Intel Xeon D-1500 Series Discussion

[NeverDie]

I suspect the E3 refresh will have the biggest impact near-term but that most of us will be moving over to the D series within 2-3 years, after SSD's get faster.

It has me curious as to when 10GigE will go mainstream in regular, non-Xeon CPU's/PC's? Maybe three years? It will be exciting to share with the server at 10GigE speeds, build faster SAN's, migrate VM's super fast, and do all that fun stuff. It's all chicken-and-egg: the Xeon's have to come first to pave the way for that, and so now we're another step closer to that inevitability.

 

[dba]

For now, the quad core version is also a 45w TDP part. There were only two launch SoCs but there are other variants coming in the next quarter or two.

Here is the Intel Xeon D-1520 page: ARK | Intel Xeon Processor D-1520 (6M Cache, 2.20 GHz)

Darn. I read the link from EffrafaxOfWug (here) which said 20W. Intel Ark is certainly more authoritative, so 45W it is. That's still quite good considering the dual 10GbE, etc. but not quite as impressive.

 

[ATS]

It has me curious as to when 10GigE will go mainstream in regular, non-Xeon CPU's/PC's? Maybe three years? It will be exciting to share with the server at 10GigE speeds, build faster SAN's, migrate VM's super fast, and do all that fun stuff. It's all chicken-and-egg: the Xeon's have to come first to pave the way for that, and so now we're another step closer to that inevitability.

Biggest issue with non-server room usage of 10GbE is 10GBase-T being a massive power hog and still rather expensive. Plus, unlike 1G where 1GBase-T was able to leverage the server infrastructure investment, 10GBase-T is very rare in the server environment because SFP+ is generally cheaper and significantly less power. Honestly, I was quite shocked when the Supermicro board had Base-T instead of SFP+.

 

[cesmith9999]

http://www.servethehome.com/asrock-rack-intel-xeon-d-platform/

@Patrick this ASRock board does look very interesting at least IMO compared to the Supermicro. I see some what looks like pigtail antenna leads on the board running down the middle, wifi? Also what are those 2 black ports next to the PCI-e 8x slot for? Extended ITX seems a bit funny of a form factor, I guess smaller than M-ATX but bigger than ITX?

what is the MB dimensions? and I am wondering the same questions.

Chris

 

[NeverDie]

Biggest issue with non-server room usage of 10GbE is 10GBase-T being a massive power hog and still rather expensive. Plus, unlike 1G where 1GBase-T was able to leverage the server infrastructure investment, 10GBase-T is very rare in the server environment because SFP+ is generally cheaper and significantly less power. Honestly, I was quite shocked when the Supermicro board had Base-T instead of SFP+.

How many watts does 10GBase-T use, and is it any less for being 14nm SOC? i.e. does that change the game at all?

 

[chinesestunna]

Biggest issue with non-server room usage of 10GbE is 10GBase-T being a massive power hog and still rather expensive. Plus, unlike 1G where 1GBase-T was able to leverage the server infrastructure investment, 10GBase-T is very rare in the server environment because SFP+ is generally cheaper and significantly less power. Honestly, I was quite shocked when the Supermicro board had Base-T instead of SFP+.

I agree, but from the looks of how cramped that board is I think SFP+ would've been very hard to fit 2 ports, in addition to reserving space for the LN4 version with 2 more 1GBps ports

 

[Entz]

Yeah there is very little room on there: Could of gotten rid of the serial port maybe? or put it on a riser. Supermicro also made the choice to go with 4 full sized DIMM slots (which is very rare on ITX boards) and that eats up nearly all the free space in the IO shield area.

mATX is going to be the sweat spot for Xeon-D i think.

 

[Patriot]

Darn. I read the link from EffrafaxOfWug (here) which said 20W. Intel Ark is certainly more authoritative, so 45W it is. That's still quite good considering the dual 10GbE, etc. but not quite as impressive.

While it is more authoritative... I have found it to be wrong on multiple occasions.

 

[NeverDie]

Answering my own question (above) about the 10GBase-T:

[Dang that's annoying. Why do these pictures so often show perfectly well while editing but not at all when posted? Here's the url: http://m.eet.com/media/1115390/plxtab2.gif]

 

[gigatexal]

Are these available at trusted sources yet? The board pictured in the article with full sized ram slots is what I'd like to buy. 16GB ECC sticks are going to cost a bit but I'll sell almost all my unused hardware just to get this little guy running.

 

[Patrick]

[Dang that's annoying. Why do these pictures so often show perfectly well while editing but not at all when posted? Here's the url: http://m.eet.com/media/1115390/plxtab2.gif]

I think is an issue with an XF feature we are using for HTTPS. When you insert an image from a non-HTTPS source if you do not proxy the request it ends up triggers all kinds of issues. Still working through what happens in these few instances where it does not work.

@gigatexal The general feedback is that mid to late April we should start seeing volume shipments. Many vendors I would expect to announce product in April.

Edit just to add - I should be getting some of the pre-production/ validation platforms from at least two vendors next week.

 

[gigatexal]

Super excited. I'll keep an eye out.

 

[ATS]

How many watts does 10GBase-T use, and is it any less for being 14nm SOC? i.e. does that change the game at all?

Most of the additional power for 10GBase-T is in the Phy. The 10GBase-T Phy is having to do a lot more work than any of the SFP+ phys because the signal environment is so much worse.

 

[Patrick]

OK I have a non-10Gbase-T pre-production sample in the lab. Hello 64GB. 47w in bios with way too much cooling and no power saving.

 

[gigatexal]

What temps???

 

[Patrick]

47.2w is the highest I have gotten
4.8w BMC only power off
26.6w Ubuntu installation screen
37.2w UnixBench 5.1.3 single thread
88.1w UnixBench 5.1.3 multi thread max

This is with 4x 16GB Samsung DDR4 RDIMMs

@gigatexal will do thermals later. Heatsink barely warm to the touch

 

[gigatexal]

Considering I have two x5670s heating my living room this is welcome news.

 

[Patrick]

Some update:

 

[EffrafaxOfWug]

Hovering around 25W idle is pretty nice for an octoproc at 64GB of RAM. Is your pre-prod sample merely without the 10Gb BASE-T PHY or does it lack 10Gb networking entirely? If there's a mITX on the horizon that can idle with a 10Gb SFP under 35W I might have to brandish a large wad of currency and angrily wave them in the face of a sales rep demanding he desist in informing me of said product and relieve me of said paper post-haste.

 

[mstone]

Biggest issue with non-server room usage of 10GbE is 10GBase-T being a massive power hog and still rather expensive. Plus, unlike 1G where 1GBase-T was able to leverage the server infrastructure investment, 10GBase-T is very rare in the server environment because SFP+ is generally cheaper and significantly less power. Honestly, I was quite shocked when the Supermicro board had Base-T instead of SFP+.

1000base-T used to be really expensive also, and it only worked some of the time, and you'd buy ridiculously hot running add-in cards that were 1000base-T only (couldn't run at 100) and actually reaching gigabit speeds was really hard even with a really expensive card. All of the sudden one day there will be just enough refinement that 10Gbase-T will in every 5 port switch at walmart and we'll wonder what the big deal was. I feel like we're just about there--I don't think it'll even be three years before it becomes standard on higher-end consumer motherboards.