Broadwell-E (X99) predictions

[Dave2150]

I wondered what you guys thought we can expect from Broadwell-E, when it launches in Q1 2016 (according to leaked roadmaps).

From what I've read so far on Broadwell, it provides a 5.5% IPC increase over Haswell. This 5.5% figure was described by Intel in their slides.

We know the 14nm process is capable of similar clockspeeds to 22nm, as demonstrated by 6700k (Skylake) retaining the 4790k's 4Ghz base clock.

IMO, Broadwell-E (for x99) will simply be clocked very similarly to Haswell-E, perhaps 100Mhz higher on all skews with the 5.5% IPC - so nothing special, but a small improvement over Haswell-E.

If anyone has any other thoughts please feel free to share them.

I'm going to be deciding between going ahead with Skylake (6700K) when it launches, or waiting for Broadwell-E. I'll more than likely go ahead with Skylake - if I was to wait for Broadwell-E, then I might as well wait for Skylake-E and I'd end up waiting forever

 

[happy medium]

I'm hoping for a 6 core, 5820k like cpu in a 100 watt package with slightly faster clocks. With an easy 4.5 overclock. Is that asking to much?

 

[Justinbaileyman]

Are these going to be LGA2011 v3 drop in compatible? If so I wont be waiting for skylake liked I planned and will do my upgrade now. I thought x99 was a dead end though and what cpu's are available is what we were going to be left with?

 

[Ajay]

Are these going to be LGA2011 v3 drop in compatible? If so I wont be waiting for skylake liked I planned and will do my upgrade now. I thought x99 was a dead end though and what cpu's are available is what we were going to be left with?

They should be, with a BIOS update. Plus, it's logical, as it provides Intel server customers with an opportunity to upgrade and recieve the benefits of lower power draw and lower cooling costs (Since HEDT processors are somewhat neutered 'EP' server CPUs).

 

[ShintaiDK]

From a stock perspective I expect significant frequency changes on the X model. Something like +500Mhz base and turbo. And the focus until more cores will be to bring stock performance up to pair.

 

[Ajay]

From a stock perspective I expect significant frequency changes on the X model. Something like +500Mhz base and turbo. And the focus until more cores will be to bring stock performance up to pair.

That would make a lot of sense for BDW-E and bring HEDT users closer to parity with SKL ST performance.

With SKL-E, the base "EP" CPU will likely have more cores, so I would expect at least 8 cores to be the min for SKL-E processors. It's also likely based on Intel's recent Server CPU roadmap that SKL-E won't be out till 2H17 (and that's just a guess since 1S CPUs are still 'in planning').

 

[FalseChristian]

Probably a 10-20% increase in performance if the last 4 generations are an indication. I'm still sticking with my"old" i5 2500K @ 4.635GHz. The GPUs are our main concern.

 

[Makaveli]

I'm hoping for a 6 core, 5820k like cpu in a 100 watt package with slightly faster clocks. With an easy 4.5 overclock. Is that asking to much?

Yes that is not realistic

 

[cbn]

I think the top SKU will remain eight core.

The middle SKU will remain hexcore with 40 PCIe lanes.

The lowest SKU will be hexcore with 28 PCIe lanes.

Clocks will be higher though.

 

[happy medium]

Yes that is not realistic

The 100 watt part? I thought they were gonna be 14nm?

 

[Makaveli]

The 100 watt part? I thought they were gonna be 14nm?

The 100 watt part.

I don't think going from 22nm to 14nm will allow for a 40 watt TPD drop just from going to a smaller nm while keeping the same core count if you want reduced clocks speeds then maybe.

 

[tenks]

I was holding out for Haswell-E/Broadwell-E but the schedule got so messed up. I personally think unless you invested in Haswell-E very early, Broadwell-E no longer makes sense if you're starting from scratch.

Broadwell-E is the last x99 chip and it's already late. Skylake/Skylake-E is bringing a bigger jump in performance than usual and will be on entirely different platform. x109 or whatever they call it to match the DT 100 series chipset. Why invest in a dead platform with no upgrade path when Skylake-E is likely 2-3 quarters away?

I mean this is all moot if you don't care about your platform upgrade path, and you have money to spend now.

I was going to invest in x99 with a 5820k and then was going to sell it and upgrade to a higher end broadwell-e sku down the line. Now, with the proximity to Skylake, not so much.

 

[Ajay]

I was holding out for Haswell-E/Broadwell-E but the schedule got so messed up. I personally think unless you invested in Haswell-E very early, Broadwell-E no longer makes sense if you're starting from scratch.

My upgrade schedule got messed up, so going in on BDW-E in ~9 months will probably work out just fine for me. SKL-E is probably 2H17 and my system is already getting pretty long in the tooth (and just don't think I could go back to a quad core with SKL).

Add in a 14/16FF based GPU and I'll be happy as a clam

 

[DooKey]

First quarter 2016 sucks. I want it now!!!

 

[tenks]

My upgrade schedule got messed up, so going in on BDW-E in ~9 months will probably work out just fine for me. SKL-E is probably 2H17 and my system is already getting pretty long in the tooth (and just don't think I could go back to a quad core with SKL).

Add in a 14/16FF based GPU and I'll be happy as a clam

Broadwell-E is a 2016 product. Skylake-E is a 2016 product. Why buy broadwell?

Brand new roadmaps just showed SKL-EX in 2H'17. EX skus usually hit 9-12 months after the EP/E skus. This puts SKL-E around Q4'16. 3 qtrs after broadwell-e.

x99 will be a dead in the water platform by the time Broadwell-E ships.

edit: if I sound bitter, I'm sorry its not directed at you. I'm just hating on Intel right now for their management of the HEDT platform.

 

[pj-]

Broadwell-E is a 2016 product. Skylake-E is a 2016 product. Why buy broadwell?

Brand new roadmaps just showed SKL-EX in 2H'17. EX skus usually hit 9-12 months after the EP/E skus. This puts SKL-E around Q4'16. 3 qtrs after broadwell-e.

x99 will be a dead in the water platform by the time Broadwell-E ships.

edit: if I sound bitter, I'm sorry its not directed at you. I'm just hating on Intel right now for their management of the HEDT platform.

Who says Skylake-E is 2016? Edit: Q4 2016 is not 2-3 quarters as you said in your previous post, more like 6, and that's if there are no delays.

 

[Makaveli]

My upgrade schedule got messed up, so going in on BDW-E in ~9 months will probably work out just fine for me. SKL-E is probably 2H17 and my system is already getting pretty long in the tooth (and just don't think I could go back to a quad core with SKL).

Add in a 14/16FF based GPU and I'll be happy as a clam

I'm in the same boat as we have similar current builds.

I was not impressed by the limited DMI interface and high price of DDR4 at the launch of X99. The 14nm cpu won't fix the DMI interface DDR4 pricing is alittle better now but still high.

So the next question is can I survive on this build with maybe just a GPU upgrade until skylake.

 

[Ajay]

Broadwell-E is a 2016 product. Skylake-E is a 2016 product. Why buy broadwell?

Brand new roadmaps just showed SKL-EX in 2H'17. EX skus usually hit 9-12 months after the EP/E skus. This puts SKL-E around Q4'16. 3 qtrs after broadwell-e.

x99 will be a dead in the water platform by the time Broadwell-E ships.

edit: if I sound bitter, I'm sorry its not directed at you. I'm just hating on Intel right now for their management of the HEDT platform.

Well, it worse. The single server CPUs (1S) are still 'in planning' on the roadmap. EP/EX platforms are merging and are 'in development'. That would seem to mean the SKL-E will be out at or later than SKL-EP/EX. So SKL-E currently looks like a 2H17 product at best :|

 

[tenks]

I'm telling you right now, Skylake-EX with 28 cores, will not be released before Skylake-E with 8-10 cores.

Skylake-E will hitting in 2H'17 is 1 1/2 years after broadwell-e. Not going to happen. Logic says 12 months or sooner after Broadwell-E. Look at the DT, broadwell and skylake are months apart. You think this is going to change for the HEDT skus?

Who says Skylake-E is 2016? Edit: Q4 2016 is not 2-3 quarters as you said in your previous post, more like 6, and that's if there are no delays.

You know thre are 4 quarters in a year right? Broadwell-E is slated to drop in Q1'16. I said Skylake-E in Q4'16. That's a difference of 3 quarters. Dunno where you got 6.

 

[Ajay]

I'm telling you right now, Skylake-EX with 28 cores, will not be released before Skylake-E with 8-10 cores.

Skylake-E will hitting in 2H'17 is 1 1/2 years after broadwell-e. Not going to happen. Logic says 12 months or sooner after Broadwell-E. Look at the DT, broadwell and skylake are months apart. You think this is going to change for the HEDT skus?

These are slides from another thread posted by mikk:

Basin Falls is the 1S workstation platform that HEDT CPUs/Chipsets come from. It lists it in 2017 and it's still in the planning stage. I'm not making anything up or guessing. This is the Intel Roadmap.

 

[cbn]

tenks,

Ajay is right. But that doesn't mean Intel couldn't launch a Skylake Xeon-D much earlier than Skylake-E and then make a higher power version of that called Skylake-D for consumers.

Main distinguishing factor for such a "Skylake-D" would be the likely limit of eight cores and dual channel memory. While someone buying X99 around the same time would have up to 24 Broadwell cores to choose from and quad channel memory.

 

[cbn]

Based on the March 2015 release of Broadwell Xeon-D, my guess for a Skylake Xeon-D would be March 2016.

 

[IntelUser2000]

tenks,

Ajay is right. But that doesn't mean Intel couldn't launch a Skylake Xeon-D much earlier than Skylake-E and then make a higher power version of that called Skylake-D for consumers.

Actually, no.

Two reasons:
1. If the Xeon D is using an "SoC" process(since its classified in roadmaps as an SoC) with inherently lower clock speeds it'll end up useless as a HEDT part.
2. We say "it doesn't make sense" when Intel segments SKUs. Here's the thing though:

"Makes sense" WRT to a customer: companies selling a product with most features enabled at the lowest prices possible

"Makes sense" WRT to a company: selling it with least features enabled at the price that will maximize the amount of revenue

One thing I can guarantee for "financially smart" companies. They won't make products that will cannibalize their product lines.

 

[cbn]

1. If the Xeon D is using an "SoC" process(since its classified in roadmaps as an SoC) with inherently lower clock speeds it'll end up useless as a HEDT part

As I mentioned in other thread here, Broadwell measured 6.93 mm2 for both Core M and Broadwell U (non blocked off portion).

Furthermore when I look very carefully at both cores, the part not blocked off on Broadwell U matches up exactly to the fully exposed part on Broadwell Core M:

Therefore my belief is that both Core M and Broadwell U processors use the same size core and that the extra boxed in area around the core in the Broadwell U processor die map is a mistake. (More support for that reasoning in this post)

So if Xeon-D uses the same process tech as Broadwell U I have to assume it will clock well enough.

P.S. When we get the processor die map for i7-5775C I will be certain to measure the core size to see how it compares to these measurements.

 

[cbn]

"Makes sense" WRT to a customer: companies selling a product with most features enabled at the lowest prices possible

"Makes sense" WRT to a company: selling it with least features enabled at the price that will maximize the amount of revenue

One thing I can guarantee for "financially smart" companies. They won't make products that will cannibalize their product lines.

Yes, but the Broadwell Xeon-D is pretty lean (160mm2 for the 14nm processor die which includes two 10GbE LAN and the usb 3.0 ). I think the only thing holding it back now is overlap with certain low power Haswell Xeon E5 parts.

P.S. The SoC also includes an on package PCH (excluded from the 160mm2 measurement) which has the usb 2.0 and the SATA.