JPM - Open discussion on next gen CPUs please?

[DrPop]

Hi JPM - and anyone else that wants to pitch in, of course!
I read an article on Anandtech recently that was talking about Intel's first big symposium on Haswell . . . the point of this post being, that they mentioned there will not be much of a performance improvement on the CPU side over what is out now - I think they were saying like 10% or around there. So it will be essentially like an Ivy Bridge update over a Sandy Bridge. That wasn't exactly much, especially considering how high the SB overclocks! Apparently huge strides are being made in this next-gen CPU, but mostly all on the built-in GPU side of things.

Now for my question: Why should we care about Haswell, etc - if the big improvements are in the onboard GPU and not on the CPU side? Or is there more to it than that? Like increased RAM, BUS timings, better chipsets, etc - that would result in the CPU feeding the GPUs faster or more efficiently than our CPUs do now?
Lastly, do you think the onboard GPUs built in to Haswell and future Intel chips will be powerful enough to crunch with? Or would that even be a consideration, given the insane heat GPUs put off? If I had the choice, I would rather NOT have the GPU built in, simply for heat exchange purposes if nothing else. Crunching on an internal GPU (if possible) must severely hamper the CPU crunching due to power consumption and heat output.
Thanks for any discussion.

EDIT: Found the Article on Haswell here.

[Mike029]

On my 2600k I run the monitor on the integrated gpu and leave the Cuda or AMD free to only crunch. Not a huge advantage but noticeable. I think Hankey did some testing and found it to be a nice improvement over having to use the crunching gpu.

[John P. Myers]

Well..a few months ago i was excited about Haswell and posted about it plenty, but since then Intel seems to have been backsliding. I'm not happy about it and would stay with sandy bridge before i upgraded to IB or haswell. They've scrapped the idea of the 2nd companion chip holding L4 cache as well as the idea of any CPUs with more than 4 cores. For those that overclock, SB is still the better deal since clock rates can be brought up higher than the IB counterparts due to Intel using garbage thermal grease between the heatspreader and the CPU instead of directly soldering it.

However, Haswell-E for socket 2011 will be decent, though it is not compatible with current socket 2011 mobos, and this is not fixable with a BIOS update. And it still won't be as good as Intel originally intended Haswell to be.

It makes me angry and disappointed at Intel, even though i know it's AMD's fault for continuously producing garbage that Intel beats without even trying very hard. And since they don't *have* to try hard, they don't. Same reason Nvidia renamed the 660Ti to the GTX 680, because it was good enough to give AMD a run for their money, but not as good as it should've been. The consumers and the overall progress of computational evolution pays the price for this.

Moore's law is a joke. It is being intentionally adhered to, as if the best they can possibly do just happens to be what Moore's law states they should be able to do. It's not the case and this is an obvious example. They dumb down their chips based on what the competition can do, not based on what they could really do. Btw, Moore was one of the co-founders of Intel. Coincidence? pfft

On the plus side, the enthusiast Haswell chips will have weaker on-die GPUs than the mid-range CPUs. Imo, it's still more GPU than one should find on an enthusiast chip (it should be none, with 2 extra cores in place of it). But as i've said, no one's opinion counts. The only thing that matters is AMD's inability to make something challenging.

[John P. Myers]

I will say that if you do video transcoding, definitely go with Haswell.

One of Intel's main arguments is no one needs more than 4 cores, and if they do, they can buy a Xeon. Well, Intel doesn't know us very well AMD is coming out with the Piledriver CPUs shortly. Can get an 8-core CPU for ~$260 (FX-8350). Sure it might *still* be 32nm, still only have 8MB L3 cache, still be 125W TDP, but for our purposes, it will be faster (not per core, but overall). October 23rd. 4GHz base clock speed.

Edit: oh yeah, along with everything else Haswell won't be doing, it won't be supporting DDR4 anymore either, last i heard.

[DrPop]

Darn. OK, that kind of sums up what I was getting from that article, then - not exactly a cruncher's dream chip like was hoped. I soooo wish AMD could fix the flaws in Bulldozer and then we'd have some competition again!
Hmmm...makes me wonder how far into the next generation of AMD CPUs ASUS will support the new mobo I got in Cheetah?

[c303a]

I think I will just stay with the I2600K that I have now. I have it running on a Asus P8Z68-V/GEN3 motherboard and with the Asus Suite overclock program it is running at 4430.0 and the temp is running at 62 degrees. That is at full load for all the cores plus Freehal, Wuprop and GPUGrid. It should start on Poem again shortly and it just keeps crunching.

[John P. Myers]

Darn. OK, that kind of sums up what I was getting from that article, then - not exactly a cruncher's dream chip like was hoped. I soooo wish AMD could fix the flaws in Bulldozer and then we'd have some competition again!
Hmmm...makes me wonder how far into the next generation of AMD CPUs ASUS will support the new mobo I got in Cheetah?

The new Piledriver CPUs are expected to be 10-15% faster than bulldozer and still compatible with AM3+ sockets

[John P. Myers]

oooooh looky, preorders

http://www.cpu-world.com/news_2012/2...iver_CPUs.html

[John P. Myers]

AMD is coming out with the Piledriver CPUs shortly. Can get an 8-core CPU for ~$260 (FX-8350). Sure it might *still* be 32nm, still only have 8MB L3 cache, still be 125W TDP, but for our purposes, it will be faster (not per core, but overall). October 23rd. 4GHz base clock speed.

Upon further thought, maybe a bit of caution is advised. It is still AMD after all lol

At the high end, best bet, Piledriver could be as much as 25% faster than the current fastest Bulldozer. 15% per clock increase + 10% higher clock. But would a 25% faster Bulldozer be faster than Haswell's version of a 2600K? Hmmm...

Of course didn't they say IB would be 10-15% faster than SB too? And didn't it turn out to actually be 3-7%? lol

So i suppose i'll end this part of the discussion with 3 famous words:

I don't know.

[DrPop]

OK...we take the wait and see approach then. Bottom line for me is that I see CPU is literally tapped out. GPU is it. But then, the problem becomes, (like in POEM for example), the CPU % utilization is way high even just feeding the GPUs!
Ahhhh...someday we're going to need 16 core monster CPUs just to keep spooning the GPUs WUs! haha