Intel Harpertown (Penryn) Quad CPUs Benchmarked

unts writes "The Intel Developer Forum is currently running in San Francisco, and Intel is showing off the up-coming Harpertown processors based on the Penryn core. HEXUS got hands on with a test system and ran some performance tests: 'Harpertown is a better quad-core processor than Clovertown: it's as simple as that. More L2 cache will gobble-up larger application data-sets and a higher FSB, on select models, will ensure that per-CPU bandwidth is less of a concern.'"

Comments

The competition is getting good

[downix]

While invariably the comparisons will bemade between this and AMD, let us not forget that Intel is getting stiff competition from left field as well. The arrival of the SPARC Niagra II processor is about to make the realm of high-end computing a lot more competitive than it has been in years. I, for one, can't wait to see a real head-to-head-to-head, AMD and Intel quads vs the 8-core monstrosty that is SPARC.

Re:The competition is getting good

[shawnce]

Don't ignore IBM's POWER processors (POWER7 is in the works) and of course Cell.

Another interesting article on the IDF and Penryn

[Null Nihils]

http://www.tgdaily.com/content/view/33929/135/
This article goes into some of the juicy technical details about Penryn/Nehalem and covers a lot of ground about what Intel had to show at the IDF.

The article is also relevant to this discussion, "End of Moore's Law in 10-15 years?". FTA:

Otellini provided an overview of the history of the insulating layer which, in modern CPUs, is only five molecular layers of silicon dioxide (SiO2) thick. He explained that as far back as 15 years ago, Intels engineers saw this layer as problematic. The continued scaling of the insulating layer could not continue forever. And, we found out later in the day with Dr. Gordon Moores keynote, that five molecular layers is about the lowest you can go in practice. Its a form of wall, and Intel was right up against it.

Much more detailed coverage available at TR

[EconolineCrush]

A much more in-depth review is available at The Tech Report: http://techreport.com/articles.x/13224

Depends what you mean by "high-end computing"

[vlad_petric]

The Niagra processor is a great idea for (web, db) server workloads, where you have a lot of inherent parallelism and very poor cache behavior. A while back, the Piranha research project figured out that for such types of applications, it's better to have many "wimpy", in-order cores than a few "beefy", out-of-order execution ones. Niagra is doing exactly this. However, outside this application realm Niagra doesn't do that well.

Bottomline: The Niagra microarchitecture is meant for a particular niche.

Aggregate RAM bandwidth is what Intel really needs

[Phatmanotoo]

Just look at thre STREAM benchmark numbers and you'll see clearly that AMD has been way ahead of Intel when it comes to RAM bandwidth. I just benchmarked a dual-Quad-Xeon myself (Dell 2900) and I could not believe the poor results I got. One app running in the system can get up to around 3,500 MB/s. Put just two tasks running together (taskset'ed to different chips), and they will each get around 2,600 MB/s. From there on, total aggregate bandwidth tops at 5,200 MB/s and stays there, no matter how many simultaneous tasks you run (it will of course degrade if you run more than eight tasks, you get the point).

Dual-socket Opteron machines from two years ago can get to 15,000 MB/s aggregate, easily.

So, I'd really like to know if Intel is planning to improve things in this department.