mdk777 wrote:OK JohanAT,
Please explain this review:
ASUS F1A75-I Deluxe Review – Llano and Mini-ITX
http://www.anandtech.com/show/4833/asus ... nd-miniitx
You use a 1000 watt PSU to determine the efficiency of a 50-120 watt system ?
You have to know this will give you totally skewed and totally inaccurate results.
The 1000 watt PSU in question will be well under 80% efficient. At the idle range it will be closer to 70%.
If you are looking at idle wattage under 100, you should be looking at a 250-400 watt PSU. several
good gold rated units in this range could have been used:
You would then record 38 watts at idle and 108 watts at load.
not the 50 and 124 you report
Why would anyone intentionally de-tune the reporting of a system
Curious minds want to know.
Smartidiot89 wrote:The writer for that article looks to be Ian Cutress, looks like Johan is the wrong guy to ask for an "explanation"
mdk777 wrote:Well, I was under the impression he had editorial review status...
However, just my assumption I guess.
Did no review of the staff organization, so you could certainly be correct.
However, If I was working for a site that put up such a review, I would be concerned about it reflecting poorly on my work by association none the less.
Pietro sk wrote:Smartidiot89 wrote:The writer for that article looks to be Ian Cutress, looks like Johan is the wrong guy to ask for an "explanation"
It doesn't matter because It's published on AT website, that means they are responsible.
mdk777 wrote:Well, I stand corrected.
And if such a 25% error was intentionally reported, and got past the editors....you could bet they would fire the writer, retract and correct the result and apologize to their readers.
However, even when multiple people point out the obvious error in the comments...
Anandtech let the article stand.
Now do you understand why everything that is retorted there is immediately suspect to those who understand the details
You asked earlier in the thread why the results reported at ANandtech would automatically be disregarded...
Now you have your answer.
JohanAT wrote:My latest article http://www.anandtech.com/show/4486/serv ... rk-session contains two benches that are very "intel CPU friendly" and I am currently investigating what made them Intel CPU friendly and if that is relevant to the real world and. I had a lot of good pointers from several people. If those people would have immediately said "hey it is an intel site", people can not be educated.
In short: you gain nothing by simply bashing a site/person, especially if that person wants to hear your arguments. You can gain a lot by reasoning and debating.
JohanAT wrote:For example: Euler3D CFD is a benchmark compiled using the Intel Fortran compiler. You could say that is "against AMD", but the fact is that the developer does not get good results with other fortran compilers on the AMD platform. The fact that Intel has produced a decent fortran compiler gives them an advantage on that part of the HPC market. Thoughts ?
duby229 wrote:Yeah, but an Intel fanboy will argue with you until hes blue in the face that the results will only be comparable if the same compiler is used. I know its redonkulious, but I've been in that argument far too many times.
But on the other hand it is my experience that results from different compilers are almost always well withing the margin of error. The simple truth is that you can get much larger gains by optimizing the code for the architecture, which puts AMD at a disadvantage. Or they can optimize the architecture for current software, which AMD has more or less refused to do...
My personal opinion is that the best results obtainable for a given application on a given architecture is the best comparison between them.
97% of the benchmarks that are being thrown around dont indicate anything at all.... NOTHING.... Well maybe it does indicate the massively pervasive incompetence in the enthusiast tech market. (Which oddly enough is being encouraged by some important names in the SERVER market) (Its almost as if they wish the death of enthusiast desktop computing)
Widening the scope of the thread slightly, I found that the same behaviour exists in the IPP library (ie fast vector code for Intel and suboptimal code for other) and again it was due to a cpuid rather than feature check. The library was nice and fast for the image manipulation I wanted to do.
Fortunately the workaround was very straightforward (though may violate the eula?). If you write a lib with the single function ippGetCpuType(*) and LD_PRELOAD it, then you can tell the IPP lib that your cpu is <whatever> and it will run the code path you wish to run. I mentioned this on AMD's dev forums a few years ago. It was instantly removed and was politely informed that it was part of the action against Intel.
The perf difference between the generic "ippCpuUnknown" and "ippCpuP4HT2" (the best at the time) was amazing (from memory it gave the Opteron box I was running it on a >60% perf increase, making it significantly faster than the P4 Xeon competitor)
* ippGetCpuType is documented here: http://software.intel.com/sites/product ... uType.html, http://software.intel.com/sites/product ... /index.htm
I think the IPP has been fixed in version 6 to work well on AMD. See my C++ manual.
Apparently, that confirms my suspision that AMD have been prevented from discussing the issue publicly.
abinstein wrote:If we compare BD core and SB core, then BD will lose IPC, on average. Since the only sensible way of comparing IPC is by running the same binary, such comparison preclude the usage of FMA and XOP, and by definition handicap BD.
jdwii wrote:Because most Programs barely make use of 3 cores
Currently it takes a 4.6-4.8Ghz on a Phenom II x4 to tie the 2500K clock at 3.3Ghz.
jdwii wrote:If a 4 Module bulldozer clock at 3.6Ghz can't beat or meet a 3.6Ghz 4 core sandy then Bulldozer is the lesser of the 2 in terms of performance. Just like a Phenom II x4 clock at 3.0Ghz loses to a 2500K clock at 3.0Ghz. Regardless of the reasons why its slower.
Fact is the only people buying bulldozer will be PC enthusiasts and Server people, Llano is for the Average PC user. If bulldozer does not win in these areas then it will be sad.
i thought they where making a beast.
scientia wrote:Not really. BD will be the core in the next generation Llano; it will be mainstream in the future.
scientia wrote:abinstein wrote:If we compare BD core and SB core, then BD will lose IPC, on average. Since the only sensible way of comparing IPC is by running the same binary, such comparison preclude the usage of FMA and XOP, and by definition handicap BD.
Well, unless you run hyper threading on SB in which case their IPC takes a very big hit.
Users browsing this forum: No registered users and 2 guests