• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Xeon D, Broadwell for cloud and web

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Ars has a very nice write-up of the CPU. Noticed this part:

As well as packing in a bunch of processing power and connectivity, the processors include a range of power management and power saving features in a bid to improve efficiency. These include a power optimized turbo system, wherein the processor tracks whether the boosted clock frequency is actually providing meaningfully increased performance. If it isn't, it'll cut back the power budget assigned to a core and use it elsewhere.

The chip can also autonomously make decisions about which power and sleep state to use, rather than depending on the operating system to control these things. Intel says that this means the system can respond more quickly, and it also enables, for example, an entire rack of systems to be configured with the same power management configuration, independent of the operating systems running in the rack.

http://arstechnica.com/information-...s-its-first-foray-into-soc-space-with-xeon-d/
 
Just by the fact this is a 14nm part, shows how high this was on Intel's priority. Cannot wait for Skylake parts after seeing this. 🙂
 
Just by the fact this is a 14nm part, shows how high this was on Intel's priority. Cannot wait for Skylake parts after seeing this. 🙂

With everything they have put on this die, this bodes very well for those 65w desktop BW CPU's.
 
With everything they have put on this die, this bodes very well for those 65w desktop BW CPU's.

I wonder if the only reason the desktop BW exists is because they can use so much from Xeon D... No way Intel is going to put 8 cores in 1150 casually. A man can hope though
 
my understanding is that there isn't any FMA improvement from HSW to BDW, what do you have in mind ?

I assume he refers to this:
Slide%2025%20-%20Overview_575px.png
 
Last edited:
Gawd...I wished these chips ran at 5ghz and were widely adopted by all the kewl cloud computing platforms that I have to use - no matter the speed of the pipe, simple database CRUD operations that used to take milliseconds on a local PIII with a noisy, hot, unhappy spindle drive now take 10-50 times as long. Or longer if it's a busy day at the DB cluster.

Yes...I've been using WP/OP/IS/GTW/S3 today...painful...
 
I think non x86 architectures have substantive market share in several server room niches such as networking and telecoms. Dual 10 GbE is attractive and additional SKU's with dedicated hardware accelerators are promised in the second half of this year. Xeon-D is broadly targeted. Is it reasonable to think these chips will to increase Intel's x86 server room market share?
 
I think non x86 architectures have substantive market share in several server room niches such as networking and telecoms. Dual 10 GbE is attractive and additional SKU's with dedicated hardware accelerators are promised in the second half of this year. Xeon-D is broadly targeted. Is it reasonable to think these chips will to increase Intel's x86 server room market share?

Intel x86 already makes up like 99.9999999% of the server room market share.
 
Intel x86 already makes up like 99.9999999% of the server room market share.

Intel has been taking share from big Unix and mainframes on proprietary hardware for years but I believe big Unix on architecture such as Power8 and Z retains >10% share based on revenue . However, I focus elsewhere. Cavium is one embedded player with niche share in Telecom. The Semiaccurate link, http://semiaccurate.com/2015/03/09/cavium-octeon-fusion-m-base-station-chip/, discusses one offering but my impression is that MIPS is well regarded in this niche with multiple well received products. As Intel develops appropriate IP and gets better at spinning custom hardware solutions, this share is likely to migrate to X86. The jury is still out on SDN but part of the idea behind SDN is to replace proprietary dedicated hardware with generic, programmable X86 solutions. I view SDN as an Intel effort to gain market share. How much server room share remains for Intel to acquire? Most switches do not support Intel's optical networking solutions. Intel must participate in the switch market to enable it's optical networking solutions.

It's easy to find pdf's reporting that ARM targets a 20% server room market share in 5 years. However, in recent years, the only area I can think of where ARM may be taking share from Intel in low end NAS devices. In mobile, Intel is going from no share to a positive share. I think it is more likely that in 5 years Intel will capture 20%+ of the current non X86 server room market share. As we get to the second half of the year, I suspect Xeon D may be part of this share acquisition effort.
 
Last edited:
It's easy to find pdf's reporting that ARM targets a 20% server room market share in 5 years. However, in recent years, the only area I can think of where ARM may be taking share from Intel in low end NAS devices.

One thing to keep in mind is that hosted providers/"The Cloud" is becoming more popular, while I imagine the traditional server market is going to decline. I imagine the providers are using Xeons but could switch to ARM based products if it made sense to them. So it's easy to see why Xeon D was such a high priority, to protect any customers from thinking about defecting. Still, there is an opening if K12 or some other small core type ARM chip could produce much better MT perf/W even if it blows at ST.
 
Cloud is a mess for a lot of things, and is still much more hype than reality, as far as replacing most anything but email, that would otherwise have been affordable in-house (one reality it does bring is lower cost for potential risks, that might later be brought in-house, sporadic short-term needs, or building out from having zero network infrastructure and no need for a domain).

What is happening with traditional servers, though, is consolidation. Between hardware and licensing costs, redundant 1S and 2S boxes can do a good job of handling most workloads well enough for internal uses, and saves space and cost to run. IoW, I really don't see the cloud causing a decline in traditional server sales, as much as, just like desktops, performance of servers has far exceeded, "good enough." Virtualization software being mature enough to not have to constantly fiddle with it, the CPUs getting faster year over year, and storage getting faster and cheaper year over year, is creating its own cycles of unit sales decline, as old server installations are refreshed. Every 5 years or so, you can basically cram about double the crap you could before into a new server of similar cost, and operational cost savings will tend to make up for any added costs for getting a beefier server, or licensing for special software (various VMWare add-ons, backup appliances, VSANs, etc.).
 
I was testing a Supermicro X10SDV board today with passmark performance test 8. I was rather impressed with the Xeon D-1540 CPU Mark score of 11977
Overall this puts it between a i7-5820K and an i7-4790K.
https://www.cpubenchmark.net/high_end_cpus.html

It ran neck and neck with the 6-core Haswell-E i7 on the individual CPU test results, excluding single thread performance (single thread score was 1574, ~38% lower)
Single thread performance was more in line with a Sandy Bridge i3
https://www.cpubenchmark.net/singleThread.html

I uploaded the results to Passmark, hopefully they will show up in the database soon.
pretty impressive for a 45W SOC, to be able to hang with 50-120W E5-2600v3 cpus
 
Last edited:
I was testing a Supermicro X10SDV board today with passmark performance test 8. I was rather impressed with the Xeon D-1540 CPU Mark score of 11977
Overall this puts it between a i7-5820K and an i7-4790K.
https://www.cpubenchmark.net/high_end_cpus.html

It ran neck and neck with the 6-core Haswell-E i7 on the individual CPU test results, excluding single thread performance (single thread score was 1574, ~38% lower)
Single thread performance was more in line with a Sandy Bridge i3
https://www.cpubenchmark.net/singleThread.html

I uploaded the results to Passmark, hopefully they will show up in the database soon.
pretty impressive for a 45W SOC, to be able to hang with 50-120W E5-2600v3 cpus

How is wall socket power usage?

Is this a true 45W or is it "45W"?
 
How is wall socket power usage?

Is this a true 45W or is it "45W"?

I'll see if I can tell you that tomorrow. I've got the system loaded up with 64GB DDR4 (4x16GB DIMMS) Hoping I can find my killawatt. BUT, you can also run this board off of 12VDC, so I may just hook it up to a bench supply.

I couldn't find drivers for the 2 10G NICS, I'd really like to factor that into a power measurement as well.

It's a very neat little board, compared to most of the mini-ITX boards that supermicro does. Loves that it has 4 full size DIMM slots vs just SoDimms on their other boards
 
This sounds like a really great NAS / home server / transcoding server chip.

Now if regular PC motherboards would just start bundling 10GbE NICs onboard. We've had Gigabit for what, like almost 10 years now, with no obvious improvement?
 
I'll see if I can tell you that tomorrow. I've got the system loaded up with 64GB DDR4 (4x16GB DIMMS) Hoping I can find my killawatt. BUT, you can also run this board off of 12VDC, so I may just hook it up to a bench supply.

I couldn't find drivers for the 2 10G NICS, I'd really like to factor that into a power measurement as well.

It's a very neat little board, compared to most of the mini-ITX boards that supermicro does. Loves that it has 4 full size DIMM slots vs just SoDimms on their other boards

What OS are you running? The NICs should be supported by the general x540ish drivers if other websites are correct.
 
Back
Top