Intel Quietly Kills Off Xeon Phi 7200 Coprocessors

[Guest]

Intel's Xeon Phi 7200 Coprocessors, the descendants of Intel's failed Larrabee GPU project, are apparently being retired.

Intel Quietly Kills Off Xeon Phi 7200 Coprocessors : Read more

 

[Vatharian]

Curious move - It is probable, that they are afraid of people buying EPYC systems and deploying multiple coprocessors with this?

 

[knowom]

If they are smart they are ditching it in favor doing a similar approach with FPGA's instead which are way more multipurpose and can be re-purposed which is great.

 

[bit_user]

It was indeed a forum post, but on Intel's own site:

https://software.intel.com/en-us/forums/intel-many-integrated-core/topic/738422

And just to clarify, they killed the PCIe card - not the socketed version that's also the host processor.

Regarding Knights Mill: that's specifically aimed at the deep learning market. It has half the fp64 performance of Knights Landing, making it less suitable for HPC workloads. So, I expect we'll continue to see both being sold until Knights Landing has a proper replacement.

 

[Vatharian]

If they are smart they are ditching it in favor doing a similar approach with FPGA's instead which are way more multipurpose and can be re-purposed which is great.

Intergrating FPGAs has different target - tighter intergration. Btw, they have very old FPGAs intergrated, it's more of a proof of concept so far, while KNL is real product.

My problem is that you can stick only so many KNLs in motherboard (I'm not convinced they support 4 sockets), so in 7U space you could fit probably 28 chips given 4S support. In PCIe mode, I've personally built a system that contains 18x 7120 (watercooled) cards in 2U chassis on PCIe backplane. Host system supported three of those backplanes, allowing total 54 Xeon Phis in 7U space.

Situation looks little better if you use narrow sleds, but they are pricey. PCIe is just so much more universal.

Of course socketed, Omnipath or not, is brilliant idea, but forcibly moving solution to another form factor 'just because'? Not very fun, oh no.

 

[bit_user]

Curious move - It is probable, that they are afraid of people buying EPYC systems and deploying multiple coprocessors with this?

Why would they be afraid of selling more product?

No, I think they'd be glad to sell 4 Xeon Phi co-processors per system instead of a single socketed processor, regardless of whether the host is a Xeon or Epyc.

My guess is that most people buying the PCIe version were probably more interested in Knights Mill. Either that, or there were technical problems with the PCIe Knights Landing.

 

[bit_user]

My problem is that you can stick only so many KNLs in motherboard (I'm not convinced they support 4 sockets),

They only support single-socket.

so in 7U space you could fit probably 28 chips given 4S support.

I think the solution to this is to put them on blades, in a blade server.

In PCIe mode, I've personally built a system that contains 18x 7120 (watercooled) cards in 2U chassis on PCIe backplane. Host system supported three of those backplanes, allowing total 54 Xeon Phis in 7U space.

That's impressive, but only useful for workloads that don't require a lot of cross-communictaion.

PCIe is just so much more universal.

For scale, they're going with OmniPath, instead.

 

[Guest]

Is Knights Mill even similar? I don't think it is, that's why they don't list core count, it's not a similar product.

 

[bit_user]

Is Knights Mill even similar? I don't think it is, that's why they don't list core count, it's not a similar product.

Announcing Knights Mill, building on top of Knights Landing
...
Builds directly on top of KNL
...
Same core config of KNL: 2 cores sharing 1MB of L2, one VPU per core
...
In KNL, two units do SP and LP

In KNM, remove one DP ports to give space for four SP VNNI units

So 0.5x DP, 2 x SP, 4x VNNI

Pitching KNM for DL but with tradeoffs, same generation as KNL

Source: http://www.anandtech.com/show/11741/hot-chips-intel-knights-mill-live-blog-445pm-pt-1145pm-utc

Core counts were not stated because it's not yet a shipping product. By all accounts (and based on what Intel has previously said about it), it's a variation of KNL tweaked to better target deep learning (i.e. because they were non-competitive with Nvidia's P100).

 

[SockPuppet]

The people that run Intel are complete idiots. I don't think the senior management could set the clock on their microwave without the help of their children.

GPUs are a BIG business and only getting bigger every day. Intel is 2nd to none at etching circuits into silicon. But management "doesn't see any value in making discrete GPU parts". Imbeciles.

 

[iLLz]

To be honest, with the recent announcement of Knights Mill, it makes sense not to out Knights landing since they have a newer version in the works. Knights Mill is supposed to have 4x Deep Learning performance, according to Intel.

 

[bit_user]

GPUs are a BIG business and only getting bigger every day. Intel is 2nd to none at etching circuits into silicon. But management "doesn't see any value in making discrete GPU parts". Imbeciles.

I can only agree that their machine learning ambitions would've been better served by using HD Graphics, as a foundation, and bolting these big vector pipelines onto its cores.

Intel learned too well from their prior attempts to deviate from x86. Now, they seem unwilling to try anything else.

That said, it will be interesting to see what comes of their machine learning & computer vision acquisitions.

 

[thelazy_1]

Intel, just IGP it with the CPU like all other CPUs you make. Problem solved.

 

[wifiburger]

haha ! Yeah not sure who believed Intel would push that many low power core cpu ! they have those expensive xeons to sell for $$$

Im sure arm can push this type of architecture any time !

 

[DavidC1]

"To be honest, with the recent announcement of Knights Mill, it makes sense not to out Knights landing since they have a newer version in the works. Knights Mill is supposed to have 4x Deep Learning performance, according to Intel."

Knights Landing is not cancelled. The co-processor PCIe version is. The socketed version that came out first is very much alive.

For general purpose HPC Knights Mill only offers 1/2 the performance of Knights Landing. Knights Mill is for Deep Learning.