Broadwell Processor

[firejuggler]

Tomshardware offer a first look at it, not performance wise but structural.
http://www.tomshardware.com/reviews/...re-m,3904.html

[retina]

Something that could improve P95 speed:

The floating point multiplier is much more efficient, now able to accomplish in three clock cycles what takes Haswell five cycles to complete.

Unfortunately the "next" page button uses JS to find the next page (because html doesn't support links as we all know) so I have no idea what more there might be of value.

[firejuggler]

Second page is about the mobile version :
http://www.tomshardware.com/reviews/...-m,3904-2.html

[retina]

Originally Posted by firejuggler

Second page is about the mobile version :
http://www.tomshardware.com/reviews/...-m,3904-2.html

Nothing there that I can see that would help P95 unless one is concerned about power usage and saving 5 cents per month.

[VictordeHolland]

Rumors:
Haswell-E 2014-Q3 (september)
Broadwell-Y 2014-Q4 (retail devices in december)
Broadwell (laptops) 2015-Q1
Broadwell (desktop/unlocked) summer 2015
Broadwell-E end of 2015/ start of 2016

Hopefully HW-E won't be overly expensive, but I expect the DDR4 to add significantly to the cost of the platform.

[ewmayer]

Originally Posted by retina

Something that could improve P95 speed:

As already discussed a few posts back.

Unfortunately the "next" page button uses JS to find the next page (because html doesn't support links as we all know) so I have no idea what more there might be of value.

Yeah, I was annoyed by that too, but clicking on the Broadwell-Y link at top of the original page took me there. One further annoyance there, though - I hate when sites like this mix up their comparison metrics:

We do have some specifics when it comes to the die and package sizes though. The Broadwell-Y chip is 82mm2, scaled down about 63% compared to Haswell-Y's 130mm2 die size.

63% "die size" reduction - very impressive. But their first set of numbers (82 vs 130 mm) is edge length, whereas the 63% is based on area reduction. So, impressive as the latter figure sounds, doing the quick math on the former pair shows that 82mm is negligibly (< 1%) smaller than the "die size reduction under process-size shrink with no transistor changes whatsoever" number, 130mm*(14nm/22nm).

[only_human]

This article doesn't have much more info on the 14nm process but is still nice: Intel’s 14nm Technology in Detail (AnandTech)

Moving on to the specifications and capabilities of their 14nm process, Intel has provided the minimum feature size data for 3 critical feature size measurements: transistor fin pitch, transistor gate pitch, and the interconnect pitch. From 22nm to 14nm these features have been reduced in size by between 22% and 35%, which is consistent with the (very roughly) 30%-35% reduction in feature size that one would expect from a full node shrink.

[ATH]

Since Broadwell is delayed and they claim they are not moving Skylake back, many will probably skip Broadwell since Skylake should be out not long after, maybe before Broadwell-E.

http://www.pcworld.com/article/26833...o-skylake.html

[ewmayer]

By way of an upcoming-birthday-present-to-self, we are considering purchasing a Broadwell NUC. Here are the main considerations in favor of the scheme:

o I get to build and run code on a Broadwell, and - epecially interesting for LL-style work - see if the 3-cycle FMA latency (vs 5 for Haswell) makes a difference. (We have dual versions of certain key FFT macros, one of which replaces all the ADD/SUB with FMA - this is slightly slower on Haswell, but may better on Broadwell);

o We get a convenient separate sandbox in which we can get the latest Debian Linux and GCC compiler without risking breaking an existing install (my Haswell has many dependencies, in particular getting the nVidia tools installed and working was nontrivial and limits which GCC version we can use). Being able to use GCC 4.9+ will also allow us to build code with AVX512 inline assembler, although the resulting binaries will not run on Broadwell, obviously;

o Tiny form factor, I can use it as a coffee mug warmer on my desk;

o Mike has kindly offered to build it for me, since he enjoys doing that sort of thing, and the sales tax savings I get from shipping to him in sales-tax-free AR versus to me in CA will cover his cost to ship the assembled unit to me, plus leave some beer money for him.


We have also considered the low-end-Broadwell-based-Ultrabook route, but don't really need yet another laptop PC. The stark minimalism of a headless NUC appeals to our ascetic nature.

Xyzzy recommends something like this combo deal from Newegg, which costs only a smidge more than a separate-pieces bundle with less RAM (2 x 2GB vs 2 x 4GB) and a smaller SSD (80 GB vs 120). We could use either a simple LAN cable (the way we plug into our headless, wifi-less ATX-cased Haswell system at home) or the builtin wifi to talk to the system.

Comments, suggestions before we part with our hard-earned cash forever?

[Xyzzy]

Having Ernst ask us for hardware advice is kinda dangerous because we like spending other people's money!

And our default answer for just about anything right now is "NUC".

But maybe there is a cheaper or better alternative?

We can attest to the NUC's mug-warming capabilities!

One issue to explore is the headless capabilities of the NUC. We have not tried that yet.



If headless doesn't work we can use one of these: http://www.mersenneforum.org/showthread.php?t=19843

A quick check of the last generation NUC: https://communities.intel.com/message/244216

[Xyzzy]

This slightly taller model would allow the use and swapping (offline) of 2.5" form-factor SSDs:

http://www.intel.com/content/www/us/...nuc5i3ryh.html

Edit: Oops! This is the one in the combo Ernst linked!