Dean Kent (dkent@realworldtech.com) on 12/4/05 wrote:
---------------------------
>Temp (Armand.Hirt@caramail.com) on 12/4/05 wrote:
>---------------------------
>>
>>Frankly, that makes one wonder if the app does check for CPU brand ID in order
>>to get to know whether it supports SSE :-)

I could only find the Sungard app benchmark mentionned in Intel promotional material, before it shipped with the Bensley review systems. Let's just recall that Intel commissioned principledtechnologies to do a competitive comparison of Intel vs AMD chips. Whatever Intel and Sungard did together, it warranted a press release by Sungard that reads like a Xeon ad: "SunGard’s Adaptiv Credit Risk is developed and deployed on powerful Intel Xeon processor-based servers that deliver the price/performance, availability and flexibility to support high trading volumes and large numbers of users.". Finally, Intel now ships this benchmark with some review systems.

target="_blank">http://www.sungard.com/software/news_5/sungardsadaptivcreditriskannouncesnewadvancesinperformance.htm

target="_blank">http://www.principledtechnologies.com/clients/reports/Intel/SunGard_ACR_perf_jun.pdf

>If that app is fairly popular, and I am trying to evaluate what a typical user
>will see, then it makes sense to use that application whether suggested by Intel
>or not. If I use only that app, then it is a crappy review. If I use several other
>apps, some of which have not been optimized, then users can see the range of performance
>they can expect.

I think you are placing the bar quite high, since I am afraid it will be difficult to find even a single other commercial application packaged as a benchmark in this market, let alone several ones. It is generally not possible to know how and to what extent applications have been tuned to a specific platform or vendor MPU either, unless in the rare cases where the vendors bang their drums as hard as Sungard did, which even only gives clues. So your points seem either quite theoretical, or would probably lead to exclude Sungard from a cross platform review that is not "crappy".

However, a bit of context around the application is always welcome, isn't it? Or do you also object to it?

>If the app happens to check for CPUID, or some set of bits - that might be nice
>to know in a comparision between two manufacturer's products - but in the end, if
>that is how the app runs in the field then that is what the user will actually experience,
>true?

If the set of bits is the CPU Brand ID, I think it is relevant to the benchmark analysis, since for instance the ISA extensions supported can be determined in other ways.

>Once again, we go back to my original position - it is better to understand the
>reason why an app was selected than to just spread innuendo and doubt.

If it is what you refer to, I understand quite well why the benchmark has been selected in David's original review. I also think I understand why the benchmark set was so close to the one used at Tecchannel. I understand because I asked, and not because I supposed to have all the answers from the start.

>The former
>indicates an open minded investigation, the latter suggests an agenda.

It is apparent that the occasional need to ask questions in order to understand what is going on may elude you because you do not always feel that need personally.

- Armand