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

[Khato]

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

Nice find. Of course my favorite part is definitely page 2 with that list of quite notable accomplishments... and then there was the i740, lol! That is definitely Tom.

That said, don't get your hopes up based off the GenX configuration shown there. It's more about showing how the scaling is supposed to work than something you're going to see any time soon.

 

[erunion]

Process node has nothing to do with the EU scaling.

In a TDP limited environment it does.

 

[mikk]

In a TDP limited environment it does.

Because of possibly higher GPU frequencies. It has nothing to do with the EU scaling itself. A same or similar GPU frequency is a must have requirement, otherwise you won't find out how goood the EUs scale.

 

[Borealis7]

dynamic refresh rate for video? has Intel implemented a form of GSynch?? i feel a patent lawsuit coming from nVidia...

 

[Haserath]

So what do they mean by "multiple refresh rate panels required?"

Why can't one panel handle it?

Then there's also the self-panel refresh they talked about before. They'd also aim for a lower framerate self-refreshing panel?

Intel needs to get into the panel business to ever have a chance at implementing all this tech.

 

[USER8000]

Intel needs to get into the panel business to ever have a chance at implementing all this tech.

They need to panel a motion at the next investors meeting!

 

[Homeles]

So what do they mean by "multiple refresh rate panels required?"

Why can't one panel handle it?

That's not what they're saying. They're saying it requires an LCD panel that supports multiple refresh rates.

 

[Arachnotronic]

So what do they mean by "multiple refresh rate panels required?"

Why can't one panel handle it?

Then there's also the self-panel refresh they talked about before. They'd also aim for a lower framerate self-refreshing panel?

Intel needs to get into the panel business to ever have a chance at implementing all this tech.

Parse carefully.

"Multiple refresh rate panels" ->

(Multiple refresh rate) panels required.

 

[Arachnotronic]

dynamic refresh rate for video? has Intel implemented a form of GSynch?? i feel a patent lawsuit coming from nVidia...

Intel has access to all of NVIDIA's patents until 2017.

 

[Homeles]

dynamic refresh rate for video? has Intel implemented a form of GSynch?? i feel a patent lawsuit coming from nVidia...

Support for dynamic refresh isn't Nvidia-exclusive. The part that Nvidia's profiting from, other than the bonus of having Kepler support it, is that they're selling display ASICs.

Nvidia is hoping that companies like Intel support dynamic refresh, because they want to sell those display ASICs to OEMs. Right now, their solution isn't exactly cost, power, and area effective for tablets and such, but it could be in the future.

 

[Borealis7]

Intel has access to all of NVIDIA's patents until 2017.

bummer, i was SO looking forward to the drama

 

[Ajay]

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.

^^THIS. Especially since Intel is sticking with a 64b memory bus. We don't have enough info to do any math here since we don't know anything about the Gen8 EUs, but I doubt Intel would have committed so much more to the C-T IGP for anything less than a 2X improvement. Anything above that would be a nice surprise

 

[ShintaiDK]

^^THIS. Especially since Intel is sticking with a 64b memory bus. We don't have enough info to do any math here since we don't know anything about the Gen8 EUs, but I doubt Intel would have committed so much more to the C-T IGP for anything less than a 2X improvement. Anything above that would be a nice surprise

64bit bus? Both BT and CT can use 128bit memory bus. Hence also why you see 17 and 25GB/sec.

 

[monstercameron]

64bit bus? Both BT and CT can use 128bit memory bus. Hence also why you see 17 and 25GB/sec.

maybe call it ChT instead of CT...

 

[Arachnotronic]

^^THIS. Especially since Intel is sticking with a 64b memory bus. We don't have enough info to do any math here since we don't know anything about the Gen8 EUs, but I doubt Intel would have committed so much more to the C-T IGP for anything less than a 2X improvement. Anything above that would be a nice surprise

Intel's tablet chips use 2x64 bit memory busses. You must be thinking of the ARMy.

 

[Ajay]

64bit bus? Both BT and CT can use 128bit memory bus. Hence also why you see 17 and 25GB/sec.

Intel's tablet chips use 2x64 bit memory busses. You must be thinking of the ARMy.

My bad, I just looked at the data width and didn't notice the channel/rank row in the chart. So, that's a nice jump in bandwidth and the CT gfx performance should scale higher. The best thing, I think, is that BT and especially CT should have better IQ than their ARM competitors. That won't be as big a deal on phones, but will be a definite plus on tablets, IMHO.

I'm going to BB today, so I'm going to annoy my wife and check out at least one BT tablet, I'm very interested to see what the real life performance is like.

 

[liahos1]

what are the chances we get real merrifield and cherrytrail information at the intel analyst day this Thursday? Maybe this information is legit given how close we are to some new presentations from INTEL management.

 

[Blitzvogel]

Intel's tablet chips use 2x64 bit memory busses. You must be thinking of the ARMy.

And AMD is still using 64 with Kabini and future Beema (AFAIK).

And 16 EUs?

Beema isn't looking too good..........

 

[jdubs03]

And AMD is still using 64 with Kabini and future Beema (AFAIK).

And 16 EUs?

Beema isn't looking too good..........

beema should be compared to haswell/broadwelll, whereas mullins is the competitor to baytrail/cherrytrail.

i think the update for amd is definitely a solid one,, however it will be competing against late cycle products from intel. when amd releases those products broadwell and ct will be right around the corner. and even then they're still behind aside from the gpu advantage of mullins and beema?

 

[monstercameron]

beema should be compared to haswell/broadwelll, whereas mullins is the competitor to baytrail/cherrytrail.

i think the update for amd is definitely a solid one,, however it will be competing against late cycle products from intel. when amd releases those products broadwell and ct will be right around the corner. and even then they're still behind aside from the gpu advantage of mullins and beema?

I dont really believe in the gpu advantage on the low end anymore...Intel has clearly overtaken AMD in ulv/ulp gpu performance...

 

[mikk]

Except for tablet Atom and other Atom based SoC, but with 16 EUs and a major GPU redesign on 14nm they should overtake AMD easily in a year.

 

[IntelUser2000]

Support for dynamic refresh isn't Nvidia-exclusive. The part that Nvidia's profiting from, other than the bonus of having Kepler support it, is that they're selling display ASICs.

They are not new technologies for Intel. Dynamic Refresh Rate Switching is shown as available with Bay Trail and its available even further back with Haswell.

I find this hilarious:

That display power saveing technology sounds ingenious.

It says DPST SIX! That means its in its sixth generation. That tech was available way far back, I think the first one was back with the 90nm Pentium M back in 2004!

 

[IntelUser2000]

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

24 EU is GT2, because a slice has 24 EUs and GT1 is "slice half". That follows the 20/33% increase per generation they had since GMA X3000.

8/10/12/16/20/24?

I really do not see 16 EU as leadership graphics. Much lower clocks and TDP limits will hold it back compared to having the same 16 EUs in the Core line. At best I see 3x. Leaders like Qualcomm are going to be not far behind. Also, who knows what Apple will do?

You can see the difference between a 850MHz 20EU with Core and 667MHz 4 EU with Bay Trail is only 2x:
-http://www.notebookcheck.net/Intel-HD-Graphics-Bay-Trail.103037.0.html
-http://www.notebookcheck.net/Intel-HD-Graphics-4200.93336.0.html

 

[Shivansps]

That will mean that Cherry Trail will have more EU than GT1, i dont think so.
GT1 are likely 16EU, and if its indeed the half as Intel has been doing, that means GT2 is 32EU.

20/40 is also possible, but there is just no way for GT1 to have less EU than Cherry Trail.

 

[IntelUser2000]

That will mean that Cherry Trail will have more EU than GT1, i dont think so.
GT1 are likely 16EU, and if its indeed the half as Intel has been doing, that means GT2 is 32EU.

There's no reason to suggest that they have a rule saying GT1 can't be less than Atom IGPs.

For instance, they can use it on Celerons and Pentiums that have better CPU, but less GPU. They have so many SKUs because they fit every gap that can be filled by the competition(AMD, Nvidia, etc).

Also, we may see segregation of Atom GPUs on Cherry Trail, with top of the line being 16EU and lower end having less than that.