Broadwell GT3 48EUs? TDP range 4.5W-47W

[mikk]

The Broadwell GPU will be based on 8th generation graphics architecture, and will have 20% more execution units than Haswell counterparts with the same GPU type. That means that Broadwell GT3 units will have 48 EUs. The graphics unit will support DirectX 11.1, OpenGL 4.2 and Open CL 1.2/2.0 APIs.

Thermal Design Power of Broadwell microprocessors will range from 4.5 Watt to 47 Watt. The "H" CPUs will have 47 Watt TDP, with some SKUs supporting 37 Watt configurable TDP. The "U" parts will have 15 Watt and 28 Watt TDP, and some SKUs will have lower 8.5 Watt and 23 Watt configurable TDP. The TDP of "Y" SoCs will be just 4.5 Watt.

http://www.cpu-world.com/news_2013/2013122001_Some_details_of_mobile_Broadwell_CPUs.html


48EUs for GT3 is exactly what Intel presented for the unknown GenX last year: http://www.highperformancegraphics.org/previous/www_2012/media/Hot3D/HPG2012_Hot3D_Intel.pdf


4.5W as default TDP for Y-Broadwell sounds nice, finally a proper suited tablet CPU from the Core line.

 

[jdubs03]

Hopefully there is no sacrifice in performance for the 4.5W SKU compared to Haswell. If there isn't, then that is a great option for $500 class 2-in-1s.

It has a to be the step up above Cherry Trail, so it'll be interesting to see the performance differential between products next fall. My guess is, it won't be that big of a difference at least in terms of multi-core performance. In fact I'd be very surprised if CT wasn't higher.

 

[bullzz]

I think we have to wait n see if its 4.5W TDP or SDP

 

[IntelUser2000]

I think we have to wait n see if its 4.5W TDP or SDP

The same article says:

4.5W TDP
3.5W TDPdown
2.8W SDP

Hopefully there is no sacrifice in performance for the 4.5W SKU compared to Haswell.

If anything, we can expect that only against Haswell Y parts, not U.

I'm guessing that 4.5W Broadwell parts will differentiate from Cherry Trail with higher performing graphics and single thread performance.

 

[jdubs03]

I should have clarified that I meant Haswell-Y. Single-core will be solid. From what I have seen Y-series only comes with GT3 graphics, so yeah way better performance than CT.

 

[ShintaiDK]

So the GT4 looks to be 96EUs with its 2Tflop+ performance.

And more chips will contain eDRAM. Hard times ahead for dGPU makers.

 

[mikk]

I should have clarified that I meant Haswell-Y. Single-core will be solid. From what I have seen Y-series only comes with GT3 graphics, so yeah way better performance than CT.

Y-series only with GT3 graphics? From where did you get this? Seems unrealistic, GT1 or GT2 seems more plausible.

So the GT4 looks to be 96EUs with its 2Tflop+ performance.

And more chips will contain eDRAM. Hard times ahead for dGPU makers.

If such a configuration will come, only as 65W model for AIO Desktop I think. There is no sign that it comes. I wouldn't bet on it. However given that 20%+ EU for GT1-GT3 coincides with the Genx slide, the Genx diagram could belong to Gen8. It looks vastly different, for example twice the samplers which should improve the efficiency and if a 96EU part reaches 2Tflop @1Ghz means GT3 could reach 1,2-1,3 Tflop assuming that it clocks with 1,2-1,3 Ghz. For comparison Kaveri's fastest GPU is rated with 737,3 Gflops and this is a 95W SKU.

 

[ShintaiDK]

If such a configuration will come, only as 65W model for AIO Desktop I think. There is no sign that it comes. I wouldn't bet on it. However given that 20%+ EU for GT1-GT3 coincides with the Genx slide, the Genx diagram could belong to Gen8. It looks vastly different, for example twice the samplers which should improve the efficiency and if a 96EU part reaches 2Tflop @1Ghz means GT3 could reach 1,2-1,3 Tflop assuming that it clocks with 1,2-1,3 Ghz. For comparison Kaveri's fastest GPU is rated with 737,3 Gflops and this is a 95W SKU.

No doubt AMDs iGPU oppotunity is completely lost with Broadwell. Its now something that works directly against them. The entire reason for buying ATI lost on the ground.

 

[jpiniero]

To be fair you won't see GT3 or GT4 on a Intel model anywhere near the prices that AMD APUs sell for.

 

[JoeRambo]

To be fair you won't see GT3 or GT4 on a Intel model anywhere near the prices that AMD APUs sell for.

Sadly it's more Intel margins than economic necessity. Intel parts will be smaller and more power efficient. If they can somehow get their drivers together, dGPU market will be in deep trouble.

 

[Ventanni]

AMD needs stacked DRAM like yesterday.

 

[jpiniero]

Sadly it's more Intel margins than economic necessity. Intel parts will be smaller and more power efficient. If they can somehow get their drivers together, dGPU market will be in deep trouble.

Really the only people who should even consider dGPU at this point is gamers and the extreme high end. It's got to be a very small market compared to even 5 years ago. Then again I don't think it's going to be easy to get gamers to ditch nVidia.

 

[mrmt]

To be fair you won't see GT3 or GT4 on a Intel model anywhere near the prices that AMD APUs sell for.

That is what Intel wants you to do. Spend more to get more performance.

 

[Lonbjerg]

So the GT4 looks to be 96EUs with its 2Tflop+ performance.

And more chips will contain eDRAM. Hard times ahead for dGPU makers.

Not as long as Intel's driver are in the state they are in

 

[Revolution 11]

Really the only people who should even consider dGPU at this point is gamers and the extreme high end. It's got to be a very small market compared to even 5 years ago. Then again I don't think it's going to be easy to get gamers to ditch nVidia.

It's been like that for a long time. Integrated graphics have been "good enough" for non-gamers for years now. Yet, Intel drivers are still spotty and I doubt Intel is putting as much focus on them as they should be.

 

[Piroko]

Really the only people who should even consider dGPU at this point is gamers and the extreme high end.

Which has been the status quo for just under 10 years, me thinks.

 

[jpiniero]

Which has been the status quo for just under 10 years, me thinks.

I'd really say until Sandy Bridge there were other reasons to get discrete. You couldn't really game at all and the dGPUs had extra features that were nice, like multiple displays and better video acceleration.

A 10 year old Intel GPU could barely do 2D.

 

[Piroko]

I'd really say until Sandy Bridge there were other reasons to get discrete. You couldn't really game at all and the dGPUs had extra features that were nice, like multiple displays and better video acceleration.

A 10 year old Intel GPU could barely do 2D.

Facebook games and flash were stuck at/only used as 2D anyways up until recently, so that point is rather moot. The overlap of casual PC users and multi display users is also minor at best and the need for proper video acceleration is a rather new problem, if you ignore the early adopters (who typically aren't all that casual either).

 

[Shivansps]

Lets see, Cherry Trail 16 EU

GT3 is 48EU, meaning GT2 is 24 and GT1 12?

Broadwell GT1 will have less EU than Cherry Trail but at a higher clocks?

 

[Sweepr]

Biggest news here is 4.5W TDP Broadwell-ULX. The most power-efficient Haswell CPU right now is a 4.5W SDP (11.5W TDP) chip.
Cherry Trail-T for 8-10'' Nexus/iPad competitors and Broadwell-ULX for premium fanless Windows tablets & convertibles. I like that.

 

[RU482]

the Y-SKU becomes a unique part with Broadwell. On Haswell and Ivy Bridge, it was just a binned U-SKU

 

[jpiniero]

Facebook games and flash were stuck at/only used as 2D anyways up until recently, so that point is rather moot. The overlap of casual PC users and multi display users is also minor at best and the need for proper video acceleration is a rather new problem, if you ignore the early adopters (who typically aren't all that casual either).

I guess my point is that even power users don't get that much of a benefit from the discrete GPU now. On the same token you have to question Intel's decision making regarding the development of GT3/GT4 if they are going to continue to charge more than what the GT2 + nVidia GPU costs.

 

[Fjodor2001]

So the GT4 looks to be 96EUs with its 2Tflop+ performance.

And more chips will contain eDRAM. Hard times ahead for dGPU makers.

I wonder how much eDRAM they intend to pair that 96EU chip with. If it's not sufficient and it'll often fall back to swapping out to DDR3 RAM (i.e. XBONE style), it'll be seriously memory bandwidth bottlenecked.

 

[ShintaiDK]

I wonder how much eDRAM they intend to pair that 96EU chip with. If it's not sufficient and it'll often fall back to swapping out to DDR3 RAM (i.e. XBONE style), it'll be seriously memory bandwidth bottlenecked.

The current 128MB they use is way too much, but was a case of better safe than sorry. So the answer is 128MB.

Intel didn’t see much benefit beyond a 32MB eDRAM however it wanted the design to be future proof. Intel doubled the size to deal with any increases in game complexity, and doubled it again just to be sure.

http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/3

 

[Fjodor2001]

The current 128MB they use is way too much, but was a case of better safe than sorry. So the answer is 128MB.

http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/3

Doesn't it depend on the use case how much is considered sufficient? I mean discrete GFX cards have way more than that (2+ GB GDDR5 RAM) for a reason I assume?