Home » Code » 8-Core Haswell 3Q14

8-Core Haswell 3Q14

Kevin Parrish, tom’s Hardware, Intel 8-core Haswell-E Slated For 3Q 2014, here, So competitive Black Scholes will be 450 to 500 million evaluations per second in 2014 on a single chip with code using 1000 long vectors, based on our previous estimate/post. But the chip cost goes from $218 to $1000. If the vector length can only be scalar or 10 (rather than 1000) drop the performance estimate by a factor of ten and four, respectively. Look at the Intel VML  Cycles versus Vector length performance graphs for CdfNorm on the i5-4670T, here, for example. Call it ~35 million evaluations per second scalar (inline with Dr. Johnson’s No Simpsons code benchmark) and 120+ million evaluations per second with vectors of length 10. There should be two performance improving effects in play. The instruction pipeline through to FMA execution units should drive the average Flops cycles time toward 0.25 cycles per flop. In the best case, from the pure pipeline perspective, you get a two double precision Adds and two Multiplies executed on each clock cycle. Then the second performance improving effect is that each register can have 8 doubles loaded for SIMD execution.  So theoretically, every 0.25 cycles you get 8 flops. On the other hand you don’t really even need to think that hard to get this result. You just run the Intel code, it’s on the internet, through ICC/MKL (probably have to buy it, although there is a 30 day free trial program) and it just falls right out of your new 3Q14 8-core Haswell desktop.

Unnamed sources in the upstream supply chain have informed DigiTimes that Intel is on course to deliver eight-core desktop processors, called the Haswell-E series, in the third quarter of 2014.

According to the sources, the Haswell-E processors will be sold at a hefty pricetag of around $1000, and will succeed the existing Ivy Bridge-E series. The company will also release the X99, the company’s next-generation high-end chipsets, to pair with the new processors.

Advertisements

2 Comments

  1. Mcd Dman says:

    so you’re ok with running “desktop” processors in a dynarack? thought it would be server grade (xeon) only in this world. maybe not.

  2. E.L. Wisty says:

    I’m thinking that depending on the FinQuant code there may be circumstances where you don’t wait for the servers and the server chips to ship and instead just use the fastest chipsets you can get into Dynarack. Maybe Phi and AVX-512 changes how to think about Dynarack.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: