IntelUser2000 (Intel_user2000@yahoo.ca) on 1/24/12 wrote:
---------------------------
>David Kanter (dkanter@realworldtech.com) on 1/24/12 wrote:
>---------------------------
>>So what is your estimate for the performance of Medfield on SPECint2000 relative to say, OMAP5, Tegra3 or S4?
>>
>
>I agree with Wilco on this one. There's no way a single >core 1.6GHz Medfield will
>surpass similarly clocked dual core A15 in ANY benchmark.

I'd be very surprised if it did too. But I'm not sure what frequency we'll see and under what circumstances.

>If we assume the Sunspider benchmark represents real world >case, Medfield is anywhere
>from equal to 10% slower per clock compared to A9. Good >thing Medfield is clocked 30-60%(depending on Burst) >higher.

>That's on a single thread benchmark. However, the competition is dual core. Hyperthreading
>on Atoms do a lot, but dual cores do way more.

Agreed.

>I've seen >benchmarks of 2 core Atom
>with Hyperthreading disabled vs Hyperthreading enabled >single core Atom. The 2 core is 30 or so percent faster.

OK. I'm not sure what that says about A9...actually honestly that tells me that Intel made a really good choice in terms of hyperthreading.

Generally HT adds about 20% power consumption, but you're saying it tends to add around 50-70% performance, so that's a fantastic trade-off. Given that info, I can see why they wouldn't bother with another core.

>Then maybe, thanks to Hyperthreading, 1.6GHz Medfield can >be on par with 1.2GHz
>A9 parts. Now you see why suggesting equivalence to higher clocked A15 parts are
>ridiculous. Assuming only 20% IPC gain, and 1.5GHz clock speed, the A15 part will
>have 25-30% advantage in single thread and about 50-60% in >multi-threads. Now most
>people are expecting IPC gains of 40%+.

What is the power consumption though? Essentially what does the curve of perf. vs. frequency look like?

David