First look at Cherry Trail's IGP: 16 EUs (Gen 8), DX11.1/OpenGL 4.2 support

[Sweepr]

Four times more execution units than Bay Trail (which has 4 EUs - Gen 7), based on a newer Gen 8 graphics architecture that will also be used by Broadwell.

Up to 25.6GB/s to feed the newer GPU.

Source: http://vr-zone.com/articles/first-look-new-features-cherry-trails-gpu/64068.html

 

[ShintaiDK]

Wow, thats something like up to 5x BayTrail GPU performance.

14nm really showing its claws.

 

[Arkadrel]

That display power saveing technology sounds ingenious.

Its one of those.... why hasnt everyone already done that things.
It just makes so much sense, youd think everyone would have it.

 

[Vesku]

Nice to see the jump in graphics, doubt the <5W SoC market needs more than 4 CPU cores for at least the next couple of years. The power saving features are interesting but I don't think I'd want to be an early adopter of devices that are using them. Sounds like a lot of potential for software and hardware bugs that will need a few revisions.

That display power saveing technology sounds ingenious.

Its one of those.... why hasnt everyone already done that things.
It just makes so much sense, youd think everyone would have it.

Because most of them require more complexity in the display, will take a big company like Intel to push display makers to take on the extra cost.

 

[Arachnotronic]

Wow, thats something like up to 5x BayTrail GPU performance.

14nm really showing its claws.

Welp, looks like the last "complaint" from the Intel-haters has been addressed.

 

[frozentundra123456]

Looks promising. Let's hope it is a widely available part at a reasonable cost, and that the frequency hasn't been lowered because they have more AUs.

I was really expecting great improvement from has well, but it seems like the igp performance of most parts is only marginally improved. The huge improvement is basically limited to Apple.

 

[mikk]

Surely the new GPU king in the tablet space next year by a big margin. I think for gaming the CPU is a bigger problem with Cherry Trail.

 

[bullzz]

we still dont know the clock freqs. so calling it up to 5x BayTrail GPU performance is absurd. I am expecting it to be more like 2.5-3x performance. with qcomm and apple going crazy on adding GPU performance every year, where this will land is anybodys guess

 

[Sweepr]

Many X360/PS3 ports should run fine @ 720p with 16 EUs (Gen 8)... in a fanless tablet/2-in-1 form factor costing quite a bit less than Broadwell-ULT/ULX solutions.

 

[cbn]

Wow, thats something like up to 5x BayTrail GPU performance.

14nm really showing its claws.

That is impressive, but then I want to see how Intel spent the xtor budget on Moorefield.

According to Wikipedia (yeah, I know not the best source of info) Moorefield will also have Gen8 graphics--->http://en.wikipedia.org/wiki/Atom_(system_on_chip)#List_of_systems

 

[teejee]

That display power saveing technology sounds ingenious.

Its one of those.... why hasnt everyone already done that things.
It just makes so much sense, youd think everyone would have it.

LG is doing this today in G2.
Please remember that if a company is first with something then they will state that very explicitly on their marketing slides, otherwise they aren't first.

 

[cbn]

we still dont know the clock freqs. so calling it up to 5x BayTrail GPU performance is absurd. I am expecting it to be more like 2.5-3x performance. with qcomm and apple going crazy on adding GPU performance every year, where this will land is anybodys guess

Yes, it does look like the Apple A7 has a good amount of real estate dedicated to CPU and GPU:

And looking at the teardown of the iphone 5s here, it looks like Apple isn't that interested in integrating that many ICs into the SOC.

^^^ Lots of ICs everywhere, but packed more tightly together than other phone logic boards I have seen.


But I really wonder about Intel plans for Smartphone SOC. They have a node advantage and some of their OEM partners are not as technologically advanced in industrial design as Apple. So what Intel ends up doing will be very interesting.

Personally, I really don't need that much iGPU. I actually think being able to buy a nicely balanced and engineered device cheap is more interesting.

 

[jdubs03]

we still dont know the clock freqs. so calling it up to 5x BayTrail GPU performance is absurd. I am expecting it to be more like 2.5-3x performance. with qcomm and apple going crazy on adding GPU performance every year, where this will land is anybodys guess

I think we'll see a larger improvement than 3x. The EU units alone are 4x, add in a jump in the generation of graphics and 5x doesn't sound too unreasonable. I'm hoping Cherry Trail still gains quite a bit on the CPU side, they need to keep their advantage there too.

 

[escrow4]

You'd think at least 11.2 would be baked in.

 

[bullzz]

4x the EUs will not be 4x in performance. between haswell hd4600 and hd5000, there was 2x the EUs but performance increase for more like 20%

@Computer Bottleneck - apple can afford to have multiple ICs. other OEMs dont prefer that. intel themselves have mentioned they wont have 3G/LTE integrated until 2015. cherrytrail is clearly a tablet SOC and i wont expect any phones from it

 

[NTMBK]

I'm going to file this under "cautiously optimistic". I expect that this is a performance/W focused change, not just raw performance- go wider with lower clocks, and you can fit a bit more performance into the same envelope. (See HD4600 vs HD5000.) But I do expect a respectable improvement. Good to see Intel focusing on Bay Trail's one clear weak spot.

 

[mikk]

4x the EUs will not be 4x in performance. between haswell hd4600 and hd5000, there was 2x the EUs but performance increase for more like 20%

This comparison is invalid. HD4600 is a standard GPU while HD5000 a low power GPU. On bigger iGPUs the bandwidth hinders the scaling. And furthermore there are massive changes made on EUs in Gen8 which should help for better scaling.

 

[cbn]

@Computer Bottleneck - apple can afford to have multiple ICs. other OEMs dont prefer that. intel themselves have mentioned they wont have 3G/LTE integrated until 2015. cherrytrail is clearly a tablet SOC and i wont expect any phones from it

I thought Moorefield Merrifield (due in 2014) has the 3G/LTE integrated.

 

[Roland00Address]

If there are moving to 16 EU with atom, how much EU are they going to move up to with the core line?

Currently the smallest EU for 4th gen core dies is 10 EU with GT1 which is on the intel hd graphics (pentium, celeron)
Low End Core i3, i5s, i7s have 20 EU with GT2 and is called HD4200 to HD4600.
High end cores have GT3 which is 40 EU and go by the name HD5000, Iris 5100, or Iris 5200.

 

[mikk]

We don't have EU numbers for Broadwell yet. If the GenX from here belongs to Broadwell then GT1 should have 24 EUs.

 

[bullzz]

@mikk - I was comparing HD4400 to HD5000. I dont expect gen8 to deliver 2x the performance over gen7

@Computer Bottleneck - merrifield will not have integrated LTE

http://seekingalpha.com/article/1835192-intels-smartphone-efforts-accelerate-but-still-lag-qualcomms

 

[cbn]

@Computer Bottleneck - merrifield will not have integrated LTE

http://seekingalpha.com/article/1835192-intels-smartphone-efforts-accelerate-but-still-lag-qualcomms

Thanks. That is a great article.

 

[mikk]

@mikk - I was comparing HD4400 to HD5000. I dont expect gen8 to deliver 2x the performance over gen7

You compared this:

between haswell hd4600 and hd5000

The bigger iGPUs often doesn't scale well because of bandwidth bottlenecks. Because Bay Trail shouldn't really bandwidth bottlenecked with its <100 Gflops means it should make it easier for performance improvements. Serious bandwidth bottlenecks on iGPUs with Dualchannel DDR3 starts with 400-500+ Gflops.

 

[erunion]

This comparison is invalid. HD4600 is a standard GPU while HD5000 a low power GPU. On bigger iGPUs the bandwidth hinders the scaling. And furthermore there are massive changes made on EUs in Gen8 which should help for better scaling.

And he's comparing with in the same node. But for the Atoms its Four 22nm EUs versus sixteen 14nm EUs.

 

[mikk]

And he's comparing with in the same node. But for the Atoms its Four 22nm EUs versus sixteen 14nm EUs.

Process node has nothing to do with the EU scaling.