Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

60 Cores

Anything under $3,000 is really, REALLY cheap for that number of cores compared to how much the Xeons cost.


I wonder what the performance is like compared to the Xeons?
 
Tucom said:
Anything under $3,000 is really, REALLY cheap for that number of cores compared to how much the Xeons cost.
Click to expand...
If you say so... :p


I wonder what the performance is like compared to the Xeons?
Click to expand...

some independent benchmarks:

Untitled11.jpg

Intel’s Xeon Phi SE10P (red) beat Nvidia’s Tesla C2050 and K20 GPUs (light and dark green, respectively) in 18 out of 22 tests. The Xeon Phi also beat dual Xeon X5680s (each with six cores for 12 cores total, light blue) and dual Xeon E5-2670s (each with eight cores for 16 total, dark blue) in 15 out of 22 tests. Source: Ohio State.
​

For the test, they chose the parallel processing operations routinely performed on large sparse matrices. Variously called eigensolvers, linear solvers and graph-mining algorithms, these applications encode vast parallelism into wide-dense vectors multiplied by the large sparse matrices.

Something I also find rather interesting is that Intel’s Xeon Phi runs the same x86 instruction set as a 64-bit Pentium. :)
 
Last edited:
Tesselator said:
If you say so... :p




some independent benchmarks:

Image
Intel’s Xeon Phi SE10P (red) beat Nvidia’s Tesla C2050 and K20 GPUs (light and dark green, respectively) in 18 out of 22 tests. The Xeon Phi also beat dual Xeon X5680s (each with six cores for 12 cores total, light blue) and dual Xeon E5-2670s (each with eight cores for 16 total, dark blue) in 15 out of 22 tests. Source: Ohio State.
​

For the test, they chose the parallel processing operations routinely performed on large sparse matrices. Variously called eigensolvers, linear solvers and graph-mining algorithms, these applications encode vast parallelism into wide-dense vectors multiplied by the large sparse matrices.

Something I also find rather interesting is that Intel’s Xeon Phi runs the same x86 instruction set as a 64-bit Pentium. :)
Click to expand...

damn impressed,
 
Impressive.....

I know. Also impressive is the price tag. I wonder how much will cost a MOBO with the 60 cores.....

:):apple:
 
phoenixsan said:
I know. Also impressive is the price tag. I wonder how much will cost a MOBO with the 60 cores.....

:):apple:
Click to expand...

The Phi is a co-processing card, like the nVidia GPU computing cards, but using x86 instructions and a smaller number of superior cores.
 
Tesselator said:
If you say so... :p
Click to expand...



If I say so? I'm sure you're more than capable of figuring out that E5 Xeons cost more than 4 to 5 grand for the higher end configurations. Given this has 60 cores, and costs much less, it's quite interestingly priced.
 
Tucom said:
If I say so? I'm sure you're more than capable of figuring out that E5 Xeons cost more than 4 to 5 grand for the higher end configurations. Given this has 60 cores, and costs much less, it's quite interestingly priced.
Click to expand...

Sure sure. Maybe reread the original sentence: "Of course the drawback for me is the $2,700.00 price tag."

If almost $3k for a co-processing card is cheap to you then let's be friends... Maybe you could help me finance the new Maserati I've got my eye on. ;)
 
Tesselator said:
If almost $3k for a co-processing card is cheap to you then let's be friends... Maybe you could help me finance the new Maserati I've got my eye on. ;)
Click to expand...

$3,000 for a co-processing card, like a few grand for a workstation, is a decent price from a business perspective. It's not like nVidia's competitors are cheap.
 
FluJunkie said:
$3,000 for a co-processing card, like a few grand for a workstation, is a decent price from a business perspective. It's not like nVidia's competitors are cheap.
Click to expand...

Absolutely true! Please send your contribution to my Maserati via PayPal to: Tesselator69 @ gmail . com

:)
 
This is one of those situations where this old axiom hold true "if you have to ask the price you can't afford it"
 
What difference does it make that it is superior in "sparse matrix" calculations for the vast majority of Mac users.
 
tamvly said:
What difference does it make that it is superior in "sparse matrix" calculations for the vast majority of Mac users.
Click to expand...

None.

It's mainly for scientific calculations, custom code, and so on. Almost nothing you commonly use can take advantage of this card AFAIK.
 
MyMac1976 said:
Not everything needs to be for all of us or none of us.
Click to expand...

I'm sure there are folks who could benefit from a supercomputer on their desk. It wouldn't affect me or my use of my MP. There are precious few apps that can take advantage of 12 cores, let alone 60.

I'm more interested in developments in I/O performance than pie-in-the-sky highly parallel processor advances.
 
Shame on all of you for forgetting that this thing could handbrake a blu-ray in 12 seconds. :eek:


note: 12 seconds is an estimate made up out of nowhere.
 
tamvly said:
I'm sure there are folks who could benefit from a supercomputer on their desk. It wouldn't affect me or my use of my MP. There are precious few apps that can take advantage of 12 cores, let alone 60.

I'm more interested in developments in I/O performance than pie-in-the-sky highly parallel processor advances.
Click to expand...

Well, with the recent developments of USB3, two versions of Thunderbolt, and SATA III all making the scene almost at once you must be pretty happy!

About 2 or 3 years ago I could have gone for and made use of, a card like that - if it were under $1k. :p These days I'm playing more games and working a lot less - such as it is for retired persons. :)

----------
ToomeyND said:
Shame on all of you for forgetting that this thing could handbrake a blu-ray in 12 seconds. :eek:


note: 12 seconds is an estimate made up out of nowhere.
Click to expand...

Handbreak works with this card? No right?
 
Tesselator said:
Well, with the recent developments of USB3, two versions of Thunderbolt, and SATA III all making the scene almost at once you must be pretty happy!
Click to expand...

When USB3, SATA III and TB show up on a MP, I'll be happy.
 
Tesselator said:
Something I also find rather interesting is that Intel’s Xeon Phi runs the same x86 instruction set as a 64-bit Pentium. :)
Click to expand...

Actually it doesn't run the same subset

"... with Intel confirming that it is indeed using an enhanced Pentium 1 (P54C) core with the addition of vector and FP64 hardware. Intel has also confirmed that Xeon Phi will offer 512-bit SIMD operations ... "
http://www.anandtech.com/show/6017/intel-announces-xeon-phi-family-of-coprocessors-mic-goes-retail

Without the vector and FP64 hardware it wouldn't turn in those numbers. If you used an old compiler from the P54C days it would never turn in these kinds of numbers.


The Pentium basis shouldn't be surprising at all. The rationale was laid out back when one of its ancestors was the precursor to the Larrabee project. The really old Pentium subset was chosen because it allowed for smaller core implementation. The process being used on these is the same 22nm that is used on Ivy Bridge and Haswell. If the x86 cores were the same size (i.e., complexity) as the mainstream x86 cores they couldn't fit 60 on a reasonable sized die.

So they chuck features ( fancy Out of Order, fancy branch prediction , etc. ), but do make the trade off to add the modern vector implementation. Where you force Phi into lots of branching, scalar code is exactly where these benchmarks fail to do better than the Xeon 2670s.
 
Last edited:
ScottishCaptain said:
Nobody has been interested enough in it to port the required drivers over.
Click to expand...

You'd think Apple might be motivated. They could sell some more Mac Pros to host these things. Especially since Phi cards aren't sold as "retail cards". They are suppose to be sold through system integrator (which would be the role that Apple would be playing).

Pragmatically, though it is more than drivers. Need the higher end Intel Parallel Compile/Debug tool chain too ( which is currently missing for OS X regardless of leveraging Phi or not. ) if want to build apps that operate on both sides of host OS and Linux on the Phi card.

Honestly, both issues would mean that Apple would have to do some work. I have doubts that would pass their Scrooge McDuck new project approval process.
 
have to wonder if a software/hardware manufacturer like Blackmagic would figure out a solution to get something like this to work with their applications? seems like it could function like/better than a RED rocket card if the software could work with it. the video market would swallow this thing up if it could work, especially if it could work with existing hardware. easily could be the best upgrade one could make to their system at (maybe) a similar price to a Kona3 card.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back