IDF: Intel unveils its Haswell Xeon E5-2600 V3 server processor ( 18 cores+)

theinquirer.net ^ | Mon Sep 08 2014, 17:30 | Lee Bell

[Ernest_at_the_Beach]

Features up to 18 cores and DDR4 memory support

SAN FRANCISCO: INTEL HAS ANNOUNCED its next generation Xeon chip for servers and workstations at the Intel Developer Forum (IDF) in California today, the Xeon E5-2600 V3 family of processors, based on Haswell microarchitecture.

Featuring up to 18 cores, the C612 chipset Xeon E5-2600 V3 CPU, previously codenamed Haswell-EP, offers improved performance in an R3 socket when compared to its predecessor, the Xeon E5-2600 V2, which was based on Ivy Bridge architecture.

The higher core count brings significant performance improvements over the previous generatons, Intel said, as well as improved peripheral expansion.

[GeronL]

They should have named it the HAL 9000

[Ernest_at_the_Beach]

ping!

[TangoLimaSierra]

They should have named it the HAL 9000

Or Skynet.

[ImJustAnotherOkie]

Any opinions on diminishing returns of core counts?

[ImJustAnotherOkie]

"From a performance perspective we are delivering worldwide perfoamnce levels, generation from generation, tripling performance [thanks to] the 18 cores [that]

Don't need no stinkin' parity check...

[deoetdoctrinae]

[Crusher138]

Will still play Quake at only 20 fps...

[Disambiguator]

"From a performance perspective we are delivering worldwide perfoamnce levels, generation from generation, tripling performance [thanks to] the 18 cores [that]

Fo' shizzle.

BTW, I've seen business cards that said "When Preformance Counts."

"Preformance? Isn't that some kind of preversion?"

[ImJustAnotherOkie]

I thought I told you not to call me here!!!

[factoryrat]

I won't be impressed until they come out with 64 core processors, one for each bit on the data buss, along with 64 bit direct memory accessing, and RISC for massive throughput. Save the GUI for all of the slow stuff on the market now, like all of the current video memory hog uP’s on the market.

BTW, I cut my teeth writing code for 6805, and 6811 uC’s. In assembly language. You would be amazed at what you can run with a program on a single 64k eprom. Fun times.

[Usagi_yo]

But where does the next performance increase come from? We’re still in the 20th century bus structure paradigm and these multi-core processors can easily (and do) swamp the I/O and memory channels and parallel processing is a very specific and precise discipline for select programmers.

The last great performance increase came from SSD drives.

[al baby]

Loved ol Kennan Wynn

[SeeSharp]

Any opinions on diminishing returns of core counts?

It all depends on the efficiency of the locks. I can remember Digital Equipment telling its customers that it was impossible to have more than 4 CPU's because the spin locks for each CPU added about 25% overhead. Of course that was a purely software based synchronization method.

Eighteen cores is a very impressive achievement. Now if only we had a way to teach programmers how to write thread-safe code. Dining philosophers anyone?

[SeeSharp]

All computers wait at the same speed.

[Zathras]

This maybe the last GREAT server part from Intel.
They ripped the guts out of the design community to get this out on time.
Many designers and management decided to retire than go thru this again.

Jobs and careers were crushed by this chip.

[Sparticus]

Any opinions on diminishing returns of core counts?

It depends entirely on the type of workload that is presented to the machine. If you're just checking email and surfing the web while listing to music, then you're not going to see any benefit from having more than a couple of cores. If the machine is a web or database or other highly parallelizable task server (like tracking the locating a million cell phones), then more cores help quite a lot. That is assuming that the other systems in the server can efficiently provide data to all those cores without there being a bottle neck somewhere.

[Dalberg-Acton]

Server farms running lots of VMs.

[daniel1212]

This is a one year old PC, with a ASRock 970 EXTREME3 mobo; AMD 6350 3.9ghz 6 core 64bit cpu; 8gb ram, replacing (thank God) a Sony vio with a 2.8ghz cpu and 4gb ram, both running W.8, and without prepackaged bloat, and tweaked somewhat for more speed (no fancy graphics, etc.).

And while the former is faster and can handle much more (like over 100 tabs in Firefox, and a dozen word docs, and other significant programs) i thought it would boot (with fast boot) in less than 20 seconds, and do basic operations much faster.

But i think Win 95 was sometimes faster opening folders, and this takes about a minute before all is loaded, yet it usually does not seem to use more than 1-6% CPU at idle, which it usually is at, except for \FlashPlayerPlugin_14_0_0_179.exe which can get it up to over 20%.

In fact i do not think i have ever seen it over 30% or 100% except when used a CPU testing software.
But why am i complaining.

I just thought it would use more horsepower launching new tasks and thus getting more speed. It does have a cheap Video Card (powerglo, Radeon HD 5400 series).

[Sparticus]

Application launch speed is largely a function of hard drive speed. Do you have a solid state drive in your machine? An upgrade from an old style spinning hard disk to a fast SSD makes a very big seat of the pants difference.