Processors have got terribly boring all of a sudden
 

AMD is dead in the high-performance single-processor space.

From Intel you can get a cheap IVB or an expensive SNB-E; they will surely show off Haswell at IDF and some review site will get hold of one at some point between IDF and Christmas. It's not clear that there's anything IVB-E-like in the pipeline - it would be nice to have an eight-core i7/4960X, but I don't see that Intel has a reason to release one.

[URL]http://www.tilera.com/products/processors[/URL]
I don`t know much about Tilera but it sounds good.

Fivemack - I hear you buddy. High five.

Sorry I don't have the source - but I'm sure I read AMD is pulling out of the pro-sumer (high end consumer) segment.

As for IVB-E: I have seen reference for it. But don't expect it until 2013 or pre-haswell.

What's interesting is performance per watt. I read today intel's NUC pricing and more specs:

Here is a good forum post, with all the links. Also there's an AMD variant as well.

[url]http://forums.whirlpool.net.au/archive/1930023[/url]

By my rough calcs, a few of these would be better than i7-3930Ks for performance per cost metrics. Both initial and ongoing. Considering i7-3930ks and others are susceptible to memory saturation.

-- Craig

[QUOTE=Andi_HB;307901][URL]http://www.tilera.com/products/processors[/URL]
I don`t know much about Tilera but it sounds good.[/QUOTE]

It sounds completely irrelevant to me until I can get it on a motherboard from one of the usual Taiwanese suspects; it's a good deal further out than the Ubiquitous Lots-Of-ARMs Server, and even that's still far enough away that I doubt I'll be replacing my 48-core Opteron with one when it becomes uncompetitive to keep giving it its 600W.

If I want massive-parallel I know how to spell 680 or 7970, but I don't know how to program them.

I'm not really interested in tiny little boxes; I've got a workshop to keep the equipment in, I'm mounting everything over NFS so gigabit-ethernet is non-optional, and the NUCs look expensive per hyperthread. I'd be buying an i7/4930 for the quad-channel memory interface, and little boxes without ethernet really aren't a substitute.

Yeah, things have become boring on the high perf CPU front. I agree with you that it's mostly due to AMD not being competitive on that segment any more.

The last thing that got me slightly excited is Intel Knight Corner, but it's rather because I would like to demonstrate that the ring bus is a bottleneck for tasks that are not specifically targetting that HW thus making x86 magic coherency completely irrelevant and even counter-productive.

But yes as a single beefy system, I'm keen for oct-core + as many DDR4 channels as will fit.

On the without ethernet option - you can get usb to ethernet adapters. Rather cheap too. I couldn't see any gigE adapters though.

Yes my experience with NFS is that it's highly drop sensitive, so wireless isn't a good match.

Where I think they excel is distributed workloads needing a high memory-to-Ghz ratio.

Rough figures
i7-3930k 6cores@4.2 GHz with quad 1600Mhz ram (this is what I have)

6.4:25.2 = 0.254 (memory:cpu Ghz, mem=4x1.6, cpu=6x4.2)

NUC
i3-3217U 2core@1.8GHz with dual channel 1333Mhz ram

2.67:3.6 = 0.742 (memory:cpu Ghz, mem=2x1.333, cpu=2x1.8)

The top system cost didn't get much change out of $1500, the bottom system estimated cost is $400.

The i7-3930k chip alone is 165W@4.2GHz. Let alone full system use.

[url]http://www.tomshardware.com/reviews/core-i7-3930k-3820-test-benchmark,3090-2.html[/url]

I'd be surprised if the NUC system goes above 40W. (say 17W cpu, 5W mboard, 5W ram, 2W msata)

Top system maxes out at 4x P-1 instances due to memory saturation. The bottom one by all accounts, will handle 2x tests easy.

(btw all prices are local Oz)

If NUC doesn't scream P-1 tests, I don't know what does :)

The gotcha, is some NUC form factors may include the celeron cpu, this isn't listed as having AVX on intel's site. (It's not listed as non supported either)

[url]http://www.tomshardware.com/reviews/chromebox-chrome-os-review,3251-2.html[/url]

-- Craig

It seems the newer consumer processors are not keeping up with Moore's law, which states that the average number of transistors in a chip doubles every 18 to 24 months. There is a strong correlation between transistor count and processing power, and let's say that computing power should have doubled six times within the last decade. This means today's processors must be about 2[sup]6[/sup] = 64 times faster than those of 2002.

Now let's take a look at the numbers. The fastest processor available at this time of the year in 2002 was the 2.8 GHz Northwood Pentium 4. A [url=http://www.overclockers.com/forums/showthread.php?t=538490]post[/url] on another forum indicates that it is a little over 5 GFLOPS, which sounds about right. In comparison, the Core i7-3960x, the fastest processor currently available, clocks in at 130 GFLOPS at factory settings. The speed increase is only by a factor of ~25, not the ~64 we should be expecting.

The fastest processor currently available is the Radeon 7970, which offers a peak of 1 teraflop DP; and indeed it achieves this by incorporating 512 double-precision ALUs on a die with 4.3 billion 28nm transistors, which is about the Moore's-law prediction.

The transistor counts on CPUs have also been increasing pretty much to schedule, though a lot of them end up as cache; Northwood has 55 million 130nm transistors, a 3960X has 2270 million 32nm transistors on a significantly larger die, IVB has 1400 million 22nm transistors on a die about the same size as Northwood.

[QUOTE=fivemack;308055]The transistor counts on CPUs have also been increasing pretty much to schedule, though a lot of them end up as cache; Northwood has 55 million 130nm transistors, a 3960X has 2270 million 32nm transistors on a significantly larger die, IVB has 1400 million 22nm transistors on a die about the same size as Northwood.[/QUOTE]

And it should be pointed out that "Moore's law" is really only a prediction on the number of transistors per CPU -- an uncannily accurate one -- which as you point out does not always lineally translate into processing speed.

And let's be honest -- when many of us wear computers on our hips which are more powerful than what used to be called super computers, just how much more computing power does the average person really need?

We now have GUIs which have transparent windows, menus which fade in and out, backgrounds which are streamed video or generated fractals...

Does any of this actually make anyone more productive?

[QUOTE=chalsall;308069]

We now have GUIs which have transparent windows, menus which fade in and out, backgrounds which are streamed video or generated fractals...

Does any of this actually make anyone more productive?[/QUOTE]

Me wants candy! ouch, my eye!

@kracker

:goodposting: My thoughts were similar. While many machines accomplish all sorts of work, they are also toys on the order of hot rod cars.

However, to address Chris's question, I have the feeling that more procession power has diminishing returns. My partner was disappointed when he got an i7-920 because it didn't seem that much faster than the previous box. I put a RAID 0 on it which helped some with that perception. If they had been affordable then I would have put in an SSD. Perhaps now an SSD cache would be the answer.

Fact is though, he does not do a lot of stuff which makes the CPU break a sweat, like rendering HD video. This may change if he gets more into the Reason synthesis and recording package.