Intel Demonstrates 65W Broadwell-K Socketed Processors at GDC 2015

[csbin]

http://wccftech.com/intel-demonstra...015-features-iris-pro-igpu-launching-mid2015/

The surprising thing is that Intel will be bringing only quad core chips to the socketed platform which will adopt the Broadwell core architecture that is based on a 14nm process technology. The Quad Core parts which have a unlocked design, feature compatibility with Z97/H97 chipset based motherboards with the LGA1150 socket and will have a TDP of just 65W. The new Broadwell-K series are supposed to feature four Broadwell x86 cores and faster Iris Pro graphics chips with eDRAM cache on board (128MB L4 Crystalwell), that is Intel’s flagship graphics processing unit designed to date integrating 48 execution units. The specific models are not known but Intel will be using the Core-5000 series branding for the new chips when they are available in the market.

The statement from Intel might hint to why they are pushing for the 65W TDP on Broadwell-K since they are trying to aim them towards Mini PCs, NUCs and desktop All-in-ones. PCPerspective managed to grab some shots off a demo rig running an Broadwell-K CPU on EVGA’s Z97 Classified motherboard using the standard Intel heatsink. Now technical bits were revealed but Intel did demo the new chip with Futuremark’s latest DirectX 12 benchmark. Since Broadwell-K will exist alongside Skylake processors which launch in August-October 2015, it is expected that Broadwell-K will stick to small form factor devices that require low temps and have smaller power requirements while Skylake-K 95W parts will be shipping towards the more performance demanding PCs. Broadwell generation of processors can still be used as normal socketed parts for overclocking and personal uses on the LGA 1150 motherboards that launched in early 2014.

 

[dave_the_nerd]

So... basically an unlocked desktop i5 with Iris Pro graphics and a giant-as-hell L4 cache?

Killer HTPC chip. I'd definitely consider it.

 

[witeken]

Why isn't BDW-K being released earlier? It should be launched in early Q2.

 

[III-V]

Why isn't BDW-K being released earlier? It should be launched in early Q2.

Parametric yield issues, if that rumor is correct.

65 W is a pretty hefty reduction in TDP. Bet I could predict its clocks within a decent range of accuracy.

 

[witeken]

65 W is a pretty hefty reduction in TDP. Bet I could predict its clocks within a decent range of accuracy.

-25%, a bit less than the IDF'13 demo.

 

[ShintaiDK]

Why isn't BDW-K being released earlier? It should be launched in early Q2.

Because mobile gets first dips on all 14nm.

 

[frozentundra123456]

I get releasing a part with Iris Pro at a lower TDP for NUC, SFF, etc, but what I dont get is why it is unlocked, especially with Skylake 95 watt on the horizon.

I think Broadwell K will be a niche product, the appeal will depend on price. If they release it at the 200.00 price point it would be appealing for some uses. If they charge a significant premium over haswell quads because of Iris pro, the the appeal goes down rapidly.

Actually, I would rather see a mainstream quad mobile chip with iris pro instead of the desktop.

 

[.vodka]

So, Broadwell is +5% per clock over Haswell, and will probably include the 128MB L4 cache as standard, even for the K parts? That's another 0-5-10% improvement depending on what you're doing. Could be an interesting proposal for those who already have a z97 board. Rated at 65w, it should overclock nicely, if the 14nm mobile first process provides for this.


Skylake will then launch shortly after, right? It'll be interesting to see what happens when the dust settles.

 

[SPBHM]

that's good news, as long as the price is not to high...

I was always curious to see an unlocked Iris Pro CPU

 

[AtenRa]

If they release it at the 200.00 price point it would be appealing for some uses. If they charge a significant premium over haswell quads because of Iris pro, the the appeal goes down rapidly.

14nm = more expensive
lower yields = more expensive
large die = more expensive
large L4 = more expensive

I dont think we should expecting less than $300, unless Intell doesn't care about margins anymore.

 

[Gikaseixas]

HTPC use? Yes I want one please

 

[Fjodor2001]

So is this what Intel had to do to keep the promise of an 1150 upgrade path to Broadwell? I.e. release a single SKU to not break it.

 

[erunion]

Parametric yield issues, if that rumor is correct.

65 W is a pretty hefty reduction in TDP. Bet I could predict its clocks within a decent range of accuracy.

Its not a TDP reduction. i7-4770R is already 65W, Intel has targeted desktop Iris Pro at SFF.

I get releasing a part with Iris Pro at a lower TDP for NUC, SFF, etc, but what I dont get is why it is unlocked, especially with Skylake 95 watt on the horizon.

When Intel launched Devils canyon they announced Broadwell Iris Pro would be unlocked and socketed, which was not the original plan but 14nm delay obviously gave them time to make changes. I can only speculate but I think its due to an increasing interest in overclocking by intel's in general (as seen in devils canyon and unlocked Pentiums). Iris Pro is a premium product (in price and in features) so should be unlocked.

 

[Idontcare]

Parametric yield issues, if that rumor is correct.

65 W is a pretty hefty reduction in TDP. Bet I could predict its clocks within a decent range of accuracy.

-25%, a bit less than the IDF'13 demo.

The drop in TDP is basically in-line with what we'd expect if Intel did a dumb-shrink* of haswell and kept the clockspeeds unchanged. (84W -> 65W)

We'd expect about a 25% reduction in power consumption (which could then be rolled into a 25% TDP reduction for the platform) for 14nm over 22nm at the socket level.

We know the chips themselves are quite delayed at this point in terms of coming to market. Haswell refresh was a stop-gap to deal with that.

But I'm wondering if they have a heat-density issue preventing them from pushing too hard on >65W SKUs? Just speculation on my part, but in my testing with 22nm IB versus 32nm SB it was the operating temperatures of the IB that really limited the clocks on a stock config 3770K (unless you delidded of course, but even then...).

So it could simply be the case that were Intel to clock broadwell high enough such that it consumed enough power to warrant a 95W (or 88W) TDP rating then the operating temperatures themselves would be well in excess of 110C or some other way-too-high temperature. Just thinking out loud here, no industry rumors or sources behind this conjecture.

* dumb-shrinks don't really happen anymore, but you get my meaning, same architecture, same circuit designs processing the same instructions, but using less power for the sake of those 14nm xtors vs. the 22n xtors in Haswell

 

[Shivansps]

TDP does not mean anything in a K part, a motherboard that can overlock a K processor can also ajust the TDP value to wharever you like.

 

[frozentundra123456]

True, but the only place a big honking igpu and edram makes sense is in a SFF, Brix, type device, and they are thermally constrained.

IDC could be right, thermals might not be so good, so they keep the tdp down, but allow overclocking, since someone who wants to overclock will probably use aftermarket cooling anyway.

 

[jpiniero]

The 4770R is 3.2/3.9. So based upon what's happened so far, the i7 Broadwell-K is probably 3.4 or 3.5 base with 3.8 turbo. Maybe overclock it to 4.1 or 4.2.

 

[crashtech]

Does anyone think this 65W Broadwell-K can eclipse the performance of a mildly OCed 4790K? Can the IGP be disabled for more thermal headroom?

 

[Shivansps]

Does anyone think this 65W Broadwell-K can eclipse the performance of a mildly OCed 4790K? Can the IGP be disabled for more thermal headroom?

Better yet, go intro the bios and set the TDP to 200W.

 

[frozentundra123456]

Better yet, go intro the bios and set the TDP to 200W.

And hook up a liquid nitrogen cooler.

Wonder how the igpu will overclock. No matter what, stock, the igpu is going to still be the limiting factor rather than the cpu.

 

[III-V]

But I'm wondering if they have a heat-density issue preventing them from pushing too hard on >65W SKUs? Just speculation on my part, but in my testing with 22nm IB versus 32nm SB it was the operating temperatures of the IB that really limited the clocks on a stock config 3770K (unless you delidded of course, but even then...).

Isn't heat density just a symptom, the underlying problem being that electrical parameters are not keeping up with density increases?

I've seen it commonly speculated that 22 nm did not overclock as well as 32 nm because of heat density issues, but heat density has been increasing for quite some period of time now. I'm skeptical of the idea that it's only just now become significant roadblock. Certainly it is the nature of exponentially growing "bad transistor behaviors" to be a non-issue one node, and a severe headache the next, like gate oxide leakage at ~130 nm. However, you'd think that the credible authorities would have highlighted that heat density is, in fact, what's holding us back, rather than whimsically pointing out that we've got heat density on par with nuclear fission reactors, and doing nothing more than that.

So is this what Intel had to do to keep the promise of an 1150 upgrade path to Broadwell? I.e. release a single SKU to not break it.

Where was that promise made? Also, who says this is a single SKU?

 

[tenks]

Anandtech is saying this isn't the high end k i7 chip, more a low power S variant.

 

[III-V]

Anandtech is saying this isn't the high end k i7 chip, more a low power S variant.

I think they're ill-informed. There was an i5 and i7 GT3e Haswell SKU, with a TDP of 65W.

 

[JDG1980]

It will be interesting to see how the Iris Pro 6200 integrated GPU performs. Its predecesor, the 5200, wasn't quite as good as the Nvidia GT 650M on most games (though it did surprisingly well in GPGPU). All indications are that the 6200 still won't be able to play most AAA titles at 1080p, or max out settings in madVR for HTPC users. The real bottleneck is still the slow memory subsystem. Since modular GDDR5 was apparently cancelled some time ago, that leaves stacked memory on-package (HBM) as the solution. A good all-in-one chip would include four or more decent x86 cores, a GPU at least as good as Pitcairn, and 8GB of stacked HBM. Basically something like the XB1/PS4 APU, but with desktop-class CPU cores and RAM onboard instead of in separate BGA packages. Imagine a Mini-ITX board with all of this already soldered in - all you have to do to complete the system is put it in a case, plug in a PSU, add an SSD, and install Windows.

 

[III-V]

though it did surprisingly well in GPGPU

Is it really all that surprising? Intel's "strength" has always been with GPGPU -- they've had a relative dearth of texturing and pixel filling performance, though.