor.

 

[Absolute0]

You have no idea what you're talking about. The only thing that COULD "fit in conroe's cache" is Super Pi 1M. And that doesn't explain why Conroe rapes at every other benchmark out. It's the attempt of a scared mind to account for performance you don't want to believe.

 

[Cooler]

You have to remember that no single program can use the full 4mb of L2 cache. The OS will take up some of it to begin with second these same tests have been performed on Merom and they have all most the exact same scores. (Merom has 2mb of L2)Thirdly the cores must share the full 4 mb of L2 cache so one using all of it is highly unlikely. If you read andtechs article on about one of archs main advantage is its ability to move load commands around depending on free processor resources which no other processor can do at this moment. It does not mater how fast you can get data from ram if you are still waiting for function to execute inside the CPU.

 

[n19htmare]

Originally posted by: Quinton McLeod
You people must realize that entire programs can fit in the Conroe's 4MB cache. Programs simply run through that and it, quite literately, runs its process on the cache. Once you use up that 4MB cache, that's where the problems start. That is what is being said. No one dared to test it out. It may be due to the fact that everyone is too scared to question the Conroe because their feelings may get hurt. Put all this hope into one processor only to find kinks in its armor.

You have been reading too much of that shriuku non-sense. That was a theory by some narrow minded people which was proven to be wrong very quickly.

Heck a 1M Pi doesn't even fit in the remainder of the canche... let alone a 32M pi, and 32M pi scores rape any other sp32m pi scores.

You think the whole 3dmark suite fits in the 4mb cache?

You are clueless.

 

[Njkid32]

Merom has 4mb of L2 cache just like conroe. Its the mobil version. You might be thinking of Yonah...

 

[d3lt4]

great just what we need... more conroe hype. I don't believe that the reviews are all that legit, and like mentioned earlier, it's only one test. Just get over it, conroe is better than A64, but I agree it isn't perfect, everything has a flaw or more, but that doesn't mean that it sucks.

 

[Cooler]

Originally posted by: Njkid32
Merom has 4mb of L2 cache just like conroe. Its the mobil version. You might be thinking of Yonah...

lower versions have 2 Mb of L2.