ITT: We discuss Xeon D SOC as a foundation for desktop (eg, Broadwell-D/Skylake-D)

[cbn]

Currently the die variants for (22nm) LGA 2011-3 are 8C, 12C and 18C. I'm thinking if 14nm (either Broadwell or Skylake) moves the smallest LGA 2011 type die variant up to 12C we will probably see Intel produce a high clocked consumer version of Xeon D for enthusiast users.

Comparing Xeon-D SOC to the current 22nm eight core HEDT/Xeon E5 + 32nm PCH platform:

1. Eight core Xeon-D has 12 MB of L3 cache while the eight core LGA 2011-3 die has 20 MB of L3 cache
2. Xeon-D has dual channel memory controller while LGA 2011-3 has quad channel.
3. Xeon-D has 24 PCIe 3.0 lanes while LGA 2011-3 has for 40.
4. Xeon-D has six SATA 6 gbps ports while LGA 2011-3 has ten SATA 6 Gbps ports.

So while features are reduced the specification is still quite good.

One advantage I can think of using a Xeon D based die (either Broadwell or Skylake) would be reduced cost motherboards (assuming Intel makes the consumer version of the chip LGA) since the PCH is no longer on the motherboard.

Since motherboard cost has been a big part of the debate between both sides of the i7-4790K vs. i7 5820K argument I think Intel moving to a Xeon D based platform (for desktop) could make the choice a lot easier for those still on the fence of mainstream vs. a platform with greater than four cores.

So what features should Intel carry over from Xeon D to the consumer version? What core/thread counts should be offered?

 

[erunion]

I'm not following you.

Xeon-D is 45W and 2.6ghz turbo, its incompatible with HEDT.

 

[cbn]

I'm not following you.

Xeon-D is 45W and 2.6ghz turbo, its incompatible with HEDT.

What am referring to would have a higher clocks and its own motherboard socket.

 

[VirtualLarry]

I would love to see the Xeon-D released as a "poor man's HEDT" CPU. It makes a lot of sense. If you only need 6 SATA ports rather than 10, and only dual-channel RAM rather than quad-channel, but still want 8 cores, it seems like this chip would be the ticket.

http://www.anandtech.com/show/9070/intel-xeon-d-launched-14nm-broadwell-soc-for-enterprise

We probed Intel on the likelihood of seeing any consumer oriented implementations of such a platform, and although they recognized that there might be some niche situations they had not considered where Xeon D might be appropriate, they were being bullish on actually aiming anything at the consumer. Xeon D is purely an enterprise play.

Doesn't look so good.

 

[cbn]

I would love to see the Xeon-D released as a "poor man's HEDT" CPU. It makes a lot of sense. If you only need 6 SATA ports rather than 10, and only dual-channel RAM rather than quad-channel, but still want 8 cores, it seems like this chip would be the ticket.

http://www.anandtech.com/show/9070/intel-xeon-d-launched-14nm-broadwell-soc-for-enterprise


Doesn't look so good.

At the moment, the top HEDT SKU is eight cores. In order for a consumer Xeon D based platform to make sense the top HEDT SKU would have to move up to something like 12 cores.

For a Skylake D I would expect it to come after Skylake S and before the Cannonlake mobile SKUs launch. At some point after that Skylake E would launch with 12 cores as the top SKU.

 

[ShintaiDK]

So this is really another "Why dont we have cheap 8 cores" thread for the nTh time?

 

[cbn]

So this is really another "Why dont we have cheap 8 cores" thread for the nTh time?

Eventually eight cores will become relatively cheap, just like six cores became relatively cheap. Remember how much Gulftown cost?

 

[SPBHM]

yes, Xeon D with increased clock and 95W TDP would be the perfect $300-350 i7, I think, a real good successor to the i7 860-2600K and so on... but unfortunately it would probably kill most 2011 sales and would complicate their life with future products, also Intel is still focused on selling huge laptop IGPs on $300 95W desktop chips that few people buying these CPUs care about.

 

[Arachnotronic]

I don't think Xeon D would be able to clock high enough to be viable as a HEDT chip.

 

[imported_ats]

I don't think Xeon D would be able to clock high enough to be viable as a HEDT chip.

Xeon-D should be able to reach around into the mid 3s given a 130ish tdp.

 

[erunion]

Maybe someone else has concrete info on this, but I suspect its on a low power version of 14nm.

 

[cbn]

Maybe someone else has concrete info on this, but I suspect its on a low power version of 14nm.

I don't think Xeon D would be on a low power version of 14nm.

For example, this 65 watt 12 core Xeon E5 comes from the same die as the higher clocked E5 Xeons.

And the ULV consumer chips come from the same die as the high clocked desktop chips.

 

[Arachnotronic]

Maybe someone else has concrete info on this, but I suspect its on a low power version of 14nm.

That's what I think, too. Lower power/density optimized metal stack for this chip.

 

[cbn]

Intel is still focused on selling huge laptop IGPs on $300 95W desktop chips that few people buying these CPUs care about.

One thing I do like about having an iGPU (at least with the Intel Z boards) is that it lets me hook at third monitor up with just a single video card running the first two monitors.

That saves me a $30 active adapter and the need for a display ported enabled video card.

Though with that mentioned, the iGPU doesn't need to be large for that purpose.

 

[DrMrLordX]

If I had to guess, I'd say all the Broadwell products prior to the 5675C/5775C are probably low power/density optimized chips. Getting 14nm to work without being optimized for low power is (possibly) one of the reasons why desktop Broadwell was mostly canned in the first place.

 

[cbn]

yes, Xeon D with increased clock and 95W TDP would be the perfect $300-350 i7, I think, a real good successor to the i7 860-2600K and so on... but unfortunately it would probably kill most 2011 sales and would complicate their life with future products

My guess is that Skylake-D would replace the mainstream socket. (ie, mainstream LGA socket becomes a SOC based one like AM1). Then some of the mainstream iGPU chips becomes BGA only.

HEDT moves on to 12 cores at the high end (with the option to run even higher core count E5 Xeons).

 

[erunion]

That's what I think, too. Lower power/density optimized metal stack for this chip.

Ashraf just addressed this today.

http://www.fool.com/investing/general/2015/04/28/intel-corporation-a-look-at-the-new-xeon-d-process.aspx

 

[CHADBOGA]

Ashraf just addressed this today.

http://www.fool.com/investing/gener...oration-a-look-at-the-new-xeon-d-process.aspx

:biggrin:

 

[erunion]

Thanks. On my phone, usually it auto formats it.

 

[cbn]

Ashraf just addressed this today.

http://www.fool.com/investing/gener...oration-a-look-at-the-new-xeon-d-process.aspx

From the article (in which Xeon-D was compared to 28 watt 2 + 3 Broadwell U )

The CPU cores run at a base clock of 2.00GHz and can turbo up to a maximum of 2.6GHz, while the Broadwell CPU cores can turbo to up to 3.4GHz -- a 30% difference.

My guess: More of the Xeon D die is made up of cache memory, which is known to be denser than logic, helping chip density. Intel may also be using high density.

Furthermore, I suspect that the Xeon D is built using a metal stack optimized for density rather than for high frequencies (hence the max turbo of just 2.6GHz) to be able to integrate as much as possible into a relatively small footprint.

Two points:

1. Ashraf mentions difference in cache confounding a direct density comparison. The chips being compared are also quite different in other ways as well.

2. Can we really use Xeon-D's low turbo clock as evidence of a density difference? Remember i7 -5960X only turbos up to 3.5 Ghz despite a 140 watt TDP. Also there are numerous other SKUs within Intel's product line-up with lower than expected 1C turbo despite a healthy enough TDP. Therefore, A product with low 1C turbo is not proof of a deficiency in silicon. In some cases the low 1C turbo may be done for product differentiation purposes.

 

[cbn]

Just to clarify post #16.

I see a Skylake-D socket being used alongside LGA 1151. This, in addition, to whatever socket Skylake E uses.

However, I think it might be that LGA 1151 becomes the last mainstream socket to require a PCH. From that point forward there would only be two sockets on the desktop:

1. SOC based socket

2. HEDT/Xeon E5 socket

 

[JDG1980]

I would love to see the Xeon-D released as a "poor man's HEDT" CPU. It makes a lot of sense. If you only need 6 SATA ports rather than 10, and only dual-channel RAM rather than quad-channel, but still want 8 cores, it seems like this chip would be the ticket.

Well, if you need 8 cores, six SATA ports, and dual-channel RAM, the AMD FX platform can do that. It will be cheap (especially if you have a local Micro Center), but power consumption will be outrageous.

Actually, from the numbers, it looks like the Xeon D-1540 is going to have quite similar performance to the good(?) old AMD FX-8350. The D-1540 turbos to 2.6 GHz, and Haswell/Broadwell has about 60%-70% higher IPC than Vishera, so that's the equivalent of ~4.15-4.40 GHz Vishera. The FX-8350 factory turbos to 4.2 GHz. Where the Xeon solution really shines is energy efficiency; the whole SoC has a TDP of just 45W. The FX-8350 has an official TDP of 125W, but I saw a difference of 145W between idle and full Prime95 load on mine. Some of that could be the antiquated chipset, though - a burden the Xeon also doesn't have. The Xeon also has more modern connectivity, including two 10Gbps Ethernet ports.

http://www.anandtech.com/show/9070/intel-xeon-d-launched-14nm-broadwell-soc-for-enterprise

Of course, it will really be up to vendors like Asus and Gigabyte to determine if a consumer-focused Xeon-D solution is released. Intel isn't going to refuse to sell them chips if they pay for them. If a D-1540 board could be released for around the $499 price point, it would be a very attractive all-in-one system for many power users.

 

[Enigmoid]

Actually, from the numbers, it looks like the Xeon D-1540 is going to have quite similar performance to the good(?) old AMD FX-8350. The D-1540 turbos to 2.6 GHz, and Haswell/Broadwell has about 60%-70% higher IPC than Vishera, so that's the equivalent of ~4.15-4.40 GHz Vishera. The FX-8350 factory turbos to 4.2 GHz. Where the Xeon solution really shines is energy efficiency; the whole SoC has a TDP of just 45W. The FX-8350 has an official TDP of 125W, but I saw a difference of 145W between idle and full Prime95 load on mine. Some of that could be the antiquated chipset, though - a burden the Xeon also doesn't have. The Xeon also has more modern connectivity, including two 10Gbps Ethernet ports.

Of course, it will really be up to vendors like Asus and Gigabyte to determine if a consumer-focused Xeon-D solution is released. Intel isn't going to refuse to sell them chips if they pay for them. If a D-1540 board could be released for around the $499 price point, it would be a very attractive all-in-one system for many power users.

It looks like a 2.4 ghz (assuming the chip can maintain full turbo which it likely can under non AVX2 loads) 8 core chip with 5% IPC gain over Haswell so it will function like a 2.5 ghz haswell 8 core with HT. Probably about 25% slower than the 5960X and around a haswell hex core. All in all a good bit faster than the 8350 on MT with probably similar ST performance.

(In your comparison you do not include HT and the module penalty).

 

[cbn]

If a D-1540 board could be released for around the $499 price point, it would be a very attractive all-in-one system for many power users.

Laptops with dGPU and four 2.5" HDD/SSD are another thing to consider. (Yes, multi-drive laptops still exist).

 

[imported_ats]

Of course, it will really be up to vendors like Asus and Gigabyte to determine if a consumer-focused Xeon-D solution is released. Intel isn't going to refuse to sell them chips if they pay for them. If a D-1540 board could be released for around the $499 price point, it would be a very attractive all-in-one system for many power users.

You aren't going to see D-1540 boards at $499 as that's lower than the price point for the actual CPU. Right now the cheapest board is the Supermicro X10SDV-F at ~$780. Cheapest I can find for the X10SDV-TLN4F(2x10G) is $890. That $890 should collapse down to being very close to the non-10G version over the next couple of months (same chips in both cases).