Xeon D, Broadwell for cloud and web

[ShintaiDK]

http://techreport.com/review/27928/intel-xeon-d-brings-broadwell-to-cloud-web-services

Its not exactly getting any easier for any hopeful upcoming ARM competitors.

In short, this thing is a 8C/16T processor with 12MB of L3 cache, 24 lanes of PCIe Gen3, and dual 10GigE links.

TDP ratings from "under 20W" to 45W

We don't yet have full pricing and specs on the various Xeon D models Intel will offer. We do know that the Xeon D-1450 will have eight cores with a base clock of 2.0GHz, an all-core Turbo peak of 2.5GHz, and a single-core Turbo peak of 2.6GHz. Meanwhile, the Xeon D-1520 will feature a 2.2GHz base frequency, 2.5GHz all-core Turbo, and a 2.6GHz single-core peak. Both chips should be available this month.
Intel has provided us with a few preliminary benchmark results for the Xeon D compared to Avoton. They show the Xeon D to be as much as 3.4x faster with up to 1.7x higher performance per watt. However, those numbers are based on pre-production hardware and look kind of shaky. I suspect we'll see better numbers published in the coming weeks.

 

[Arachnotronic]

Did you see AT's comparison of X-Gene v.s. Avoton? Avoton offers >2x the perf/watt. Xeon D should be even more competitive.

I would like to see future Xeon D products integrate everything into a single die, though, rather than the multi-chip package.

 

[bullzz]

http://finance.yahoo.com/news/intel-packs-power-intelligence-xeon-160000140.html

$581 for 8 core. not bad

 

[NTMBK]

Looks like a great chip. I wonder what the die size is like compared to a Xeon E3? Dropping the GPU will leave plenty of space for extra cores.

 

[ShintaiDK]

2 of the models can be found on ark.intel.com now:

http://ark.intel.com/compare/87039,87038

 

[Headfoot]

I can't seem to find what socket this is -- BGA only? 2011-3?

 

[ShintaiDK]

I can't seem to find what socket this is -- BGA only? 2011-3?

Custom socket I guess. The chipsize is the same as LGA115x chips.

 

[monstercameron]

I find it interesting that some posters claimed arm wasn't a factor in the server market...yet Intel has chosen to respond to this alleged non-threat.
Intel is just hedging their bet and it'll be interesting to see how the market responds.

 

[ShintaiDK]

I find it interesting that some posters claimed arm wasn't a factor in the server market...yet Intel has chosen to respond to this alleged non-threat.
Intel is just hedging their bet and it'll be interesting to see how the market responds.

So you think this product is to act on ARM, rather than simply requested by OEMs/hosting companies?

I think you value ARM servers a tiny tad too high.

 

[erunion]

I find it interesting that some posters claimed arm wasn't a factor in the server market...yet Intel has chosen to respond to this alleged non-threat.
Intel is just hedging their bet and it'll be interesting to see how the market responds.

Youre conflating ARM servers with dense servers. Obviously ARM vendors have been targeting dense servers because that's where their weakness matter least, but Intel has been present in dense servers for years. I'm sure you remember seamicro.

 

[Phynaz]

Wow, Intel dropped the hammer again.

Definitely earning the Chipzilla monicker.

 

[Burpo]

The first ever Intel Xeon SoCs, the D-1520 and D-1540, are now in production. These processors provided up to 3.4X the performance of an Intel Atom C2750 in a dynamic web serving platform test undertaken by Intel.* The new Xeons are said to boast up to 1.7X better performance per watt in the same example task.*

The Intel Xeon D is powered by the newest generation Broadwell cores. That means that the SoCs carry Intel's latest technology. This includes AVX 2.0, FMA improvements, AES-NI improvements, and functioning TXT and TSX. TSX was a feature first announced with Haswell generation parts, but a late bug discovery led to TSX being labeled a "developer only" feature in Haswell. Broadwell has the fixed version that is apparently ready for production.

These 64-bit SoCs have the following key configurations:

Intel Xeon D-1520: 4 cores / 8 threads, 2.2GHz, 45W
Intel Xeon D-1540: 8 cores / 16 threads, 2GHz, 45W
Further specs the two new Xeon SoCs have in common include:

Up to 5X network bandwidth thanks to integrated 2 x 10GbE Intel Ethernet
Integrated I/O such as 24x PCIe 3, 8x PCIe 2, 6x SATA3, 4x USB etc
Memory architecture based upon DDR4 DIMMs with up to 12MB L3 cache
Up to 128GB of addressable memory

"It's interesting to see Intel use its Xeon branding and technology in this SoC market space for network appliances and IoT devices. ARM is also targeting the billions of devices which are expected to make up the IoT in coming years. With so many devices expected to be connected to the internet in the coming years there's probably plenty of room for both corporations to prosper - but of course they have to fight tooth and nail to carve out their respective territories and that battle - with both hardware and software - has definitely started.

*Up to 3.4x better performance on Dynamic Web Serving

Intel® Xeon Processor D-based reference platform with one Pre-Production Xeon Processor D (8C, 1.9GHz, 45W, Turbo Boost Enabled, Hyper-Threading enabled, 64GB memory (4x16GB DDR4-2133 RDIMM ECC), 2x10GBase-T X552,
3x S3700 SATA SSD, Fedora* 20 (3.17.8-200.fc20.x86_64, Nginx* 1.4.4, Php-fpm* 15.4.14, memcached* 1.4.14, Simultaneous users=43844

Supermicro SuperServer* 5018A-TN4 with one Intel Atom Processor C2750 (8C, 2.4GHz,20W), Turbo Boost Enabled, 32GB memory (4x8GB DDR3-1600 SO-DIMM ECC), 1x10GBase-T X520, 2x S3700 SATA SSD, Ubuntu*
14.10(3.16.0-23 generic), Nginx* 1.4.4, Php-fpm* 15.4.14, memcached* 1.4.14, Simultaneous users=12896

http://hexus.net/tech/news/cpu/81472-intel-launches-xeon-d-1520-d-1540-64-bit-socs/

"Pricing & Availability

Intel Xeon Processor D-1520 (4-core) $199.00
Intel Xeon Processor D-1540 (8-core) $581.00

The products are available today. The extended Intel Xeon processor D product family including microserver, network, storage and IoT optimized SoCs is expected to be available in the second half of this year."

 

[Enigmoid]

The pricing is extremely attractive considering the chips have dual 10 gb ethernet capabilities.

$199 for the 4 core and $581 for the 8 core align pretty well to the consumer market.

That die looks tiny. No wonder they are so cheap.

 

[mrmt]

Well, this is the Microserver chip done right. I think AMD and others have a bright future becoming OEMs and selling Intel-powered servers on the market.

 

[DrMrLordX]

That's interesting. Doubling the core count and dropping 10% clockspeed keeps them at the same TDP? I wonder how far over TDP these chips run in "all core" turbo mode (2.5 ghz)? How long can they stay there?

 

[lefty2]

So, it's ment to give ARM microservers a run for their money. How would it compare to the Cavium 48-core Thunder X on price, performance/power, features?

 

[Arachnotronic]

So, it's ment to give ARM microservers a run for their money. How would it compare to the Cavium 48-core Thunder X on price, performance/power, features?

In anything that requires high per core perf, 48-core Cavium would not look too good

The big selling points for the Cavium ThunderX, IMO, isn't the general purpose CPU performance -- they're pretty weak cores. It's all of the accelerators and such that come integrated into the SoC which should make them good for targeted workloads (i.e. networking, storage, etc.)

 

[Vesku]

So you think this product is to act on ARM, rather than simply requested by OEMs/hosting companies?

I think you value ARM servers a tiny tad too high.

It all works together:

ARM Vendor: "Our upcoming product is efficient X Benchmark/W and Y cores, better than Avoton, and even has dual 10G Ethernet built in compared to the current 1G in this price category."

OEM: "Interesting, we'll think about it."

Later OEM is talking to their Intel rep.

OEM: "So company X has an interesting product with dual 10G for about $XXX. We are kind of interested but would obviously prefer to source from you."

Intel Rep: "We've got some very exciting products in the pipeline I'll see if I can get permission to provide you some more concrete details."

Bunch of Intel Reps send messages up the management chain "Company X is offering dual 10G and X Benchmark/watt with Y cores."

Xeon D plans adjusted accordingly and some details sent to OEMs.

 

[Burpo]

Yeah, imagine that.. intel redesigns BW just because

 

[SPBHM]

to be perfect it should be unlocked, without the lan stuff, and support for higher voltage DDR3, this would be a great regular desktop CPU, but it would kill the -E.

 

[MarkLuvsCS]

Looks like a pretty sweet chip for a custom/OEM NAS build. Anything to bring 10g more common/cheaper is awesome in my book!

 

[Nothingness]

Yeah, imagine that.. intel redesigns BW just because

Come on, they didn't have to touch the core, just use the right IP to complement the core.

What is funny is how well Intel can do server chips while at the same time they keep on making obsolete or badly targeted SoC for the mobile market. They really should concentrate on what they do the best (the same applies to ARM BTW).

 

[itsmydamnation]

Still no crypto accel, web farm doesnt want to waste its ipc with crypto operations. Amd needed to go bigger with its arm soc, not enough cores. If they did it would likely be preferable to thid. But now its not likely.

 

[Thala]

In anything that requires high per core perf, 48-core Cavium would not look too good

The argument here of course is, that you do not need high per core perf, because typical server workload scales nicely with cores. In the big picture, with scaling workloads, it seems a better idea to bet on thread level paralellism than on instruction level parallelism given a target die size and power envelope.

 

[mrmt]

The argument here of course is, that you do not need high per core perf, because typical server workload scales nicely with cores. In the big picture, with scaling workloads, it seems a better idea to bet on thread level paralellism than on instruction level parallelism given a target die size and power envelope.

Depends on the server. You do need some level of IPC according to the workload. If it's a network appliance then yes, 48 cores might cut it as well as a beefier core. But what if you are running a SAP ECC server? Then you want more cores but you don't want mid-range phone-level IPC on these cores.