Intel ships 60 Core Supercomputing chip

[Hammer_Time]

http://news.cnet.com/8301-1001_3-575487 ... -it-first/

Intel's 60-core chip ships; elites like Hawking get it first

Intel's first chip to integrate 60 processor cores is for high-performance computing applications -- with a high-end price to match.

by Brooke Crothers November 12, 2012 8:20 PM PST

Intel's first 60-core chip is here, but only elite institutions like Cambridge University will get early access.

The chip, generally called Phi, isn't your typical Intel processor. It's less like a central processing unit (CPU) -- which is the chipmaker's bread and butter -- and more akin to a graphics processing unit (GPU) that works in conjunction with the CPU.
In fact, Intel tried to bring out a consumer GPU card called "Larrabee" based on the technology but canceled the project in 2009.
But the company still needs the technology to compete with Nvidia in the supercomputing market because specialized GPU-like processor cores can handle certain supercomputing tasks more efficiently, reducing the prodigious power requirement for supercomputers as well as increasing performance.
Expected to be commercially available early next year, the Xeon Phi coprocessor 3100 family will offer more than 1,000 gigaflops (1 teraflops) double-precision performance -- a rating for chips that do scientific calculations.
By comparison, Intel's consumer processors don't offer anything close to that level of performance when doing specialized computing tasks.
Professor Stephen Hawking and the Cosmos Lab at the University of Cambridge have been given early access to the technology for use in their SGI supercomputer, Intel said.
"With our powerful and flexible SGI UV2000, we can continue to focus on discovery, leading worldwide efforts to advance the understanding of our universe," Hawking said in a statement.
Stateside, Intel customers such as Texas Advanced Computing Center are getting early customized products.
The first commercial versions of the coprocessor, which ship on a PCIe circuit board, are expected to be available January 28 with a recommended customer price of $2,649.
The 3100 product family will be available during the first half of 2013 with a recommended customer price below $2,000.

It is actually a "coprocessor on a PCIe card" that works in tandem with Xeon processors for supercomputing ( parallel tasks ) operations.

nVidia has already pioneered this of course...

http://www.nvidia.ca/object/personal-su ... uting.html

TESLA PERSONAL SUPERCOMPUTING WORKSTATION SOLUTIONS

NVIDIA® Tesla® 20-series GPU computing processors turn standard PCs and workstations into personal supercomputers. Based on the NVIDIA CUDA® GPU architecture codenamed "Fermi", Tesla 20-series GPUs feature more than 500 gigaflops of double precision performance, 1 teraflop of single precision performance, ECC memory error protection, and L1 and L2 caches. Tesla 20-series GPGPU processors for workstations deliver cluster level performance right at your desk.

Kepler based Tesla GPUs are now available for servers

Feature Tesla C2075
GPU Computing Applications Reservoir simulation, CAE (structural analysis), Molecular dynamics, Numerical analytics, Computational visualization (ray tracing),
Peak double precision floating point performance 515 Gflops,
Peak single precision floating point performance 1030 Gflops
Memory bandwidth (ECC off) 148 GBytes/s
Memory size (GDDR5) 6GB
CUDA cores 448

So for Peak DOUBLE precision floating point performance, the new Intel Xeon Phi Coprocessor does "over 1000 Gigaflops, twice as much as Tesla's 515 Gigaflops quoted above... in single precision the Tesla can match it, but not in Double Precision...

http://www.newegg.com/Product/Product.a ... 6814132007

NVIDIA Tesla C2075 6GB 384-bit GDDR5 PCIe 2.0 x16 1.03 Tflops Workstation Video Card

$2,269.99

The Xeon Phi costs $380 more than the Tesla, but also has twice the Double Precision flops performance...

I wonder what nVidia has up their sleeve to counter this??

[Stupify]

i was looking at the specs of K20, S10000 and Intel's Phi. Among them K20 seemed to be the best bang for performance/power solution by a decent margin, while Intel was the worst. AMD had highest TFLOPS, however, consumed almost 100W more than K20 to push out only a handful more GFLOPS. Intel, even with their better process technology and what not, just doesn't deliver as much as K20, but it still is respectable for first such product. Nvidia better watch out for future products from Intel. Intel's compiler optimization should beat out Nvidia any given day and twice on Sunday imo, AMD being no chance against either.

[Hammer_Time]

AMD has "no chance" against nVidia and Intel in the supercomputing coprocessor space?? Really?:

http://www.theregister.co.uk/2012/11/12 ... _gpu_card/

The S9000 costs $2,499 at list price while the S10000 costs $3,599, and if you do the math the S10000 delivers $6.48 per DP teraflops per watt, while a pair of S9000s will give you $6.89 per DP teraflops per watt.

Because the card is double-stuffed, it can deliver a very impressive 5.91 teraflops SP and 1.48 teraflops DP in peak floating point oomph.

Yes the S10000 card is expensive, but it does manage to beat both Intel Xeon Phi and nVidia Tesla in DP flops!!! 1.48 Teraflops in DP as stated above! That is better than the Phi's 1.0 ( or slightly more ) DP performance, and well above nVidia Teslas 0.515 Teraflops DP performance... course you have to cough up another grand for the card, and it is a beast and sucks a lot of power , but still... it is "in the game" as they say...

nVidia just released their new K20 and K20X Tesla cards, and AMD S10000 STILL beats both those cards in peak DP flops performance!! :

K20 : 1.17 Teraflops DP

K20X : 1.31 Teraflops DP

http://www.zdnet.com/nvidia-and-amd-unv ... 000007301/

Nvidia and AMD unveil new supercomputer GPUs

Summary: Nvidia's Tesla K20 and Tesla K20X, and AMD's FirePro SM10000 GPUs can each deliver more than a teraflop of double-precision performance.

By Adrian Kingsley-Hughes for Hardware 2.0 | November 13, 2012

Chipmakers Nvidia and AMD have debuted new graphics processing units (GPUs) aimed at the supercomputer market.

Nvidia's GPUs are the Tesla K20 and Tesla K20X. The Tesla K20 features 2,496 stream processors and 5GB memory, all of which provides 3.52 teraflops of single-precision and 1.17 teraflops of double-precision peak performance. The Tesla K20X has 2,688 stream processors and 6GB of GDDR5 memory offers 3.95 teraflops single-precision and 1.31 teraflops double-precision peak floating point performance.

At the heart of these cards is Nvidia's new GK110 silicon packed with 7.1 billion transistor; the biggest GPU in terms of transistor numbers. Based on the Kepler architecture, the GK110 features 64 double-precision CUDA cores, giving it eight-times the double-precision performance of a earlier GK104 SMX.

The Tesla K20X GPU delivers three times greater energy efficiency than previous-generation GPUs, and offer better efficiencies compared to CPUs.

Because of the increased double-precision floating point performance, Nvidia says that the Tesla K20-series GPUs are ideally suitable for tasks such as seismic processing, financial computing, CFD, CAE, computational chemistry and physics, data analytics, satellite imaging, and weather modeling.



The Tesla K20 and Tesla K20X GPUs are also shifting fast, with Nvidia reporting that more than 30 petaflops of GPUs have already been delivered to various supercomputer clients during the last month. That 30 petaflops of power is equivalent to the computational power of last year's 10 fastest supercomputers combined.

Tesla Nvidia K20-series GPUs are shipping today and available for order from leading server manufacturers, including Appro, Asustek Computer, Cray, Eurotech, Fujitsu, HP, IBM, Quanta Computer, SGI, Supermicro, T-Platforms and Tyan, as well as from Nvidia reseller partners.

Also out is AMD's FirePro SM10000 dual-GPU workstation graphic processor, designed to be used in virtualized environments and targeted at high-performance computers and servers.

AMD reports that the FirePro SM10000 delivered 1.48 teraflops of peak double-precision performance and features 6GB of GDDR5 memory. The FirePro SM10000 delivers up to 1.3-times the single and up to 7.8-times the double peak precision floating point performance of Nvidia's Tesla K10 GPU.

The FirePro SM10000 can be used on cloud servers to accelerate graphics on the server side and deliver full high-definition virtual desktops to client devices.

That chart above ( yes it is Daamit marketing material so take with grain of salt - they must be comparing their new card to the older Tesla K10 card of course to make that statement, but still.. ) says the new S10000 delivers highest performance per watt for peak DP flops!! Certainly it delivers the highest peak at 1.48 Teraflops DP, and we know GCN architecture is reasonably energy efficient so I do not doubt this claim that much... Yes, it sucks 100 Watts more to gain about ~ 0.2 extra Teraflops DP, but still...

AMD S10000 is also much stronger in SP flops performance, 5.91 SP flops vs only 3.95 SP flops of the nVidia K20X !! That is 33% faster!!!

Granted Intel also has advantages here, compiler design etc etc but I don' think you can totally rule out AMD here...

[Stupify]

i very much doubt their perf/watt claim as the numbers definitely support Nvidia being the top perf/watt. plus from what i recall AMD drivers/support aren't as optimal as Nvidia and so their real performance suffers a lot in comparison to their theoretical performance.

[Hammer_Time]

That could very well be true...

[Sauron_Daz]

i very much doubt their perf/watt claim as the numbers definitely support Nvidia being the top perf/watt. plus from what i recall AMD drivers/support aren't as optimal as Nvidia and so their real performance suffers a lot in comparison to their theoretical performance.

Don't know about that, but I do like AMD's consumer cards.
And I used to favor NVidia before..

[DIREWOLF75]

i very much doubt their perf/watt claim as the numbers definitely support Nvidia being the top perf/watt. plus from what i recall AMD drivers/support aren't as optimal as Nvidia and so their real performance suffers a lot in comparison to their theoretical performance.

Don't know about that, but I do like AMD's consumer cards.
And I used to favor NVidia before..

Same here.

[Hammer_Time]

i was looking at the specs of K20, S10000 and Intel's Phi. Among them K20 seemed to be the best bang for performance/power solution by a decent margin, while Intel was the worst. AMD had highest TFLOPS, however, consumed almost 100W more than K20 to push out only a handful more GFLOPS. Intel, even with their better process technology and what not, just doesn't deliver as much as K20, but it still is respectable for first such product. Nvidia better watch out for future products from Intel. Intel's compiler optimization should beat out Nvidia any given day and twice on Sunday imo, AMD being no chance against either.

For Shame Stupe, For Shame!! :

http://www.xbitlabs.com/news/other/disp ... puter.html

AMD FirePro S10000 Card Powers World’s No. 2 Greenest Supercomputer.

AMD’s FirePro S10000 Accelerator Speeds Up Saudi Arabian Supercomputer

[11/15/2012 01:22 PM]

by Anton Shilov

Advanced Micro Devices said Thursday that its recently introduced AMD FirePro S10000 server graphics card powers Sanam supercomputer in Saudi Arabia. While the system in not in the Top 10 most powerful machines in the world, Sanam is the world’s second most power-efficient supercomputer.

"There is a massive shift in the HPC industry from defining supercomputing performance in terms of FLOPS to performance-per-watt and developing environmentally responsible technologies," said John Gustafson, senior fellow and chief product architect at graphics business unit at AMD.

Based on 4.8 thousand of Intel Xeon E5-2650 (8 cores, 2.0GHz) microprocessors and 420 AMD FirePro S10000 accelerators, the Sanam delivers 421.2TFLOPS Rmax Linpack performance or 1098TFLOPS Rpeak Linpack performance. Capable of providing a system energy efficiency of over 2.3GFLOPS per watt and performing 2351 million calculations per second per watt (2351MFLOPS/W), the Sanam supercomputer is the world’s second most energy-efficient supercomputer, according to Green500 list. Ironically, only Beacon, an Intel Xeon- and Intel Xeon Phi-based supercomputer is more efficient with 2499MFLOPS/W; even the Titan, the world's highest-performing supercomputer with AMD Opteron chips and Nvidia Tesla K20 accelerators inside, is at 2142MFLOPS/w. Sanam is currently the world’s No. 52 most powerful supercomputer.

“The Sanam supercomputer, equipped with AMD FirePro S10000 GPUs based on our Graphics Core Next architecture, consumes a mere 180kW power while running, which is incredibly low for a supercomputer. With solutions like the FirePro S10000 along with innovative software from AMD and technology partners […] we enable institutions engaged in cutting-edge research […] to achieve massive compute performance and processor density, while staying within their power budget,” added Mr. Gustafson.

Sanam is a joint project of Frankfurt Institute for Advanced Studies and King Abdulaziz City for Science and Technology (KACST) in Riyadh, Saudi Arabia. As an integral part of the high-speed cluster (HSC) project, Adtech Global provided technical research to identify a systems solution that would not only work for the high-speed cluster HSC application, but could also provide long-term development and modifications opportunities.

“We focused on utilizing technologies which would provide the ability to continuously improve performance, lower costs and reduce power consumption in the future. We knew at the outset that the Sanam supercomputer we worked on with University of Frankfurt and AMD was going to be special, but knowing is one thing and seeing is another. We have worked on many high-end, custom solutions; however, it is always exciting to flip the switch and witness the incredible performance of today's supercomputers. The performance-per-watt capability of this system is simply amazing,” said Rick Rozalsky, vice president of global business development at Adtech Global.

Sorry Stupe, just had to rub this in a wee bit bro! No it is not the most powerful, but it is super-efficient!!!

Since you happen to live in Toronto, the birthplace and HQ of ATi, I think you better take your statement back before hordes of angry Daamit engineers and staff descend upon you and smash you up with video card prototypes!!





Excuse me, they renamed it the "AMD Campus" back in 2006 when they bought ATi...

http://en.wikipedia.org/wiki/Advanced_Micro_Devices



Here is a pic of an angry ATi/AMD engineer , expect many more of these unless you change your tune about AMD workstation cards!!

[Sauron_Daz]

Only 420 FirePro's?

[Hammer_Time]

4800 Xeon cpu's and yes, just 420 FirePro S10000 cards ( used as coprocessors of course ) :

Based on 4.8 thousand of Intel Xeon E5-2650 (8 cores, 2.0GHz) microprocessors and 420 AMD FirePro S10000 accelerators

Funny way of saying 4800 = "4.8 thousand"

[DIREWOLF75]

4800 Xeon cpu's and yes, just 420 FirePro S10000 cards ( used as coprocessors of course ) :

Based on 4.8 thousand of Intel Xeon E5-2650 (8 cores, 2.0GHz) microprocessors and 420 AMD FirePro S10000 accelerators

Funny way of saying 4800 = "4.8 thousand"

Translation issues most likely. I was wondering if they meant 4800 cpu or 4800 cores though. Strongly leaning towards the latter. That would mean 420 GPGPU linked with 600 cpu, much more reasonable.

[Hammer_Time]

Yes, I believe you are right, that makes much more sense... 8 cores per Xeon E5-2650 cpu, so 4800 total cores divided by 8 = 600 Xeon cpus as you surmised... with 420 coprocessor FirePro cards...

[Stupify]

H_T that's not apples to apples comparison, now is it? I was comparing the three GPGPU performance/watt and what you have posted is about the entire system which obviously would be better than the standard pure CPUs only and hence would have no trouble beating those out in terms of perf/watt. Setup 3 systems with these solutions while rest being the same and I bet it would be the worst of the bunch, theoretically and even worse in reality.

[Sauron_Daz]

4800 Xeon cpu's and yes, just 420 FirePro S10000 cards ( used as coprocessors of course ) :

Based on 4.8 thousand of Intel Xeon E5-2650 (8 cores, 2.0GHz) microprocessors and 420 AMD FirePro S10000 accelerators

Funny way of saying 4800 = "4.8 thousand"

Translation issues most likely. I was wondering if they meant 4800 cpu or 4800 cores though. Strongly leaning towards the latter. That would mean 420 GPGPU linked with 600 cpu, much more reasonable.

And that's using up 180 kW, so it almost can't be 4800 CPU's.

[Sauron_Daz]

4800 Xeon cpu's and yes, just 420 FirePro S10000 cards ( used as coprocessors of course ) :

Based on 4.8 thousand of Intel Xeon E5-2650 (8 cores, 2.0GHz) microprocessors and 420 AMD FirePro S10000 accelerators

Funny way of saying 4800 = "4.8 thousand"

Quite normal here. You use the comma as seperator which is funny/confusing for us.

[DIREWOLF75]

4800 Xeon cpu's and yes, just 420 FirePro S10000 cards ( used as coprocessors of course ) :

Based on 4.8 thousand of Intel Xeon E5-2650 (8 cores, 2.0GHz) microprocessors and 420 AMD FirePro S10000 accelerators

Funny way of saying 4800 = "4.8 thousand"

Quite normal here. You use the comma as seperator which is funny/confusing for us.

Those crazy transatlantics... Soon they´re gonna try and pay for things with farthings, you know it!

[Hammer_Time]

H_T that's not apples to apples comparison, now is it? I was comparing the three GPGPU performance/watt and what you have posted is about the entire system which obviously would be better than the standard pure CPUs only and hence would have no trouble beating those out in terms of perf/watt. Setup 3 systems with these solutions while rest being the same and I bet it would be the worst of the bunch, theoretically and even worse in reality.

True, but it is still the #2 most efficient supercomputer on the planet right now, and Adtech Global ( makers of this SANAM supercomputer in Saudi Arabia ) could have went with the Intel Phi or nVidia Tesla type of solution, but they chose AMD FirePro... they must have had a good reason to do so...

I suppose we need to see more of these GPGPU style of solutions out in the real world before we can really state whose solution is "the best" or, "the most efficient", or "best performance per watt" etc etc.

My point is that at least AMD is still in the ballgame here with their latest FirePro cards. Mebbe not the "best" but certainly still in the ballgame...

[Hammer_Time]

4800 Xeon cpu's and yes, just 420 FirePro S10000 cards ( used as coprocessors of course ) :

Based on 4.8 thousand of Intel Xeon E5-2650 (8 cores, 2.0GHz) microprocessors and 420 AMD FirePro S10000 accelerators

Funny way of saying 4800 = "4.8 thousand"

Quite normal here. You use the comma as seperator which is funny/confusing for us.

Please send me 1.00009,000.3.14,159 dollars !!

[Sauron_Daz]

Based on 4.8 thousand of Intel Xeon E5-2650 (8 cores, 2.0GHz) microprocessors and 420 AMD FirePro S10000 accelerators

Funny way of saying 4800 = "4.8 thousand"

Quite normal here. You use the comma as seperator which is funny/confusing for us.

Those crazy transatlantics... Soon they´re gonna try and pay for things with farthings, you know it!

[Sauron_Daz]

Funny way of saying 4800 = "4.8 thousand"

Please send me 1.00009,000.3.14,159 dollars !!

How much is that is Euro's? Note that we don't use cent and two cent pieces here.