Replies to post #49611 on Intel Corp (INTC)

[wbmw]

Re: only when that opposing viewpoint is not properly represented.

If you think that people's descriptions of AMD are under-representing their capabilities, why don't you say, "I think your description is unfair and does not properly take into account Barcelona's strengths?" Why must you use hyperbole, and say things like, "It will be like rolling thunder month by month and its Intel's server share it's going to rain buckets on?" #msg-23265129

Do you really believe that Intel's server share is going to plummet as Barcelona ramps, or are you merely using blunt rhetoric as an attacking point to make sure that there are as many over-representative posts of Barcelona as there are under-representative posts...?

I don't mind you setting the record straight when someone fails to give Barcelona its due respect (in fact, I encourage it), but I strictly detest using unwarranted language with the excuse that someone's misrepresentative negative comments give you the license to say something equally misrepresentative.

Re: You guys are writing off Barcelona without seeing its full performance profile as I'm sure it will do better than Clovertown in TPC-C and SAP at least per clock as well as SpecWEB.

I'm not doing this. However, I am calling attention to the fact that AMD has not yet employed these benchmarks to show Barcelona performance, and instead they concentrated on a small subset of scores that I believe misrepresent the overall performance profile. When I see 3rd parties like the Tech-Report showing Barcelona failing in every benchmark, and then I see you arguing that these benchmarks are not as important as SPECfp2006_rate, which happens to be one of only a small handful of applications that AMD has actually released so far that show Barcelona in a positive light, then I feel like the situation is again being misrepresented. I'm happy to revisit TPC-C, SAP, and SPECweb2005, later when AMD has submitted these scores, but until then, I can't assume that Barcelona will do well in these, because my first reaction is to assume the opposite, since I would think AMD would be proud to display these achievements, if in fact they did well in them.

Re: You are also ignoring totally its virtualization features.

Barcelona comes with improved Virtualization, but so does Penryn. AMD hypes up their capabilities, but so far, no one has tested Clovertown and Barcelona side by side in a virtualized environment. So that puts this feature right in the same bucket as AMD's so-called performance with TPC-C, SAP, and SPECweb2005.

Re: Concentrating on basically workstation benchmarks and saying it is a failure is not doing your proper DD and I'm telling you early for your own benefit as I already am aware of Barcelona's all round performance and how good it is.

Pointing out Barcelona's weakness in workstation benchmarks goes right in the face of AMD's claimed benefits in SPECfp2006_rate based applications. I've looked at the descriptions of SPECfp2006_rate, and I've looked at the descriptions of benchmarks used in the Tech-Report article, and there are many similarities. I have pointed this out before, and it's part of what I consider DD (Due Diligence). For you to ignore these while pushing SPECfp2006_rate as a quintessential qualifier for HPC workloads is demonstrating a lack of your own DD. In the end, if you are posting to set the record straight and make sure descriptions of Barcelona is fair, then you should start by leading by example.

[mmoy]

> You are also ignoring totally its virtualization features
> as well as its better SpecCPU performance using non-Intel
> compilers.

You shouldn't have to make that caveat. Intel's compilers have been shipping for a while with support for a processor that isn't yet shipping. Intel has its act together on compilers. That's more than I can say for GCC and Microsoft in the performance realm. One could say that Intel's compilers are part of the chip package along with their chipsets.

A few of the folks asked my opinion on Core 2 Duo back when it was introduced on whether performance would be quite a bit better when compilers came out to support Core 2 Duo. Intel shipped their compiler with Core 2 Duo support and I think that they had somewhere around a ten percent improvement. GCC and Microsoft aren't there yet. Core 2 Duo has been out for 18 months and a company like Microsoft still only gives you NetBurst and K8 optimization. Perhaps Microsoft will add Barcelona and Core 2 Duo support in VS2008. I don't even follow the GCC release schedules as with GCC, you just wait and wait and wait. AMD has put some efforts into GCC for Barcelona so they should have support one of these days but GCC releases seem to spend a lot of time in Beta.