• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Predictions thread: How large will Ontario's IGP be?

How many stream processors do you think AMD will include with Ontario APU?

  • 40

  • 80

  • 120

  • 160

  • 200

  • 240

  • Greater than 240


Results are only viewable after voting.
The other questions are what clockspeed and what memory bandwidth will the IGP operate with.

Given the process node (TSMC bulk 40nm), the targeted TDP (<20W), and rumored die-size (~100mm2) the IGP has got to be rather underwhelming in terms of circuitry and targeted clockspeeds.

Llano's IGP is the real question since there are so many unknowns.
 
The other questions are what clockspeed and what memory bandwidth will the IGP operate with.

Yep, all we know is that it will use DDR3. (I am guessing one DIMM as well which would be half the total system bandwidth of Llano).
 
80, I expect it to be an integrated 5450

edit: the 5450 has a TDP of 19.1 watts max, so take off the RAM from that, replace it with a low power CPU ala atom, and there you have it.
 
Last edited:
Yep, all we know is that it will use DDR3. (I am guessing one DIMM as well which would be half the total system bandwidth of Llano).

prolly low-speed low-Vdimm DDR3 at that, since it is targeting low-power

Are we sure llano will only be dual-channel? Given the bandwidth needs/benefits of the IGP, if there was ever a case to be made for triple and quad channel memory controllers on consumer platforms...
 
80, I expect it to be an integrated 5450

edit: the 5450 has a TDP of 19.1 watts max, so take off the RAM from that, replace it with a low power CPU ala atom, and there you have it.

http://www.amd.com/US/PRODUCTS/NOTEBOOK/GRAPHICS/ATI-MOBILITY-HD-5400/Pages/hd-5430-specs.aspx

http://www.amd.com/US/PRODUCTS/NOTEBOOK/GRAPHICS/ATI-MOBILITY-HD-5400/Pages/hd-5470-specs.aspx

I noticed the TDP varies by a quite a large amount for these 80 stream processor mobility chips.

In the first link we have the 5430 mobility radeon at 550 Mhz core clock and 1.6 Gbps GDDR3 memory for 7 watt TDP.

In the second link the 5470 mobility radeon with the 750 Mhz core clock and 32.2 Gbps GDDR5 scales to 15 watts TDP.
 
http://www.amd.com/US/PRODUCTS/NOTEBOOK/GRAPHICS/ATI-MOBILITY-HD-5400/Pages/hd-5430-specs.aspx

http://www.amd.com/US/PRODUCTS/NOTEBOOK/GRAPHICS/ATI-MOBILITY-HD-5400/Pages/hd-5470-specs.aspx

I noticed the TDP varies by a quite a large amount for these 80 stream processor mobility chips.

In the first link we have the 5430 mobility radeon at 550 Mhz core clock and 1.6 Gbps GDDR3 memory for 7 watt TDP.

In the second link the 5470 mobility radeon with the 750 Mhz core clock and 32.2 Gbps GDDR5 scales to 15 watts TDP.

hmm yeah you're right... there is a lot of variation there. I don't think we're going to see anything resembling resembling a redwood core though. It's going to be something closer to cedar, and both of those you linked to are cedars.
 
hmm yeah you're right... there is a lot of variation there. I don't think we're going to see anything resembling resembling a redwood core though. It's going to be something closer to cedar, and both of those you linked to are cedars.

I don't think we will see a Redwood core either. Juniper (800 stream processors) is 180mm2 on 40nm. So half that would be ~90mm2. That sounds just too big if the entire APU is only 100mm2.

EDIT: The 100mm2 reference I made here is incorrect (i'm not sure where I got that from).

According to the Anandtech Bulldozer and Bobcat article we are looking for less than 60mm2 with dual core Bobcat.

http://www.anandtech.com/show/2872

AMD provided very little detail here other than it delivers 90&#37; of today&#8217;s mainstream performance in less than half of the silicon area. If AMD views mainstream as an Athlon II X2, then Bobcat would deliver 90&#37; of that performance in a die area of less than 60mm^2.
 
Last edited:
It'll be OVER NINE THOUSAND!

But seriously, somewhere between 40 and 80 (up to and including 80, imho).
 
I have an extremely limited understanding of this but, can a CPU and GPU an APU run at same clock speeds, while consuming low power(<20W)? Is it practically feasible? I am talking in the range of 1.4-1.6Ghz(Ontario's rumored speed range).
 
I have an extremely limited understanding of this but, can a CPU and GPU an APU run at same clock speeds, while consuming low power(<20W)? Is it practically feasible? I am talking in the range of 1.4-1.6Ghz(Ontario's rumored speed range).

Pretty unlikely. It's debatable its even reachable using 32nm SOI for the lower bound estimate(1.4GHz). 40nm TSMC which is same as current graphics cores?

The limitation isn't just power, it has to do with how GPUs are designed. Though its clear it won't do such low TDP at such high clocks anyway.

I don't think we will see a Redwood core either. Juniper (800 stream processors) is 180mm2 on 40nm. So half that would be ~90mm2. That sounds just too big if the entire APU is only 100mm2.

I estimated 80SPs, but maybe we'll see even less. The 80SP 5450 is at 59mm2. The I/O starts to dominate in space with lower end GPUs.
 
40nm TSMC which is same as current graphics cores?

Good question. We know TSMC is/was working on a 40nm node variant designed for really high performance so they could fab SUN's 40nm generation CPU's. (very much like glofo's 32nm SOI+HKMG for AMD and not for anyone else)

Not clear whether Oracle will pursue that now but it is possible that AMD signed up to use that same sub-node for their 40nm Fusion product. It would not be the best for low-power though, SUN's requirements never prioritized that.

Alternatively AMD could be using TSMC's low-power 40nm sub-node which is a notch lower in power and clockspeeds for their bobcat stuff. If that is the case then we can expect low-power chips but the clockspeeds are really going to be quite low, like Atom.
 
That was actually a rhetorical question, but yea I don't think they'll use HP because low power would be a priority. Clock speeds on the GPU won't be affected much if its bound by the architecture.
 
Last edited:
psssst

some of the values you chose are impossible. each SIMD has 16 shaders with 5 alu's for 80sp's. you have to have multiples of 80.
 
psssst

some of the values you chose are impossible. each SIMD has 16 shaders with 5 alu's for 80sp's. you have to have multiples of 80.

Dividing by zero is impossible, but this isn't one of those situations. They could pull something similar to Southern Islands and do 4 ALU config.

How about cache?

Or is the CPU's cache inaccessible to the GPU?

That depends. Does Ontario have something that allows cache sharing? Like for example a common interface or a router. If it has 512KB L2, it might be bad for the CPU to share it with the GPU.

As for area, the CPU portion on Atom Pineview takes ~20mm2. The GPU will have to be pretty compact to fit in under 75mm2 with a dual core Bobcat.
 
I am leaning toward 40 as this is made for the netbook crowd, not laptop. The current IGP is 40 and its abilities are still heads above what Intel has for the netbook.

Now what may make sense is build it at 80 but turn off half for power/heat needs. For larger netbooks/HTPC type devices then turn everything on.
 
Dividing by zero is impossible, but this isn't one of those situations. They could pull something similar to Southern Islands and do 4 ALU config.
then it will be multiples of 64. also i'm pretty sure they announced that 1st gen fusion graphics would be evergreen based.
 
As for area, the CPU portion on Atom Pineview takes ~20mm2. The GPU will have to be pretty compact to fit in under 75mm2 with a dual core Bobcat.

Maybe Ontario will be larger than 100mm2? Wasn't the quote about less than 50 percent die size, but 90 percent performance pertaining to just the cpu cores?

Regarding the GPU to CPU ratio I found these Anandtech charts rather interesting.

So I am interested to see what AMD chooses as the GPU/CPU sweetspot for this chip.

I am leaning toward 40 as this is made for the netbook crowd, not laptop. The current IGP is 40 and its abilities are still heads above what Intel has for the netbook.

Now what may make sense is build it at 80 but turn off half for power/heat needs. For larger netbooks/HTPC type devices then turn everything on.

I see your point about netbooks. If basic needs are really light, what kind of benefits could ARM coupled to a smaller GPU produce? For example, I would think being able to connect such a device to a 1080p monitor (for watching video) would be pretty nice. But how large of an IGP is really needed for that?

That being said I am really interested to see what kind of productivity applications AMD has planned for these relatively powerful Ontario devices.
 
Last edited:
Back
Top