25w Pentium D-1508

Patrick

Administrator
Staff member
Dec 21, 2010
11,971
4,932
113
Personally I'm very happy seeing SM offering SFP+ instead of much more expensive copper. There's really NO argument for choosing Base-T instead.
Well the fact that they are backward compatible with 1Gb networks and there are a lot of racks with Cat 6 already installed are two reasons. You would be surprised how much I hear those two cited as reasons for wanting 10Gbase-T.
 

mstone

Active Member
Mar 11, 2015
505
117
43
42
Well the fact that they are backward compatible with 1Gb networks and there are a lot of racks with Cat 6 already installed are two reasons. You would be surprised how much I hear those two cited as reasons for wanting 10Gbase-T.
Yeah, if you're building a cluster incremental upgrades aren't a factor but if you're running a heterogeneous environment it's important.
 

JimPhreak

Active Member
Oct 10, 2013
553
55
28
Well the fact that they are backward compatible with 1Gb networks and there are a lot of racks with Cat 6 already installed are two reasons. You would be surprised how much I hear those two cited as reasons for wanting 10Gbase-T.
Can you not plug an SFP transceiver into an SFP+ port for backward compatibility (assuming your 1Gb switches have SFP ports)?
 

eroji

Active Member
Dec 1, 2015
268
44
28
37
Wouldn't a 4C/8T version be slightly better suited for the use case? For the rest of the hardware on the board, which are great, it just seems a little low on the core number to handle multiple file transfers concurrently.
 

JimPhreak

Active Member
Oct 10, 2013
553
55
28
Wouldn't a 4C/8T version be slightly better suited for the use case? For the rest of the hardware on the board, which are great, it just seems a little low on the core number to handle multiple file transfers concurrently.
There is a 4c/8t version in the D-1518. However depending on how many simultaneous transfers you'll have going I would think the 1508 would work well for a lot of users.
 

eroji

Active Member
Dec 1, 2015
268
44
28
37
There is a 4c/8t version in the D-1518. However depending on how many simultaneous transfers you'll have going I would think the 1508 would work well for a lot of users.
Possibly. Wonder what the price difference between the 2 is.
 

Evan

Well-Known Member
Jan 6, 2016
3,128
522
113
Sometimes. Depends on the NIC.
And on the switch side the answer is generally yes, but the switch asic must support the lower speed.

Example of it not working... Copper GLC-T= into a N9372PX, so it should support 10/100/1000 but since the switch asic only supports 1000/10G/40G it won't allow the SFP to run at 100 speeds (sometimes you have OBM devices still at 100)
SFP is on the supported list for switch just not at all speeds.

On the card side I personally have never tried anything but the intended SFP modules for the card... A 10G SFP though will often run at 1G but not all. I found some IBM equipment where it's 10G only. (Again I assume that's all the asic support)

I will be getting some Intel (HP branded) cards shortly, I may true a few different SFP's see what works but most SFP I have sitting around are all Cosco branded so not sure if I really have much luck.
 
Last edited:

Evan

Well-Known Member
Jan 6, 2016
3,128
522
113
We are all eagerly awaiting the pricing details.
D-1518 boards despite being $35 more for the chip are running at the same price as d-1520 ones so maybe that's a good sign.

It would seem seem like d-1508 would easily outgun the c2550 and the d-1518 the c2750 and I am going to Guess the prices are not that different we hope, power consumption slightly greater but I wonder if we see a succession plan for c2??? CPUs still ?
 

Patrick

Administrator
Staff member
Dec 21, 2010
11,971
4,932
113
Still no word on the D-1508 but the D-1557 seems like it is getting closer.

@Evan Denverton is the successor for Rangeley/ Avoton.
 

Evan

Well-Known Member
Jan 6, 2016
3,128
522
113
Still no word on the D-1508 but the D-1557 seems like it is getting closer.

@Evan Denverton is the successor for Rangeley/ Avoton.
But will it happen ?
Shouldn't we be seeing a release already, any chance it's canceled and just not yet public info.

[edit]
Still on the public road map,
http://www.intel.sg/content/dam/www/public/us/en/documents/roadmaps/public-roadmap-article.pdf
10g, 16 lanes PCIe, 16core, extra cache, 128gb ddr4 support.
Will it be a lot lower power than d-1500 platform ? :)
 
Last edited:

mstone

Active Member
Mar 11, 2015
505
117
43
42
Avoton was a play against arm. We probably won't see much emphasis on its successor unless arm actually does something in the server space that Intel sees as a threat. The market seems a lot more interested in something with more single thread horsepower. Between the Xeon d and the skylake e3's there just isn't much room for the avoton/rangeley style machines.
 

Patrick

Administrator
Staff member
Dec 21, 2010
11,971
4,932
113
@mstone - we are going to see ARM later this year from what I am hearing, and not in the little RPi 3's.

@Evan - it is coming.
 

mstone

Active Member
Mar 11, 2015
505
117
43
42
@mstone - we are going to see ARM later this year from what I am hearing, and not in the little RPi 3's.
There's a difference between "finally selling what they've been talking about for years" and "threat". There's a latency tradeoff when moving from a few fast cores to many slower cores, and thus far it seems that the marketplace cares more about better latency than saving a little bit more power. Moving to ARM has a switching cost, and they'd need to have a significant advantage to justify it (order of magnitude maybe?) -- especially with Intel's process advantage and the ability to fund multiple approaches simultaneously. If they're only a little bit better at something, Intel will likely come out with a new competitor before the switching cost can be amortized. That's the context I see the avoton & successor in: they showed that intel can go in that direction, they established a new goal post, and they're a model intel can dump more resources there if they need to--but I don't see intel really going all-in on that model unless the new ARM chips are far enough ahead to pose a real threat because the marketplace isn't signaling that's the direction it wants to go. And even then, intel would probably just be hedging with that line while putting more effort into tiger lake.

Especially now that we're on the edge of a paradigm shit to NVMe storage, persistent memory, etc., which could radically alter our ability to keep fast processors busy; this is probably the wrong time to be coming out with a product that has relatively low bandwidth per core.
 

Patrick

Administrator
Staff member
Dec 21, 2010
11,971
4,932
113
@mstone from what I hear the Cavium 48 core multi-socket parts are shipping and sound very good. Still not volume enough that we have them yet, but supposedly we will get them when volume picks up.
 

mstone

Active Member
Mar 11, 2015
505
117
43
42
@mstone from what I hear the Cavium 48 core multi-socket parts are shipping and sound very good. Still not volume enough that we have them yet, but supposedly we will get them when volume picks up.
From what I hear they would have been great a year or two ago. (E.g., focusing on large numbers of SATA III interfaces seems more appropriate to 2015 than 2017.) By the time they're readily available we'll be comparing them (I suspect rather unfavorably) to Intel's purley. I'm sure there are workloads that fit them well, but it's going to come down to a question of whether those workloads are important enough, and the performance differences significant enough, and whether the ARM vendors can refresh quickly enough to justify switching architectures. Some people will switch just to get away from x86, but I'm not sure there are enough of them to pay the bills to keep driving the refresh cycle.

But once we see real performance numbers this will be more clear. It's somewhat telling that Cavium hasn't really been pushing on that...
 

Davewolfs

Active Member
Aug 6, 2015
337
31
28
Was thinking about this. It would make a great backup machine. Something simple, minimal ram, no zil. Just backup.
 

Evan

Well-Known Member
Jan 6, 2016
3,128
522
113
If one ever ships... Problem is for the arm guys that unless it's decent cheap and has a whole lots of wanted features like lots of sata for a storage server at a few watts of power then it's easier to use Intel and have good performance plus great compatibility.

Where is our d-1508's ? Must drop soon.