Haswell - Dolphin

[Mikey]

(06-04-2013, 09:45 PM)AnyOldName3 Wrote: I don't think you'e dreaming... so far.

Cheers AnyOldName3, are you considering Haswell at the moment to replace the 2320? I just hope I luck out and get a CPU that clocks nicely

[admin89]

Quote: and roughly 70% of 4770K's hit 4.5GHz

i7 4770k can hit 4.6GHz or even 4.7GHz without breaking a sweat and you only need 1.2Vcore for that (Higher Vcore is not a good idea since Haswell run hotter than Ivy)
http://forums.dolphin-emu.org/Thread-crazy-haswell-valdation?page=3

[Mikey]

Yep just quoting what the guys at Hardocp said ASUS had told them. I'll be stoked if mine 4.7 under water at 1.2V! I just hope it does it without going over 85 Degrees C
Do you know anyone on the forums who's got one yet?

[NaturalViolence]

AnyOldName3 Wrote:I can't remember the source, which is why I put the word somewhere in there.

Well then I can't take that claim seriously since I've never heard that and can't find a source myself.

AnyOldName3 Wrote:I mean the publicised roadmap that Wikipedia etc. refer to. We can't guarantee that after that point things will happen at a predictable rate because as soon as a new system is introduced, there are new factors introduced which are difficult to predict.

We can only go by what the people who are working on it say. And they say they've got it nailed down. Considering they've never missed a deadline for manufacturing technology on their x86 lineup since it was introduced 35 years ago I would say they've got a pretty good track record of being right about this sort of thing.

AnyOldName3 Wrote:Every Ivy-Haswell comparison I've seen has shown an up to 8% improvement. For example, hexus, if you need a specific source.

I've looked at Hexus, tomshardware, and anandtech so far. None of them seem to agree with that. So I still don't know where you got that number from. They're all showing around 5-13% gains depending on the application. I suppose an exact average would depend on which applications you include in your average and how you weigh them relative to each other. Still there are quite a few applications that show a 10% improvement. So 10% per year is certainly still not out of the realm of possibility.

[haddockd]

You also have to remember that Haswell was a "tock" which is generally an increase in features and efficiency. I always felt the "tick"s had the greatest improvement in IPC/performance metrics. That isnt to make a blanket statement, just a general rule of thumb

[NaturalViolence]

haddockd Wrote:I always felt the "tick"s had the greatest improvement in IPC/performance metrics. That isnt to make a blanket statement, just a general rule of thumb

Nope. That's the exact opposite of the tick-tock model.

Tocks are by definition where the big improvements in IPC/performance are supposed to happen. Ticks are mostly just die shrinks.

[AnyOldName3]

An example of 8% being the maximum

[pauldacheez]

Obviously doesn't mean it's gonna be the maximum everywhere. Not every benchmark is identical, and none are perfectly representative of Dolphin performance.

[NaturalViolence]

^Bingo.

[AnyOldName3]

But out of the range of tests hexus did, it only did more than 8.5% better on IGP tests (and a memory bandwidth test). NV said when he looked at hexus, he didn't see 'an up to 8% improvement', implying there was a greater improvement, which there wasn't.

One of the main point I was trying to make, but didn't succeed in making was that Ivy was a tick, and got a ~12% improvement, and Haswell was a tock, and got a ~8% improvement. If everything was going according to plan, Haswell would have shown greater improvement than Ivy, but it was smaller, implying a large deceleration in improvement, meaning 30% may be more than 3 years away.