ScienceMark Author Talks SysMark

Posted By Van Smith

Date: August 16, 2001

The author of one of the leading independent scientific benchmarks speaks out regarding SysMark.  In a note sent to several members of the media, Dr. Tim Wilkens, author of ScienceMark, responded to recent criticism sent toward VHJ regarding our SysMark article.

With the author's permission, we are publishing the note below.



Van posted an article in rebuttle to yours here:


I have to side with Van on Sysmark's usage. I see it used on so many sites.. and it's like the ignorant person's benchmark of choice. I wonder what is going on actually.. are the numbers checked to make sure that multiplication, addition etc are being done accurately (this is a super big concern), also how does on refute the obvious bias towards SSE and possibly SSE2 with no 3DNow!, though the later was the first to market. Van's table or chart in his first article clearly says something, and it's strong evidence. A benchmark simply can not be biased. That is why on ScienceMark:


I used "reliable", "publicly available", "fair" optimization. There are no ASM instruction in my benchmark, the only optimizations are those chosen at compilation time, and I've tried using Intel's compiler... but it is taking 9 months to get some action there. In fact, here is my latest message to intel on their support page, expressing my concerns about their great "SPEC" compiler, and their faulty-unstable product:


Thanks XXXXXX,

Though I think it is fair to everyone that I express my frustration with these multiple issues that have arose as a result of compiling ScienceMark with a "publicly" available compiler. I'm very frustrated with this product, which seems to have the ability to break easily.

I tried compiling ScienceMark on an Itanium, which I have access to, and "of course" this chip won't even compile ScienceMark. MKL is broken and it's performance is poor compared to a 3 man team ATLAS (http://www.netlib.org/atlas), but all I hear is how Itanium is this and that and how the compilers work flawlessly from others in the know. But what I find is the exact opposite and I suspect any results quoted by Intel concerning P4 and Itanium performance are likely hand optimized and compared to competitor's results.. obtained from stable compilers without hand optimizations.

Excuse me for being a bit upset, and pass this on to whoever you choose, but I think it's "extremely" distasteful that this product, which compiles benchmarks such as SPEC and others posted by Intel without a flaw, and still, after 9 months, has issues compiling ScienceMark, a code that's compiled and run on IBM, DEC, CVF, PGI, SGI, etc. I am also disgusted that this compiler, which evidently is worthless, has to be replaced with one, not developed by Intel but bought from Compaq. I can just imagine how flexible and dependable that compiler will be come. What will become of support for other chips besides Intel's. Surely you must share some of these frustrations and concerns? They are not arising as a consequence of my favoring Athlon over P4, but from the proven performance advantages and compiler surperiorities of the Athlon + CVF over PIII/P4 + Vtune.

I'd like to get it compiled on the Itanium and see what performance I can truely hope to get, but I must compile it myself, with my own source code. If anybody at Intel can help me out with this or in obtaining a P4, let me know. The hd ware sites that presently use my benchmark are:

http://www.tech-report.com, http://accelenation.com/?doc=41&page=1,





http://www.hardocp.com(on one occasion),

and a few others, but the list is growing.

Thanks again XXXXXX for your hard work, but I felt I had something to say concerning the past 9 months of work on this issue.. I'll check in the future for new developments..

Tim Wilkens =======================================================

So I'm a bit upset with the misinformation, faulty biased benchmarks available for Intel and "many" hardware sites who use SYSMARK. AMD just made a better product, they are beating intel with brains at the moment, but if Intel can finagle some sort of biased benchmarks then they will succeed. That's the answer to your question on why the Athlon is beating Intel. But if you want to see some real pretty biased benchmarks.. scruitnize the Itanium results when they come out and how they were obtained.

But as a hardware site, it is the editors job to "intelligently" educate the viewer, and I thought your post didn't. I'm sorry but thats my view, you have yours. Nice article Van,


Tim Wilkens


Pssst!  Our Donation Page is up.