Intel Xeon D-1500 Series Discussion

[AmshTemp]

While seeing OpenSSL charts, can't help but wonder why 1518 is not kicking 1520 in the nuts?
I thought 15x8 parts are supposed to be specifically optimized for such operation.

Why would I pick a more expensive 15x8 if it performs, somehow, similarly to its 15x1 counterpart?

 

[Evan]

While seeing OpenSSL charts, can't help but wonder why 1518 is not kicking 1520 in the nuts?
I thought 15x8 parts are supposed to be specifically optimized for such operation.

Why would I pick a more expensive 15x8 if it performs, somehow, similarly to its 15x1 counterpart?

35 watt tdp for the 4 & 6 cores
Otherwise don't see the advantage...

 

[Patrick]

While seeing OpenSSL charts, can't help but wonder why 1518 is not kicking 1520 in the nuts?
I thought 15x8 parts are supposed to be specifically optimized for such operation.

Why would I pick a more expensive 15x8 if it performs, somehow, similarly to its 15x1 counterpart?

Intel did a bit of marketing on this. The D-15x8 chips do not have a built-in QuickAssist engine. They need a Coleto Creek chipset or add-in card.

 

[AmshTemp]

Can I please "unsee" that statement, Patrick?
You will crush some hearts @ pfSense.

 

[vaga]

But D-15x8 has DPDK, isn't it?
At Data Plane Development Kit - Wikipedia, the free encyclopedia I found:
The pfSense project published a road map on 25 February 2015, in which developer Jim Thompson announced the rewriting of the pfSense core—including pf, network packet forwarding and shaping, link bonding, IPsec—using Intel's DPDK: "We have a goal of being able to forward, with packet filtering at rates of at least 14.88Mpps. This is 'line rate' on a 10Gbps interface. There is simply no way to use today's FreeBSD (or linux) in-kernel stacks for this type of load."

So, looks like D-15x8 and pfSense combination has some advantages (at least in the future) over D-15x0 and D-15x1.

But does it have any advantages over D-15x0 and D-15x1 in nowadays Linux and FreeBSD?

 

[Davewolfs]

Feature request for future reviews. Please include a single thread test.

 

[Patrick]

@AmshTemp the pfSense team actually pointed this out when I spoke to them but QA on pfSense I do not believe is officially supported yet. I am planning to do a piece with them when it is. I am also planning on having another discussion with the Intel team on doing QA benchmarking this week. Something I have been working on for awhile.

@Davewolfs a few ideas here:

1. We are running single threaded tests (UnixBench Whetstone/ Dhrystone, Sysbench CPU as examples)
2. We are publishing the more complete set of results on Linux-Bench Linux CPU Benchmarks by ServeTheHome and ServeThe.Biz
3. If there is something you have scripted that you would prefer we run, happy to add in next evolution

 

[C@mM!]

Still waiting on X10SDV-16C-TLN4F availability.... (or at least 1541 availability)

 

[JimPhreak]

Still waiting on X10SDV-16C-TLN4F availability.... (or at least 1541 availability)

Same. I'm interested in both.

 

[Davewolfs]

@AmshTemp the pfSense team actually pointed this out when I spoke to them but QA on pfSense I do not believe is officially supported yet. I am planning to do a piece with them when it is. I am also planning on having another discussion with the Intel team on doing QA benchmarking this week. Something I have been working on for awhile.

@Davewolfs a few ideas here:

1. We are running single threaded tests (UnixBench Whetstone/ Dhrystone, Sysbench CPU as examples)
2. We are publishing the more complete set of results on Linux-Bench Linux CPU Benchmarks by ServeTheHome and ServeThe.Biz
3. If there is something you have scripted that you would prefer we run, happy to add in next evolution

Just my opinion but what I think is important is to always have a very basic metric that shows you where on the scale a CPU scores relative to everything else that is out there. The current test obviously displays that this CPU outperforms every CPU that it is compared against. But we still don't know exactly how it compares to say a E5-2670 V1 at single threaded tasks.

One test that I do like (even if it is not perfect) is Passmark. I like it because you get to see a score for single threaded, multi threaded and dual CPU capable systems. The number you get allows you to see exactly how your CPU relates to the rest of the pack both old and new. I think if you could include Passmark as part of your numbers it would be useful as it would easily allow us to see where exactly these CPU's sit and what kind of relative performance can we expect compared to everything else out there.

Additionally with something like a Passmark score one can easily take that and compare it to the database which is published on their website. I don't know about others but I find this useful in making decisions, in fact it really helped solidify my decision that buying the E5-2670 was a no brainer.

 

[MiniKnight]

You can do single thread comparisons using the Linux Bench site? It's getting slow but there are lots of runs there

 

[Davewolfs]

That site needs a group by + average view.

 

[Patrick]

That site needs a group by + average view.

Yes. Still not 100% happy with the development version that will replace beta and the original. The average view for this community is much harder. We have people running Docker containers, VMs and all sorts of other stuff which clouds the results quite a bit. I do want a grouping and a leaderboard by benchmark. I also want to have graphs for the main site and elsewhere automatically generated and auto-scaled.

@MiniKnight we are working on it. Complete DB re-do is going on on the dev site.

 

[mstone]

But D-15x8 has DPDK, isn't it?

Did intel actually disable that on the other Ds? They've done stranger things, but it wouldn't make much sense. I think the draw of the D...8 series is the lower TDP for many SKUs, as well as the extended availability. (Something that matters for people developing embedded products, not so much for people buying them in low volume.)

 

[AmshTemp]

That's really great.
Can you also ask Intel about what is the real difference between 15x1 vs 15x8, the REAL one.
If I went with the highly available 1541, am I for example sacrificing network DPDK performance?

 

[Evan]

Intel offers cpu's that so far that we don't see in any retail 'boards'. I am assuming some vendors must use Xeon-D in appliances but so far I don't hear of anything. Anybody know who/what else is using them ?

Do we expect any more different supermicro, gigabyte, asrock etc boards in the near future ?

Cooling question ... Would a 120mm fan blowing directly at one of the mini-ITX boards provide enough cooling for the passive heatsinks being used by SM ?

 

[JimPhreak]

Intel offers cpu's that so far that we don't see in any retail 'boards'. I am assuming some vendors must use Xeon-D in appliances but so far I don't hear of anything. Anybody know who/what else is using them ?

Do we expect any more different supermicro, gigabyte, asrock etc boards in the near future ?

Cooling question ... Would a 120mm fan blowing directly at one of the mini-ITX boards provide enough cooling for the passive heatsinks being used by SM ?

I'd really like to know the answer to both these questions as well as I'm still on the hunt for a server case that will quietly cool a D-1541 (waiting for a MicroATX option) that gets a LOT of workload.

 

[hjfr]

Thanks for your answer.


That's it ! I bought, this morning, my "X10SDV-4C-TLN4F" (D-1518) from Sona.de after they confirmed they have some boards in stock (limited). Price: 555€ without shipping
UPS transit now !
I think, it's very difficult to have Xeon D boards (Supermicro or others) in Europe. Hard to find resellers who have stock.

Other Info: all Xeon D "wave 2" (1541/1528/1518/etc) boards from Supermicro have PCB 2.0 (1.0x for 1520/1540).

I received it this monday.
First test OK. It boots. But very hot after 10 minutes of operation on a table without any fan. The red led lights up

 

[Evan]

I received it this monday.
First test OK. It boots. But very hot after 10 minutes of operation on a table without any fan. The red led lights up

Any chance you could run it with some load and just a larger case fan pointing at the board and see what the temps are ?

@JimPhreak i was thinking to use the Ncase M1 for my Xeon-D build and just having the 120mm fan on the side of the case blow directly onto the motherboard.
Ideal would have been a case like a 1u server that has all blow through cooling but none to be found that done have holes in all the side panels rendering the blow through cooling mostly useless.
One option I can exercise prior to purchase is simply to go with the active cooled version but I think the passive is better.

 

[Patrick]

I received it this monday.
First test OK. It boots. But very hot after 10 minutes of operation on a table without any fan. The red led lights up

Yes, these are meant to be cooled via chassis fans.