Ivybridge should match LLano in graphics

[Mopetar]

Not sure how easy that will be for Intel. They've certainly been able to get better performance, but they're not doing it very efficiently.

Here are some charts from a recent Techreport review:

Performance:

Power Consumption:

The idle measurements were taken at the Windows desktop with the Aero theme enabled. The cards were tested under load running Bulletstorm at a 1440x900 resolution.

We haven't seen numbers for Llano yet, so it's hard to judge how well it performs in comparison, but for graphically intensive workloads, I can't see Intel doing better than AMD, even with IB.

 

[Edrick]

IGP wont stop growing with APUs of the future,.. mainly because of GPGPU computeing (theres no loss in continued GPU growth).

I am under the impression that the IGPs on current products can not be used for GPGPU computing. I asked this question when SB first came out.

Hopefully, this will change in the not so distant future.

 

[Riek]

This is SWAG (stuff we all get), right?

If I had any idea I needed to include a bevy of obvious stuff that needs to be incorporated to go along with an increase in SP's and EU's (higher bandwidth DDR, more local cache, yaddi yadda) then I would have definitely increased the word-count on my posts.

If Llano is already "very heavily memory-limited" then I would question why they pursued increasing the SP's to the point they did on an intentionally lower-cost product.

"No" improvement, your words not mine, beyond the existing SP count would imply they fully saturated the diminishing returns curve...for a lower-cost product surely they could cut the SP's by a healthy 25% or more, have a product that costs all the less, and not lose but a few percentage points of performance (if even that much, diminishing return curves have pretty long tails, and you claim they are all the way out at the end of it already).

Ofcourse the ondie gpu is not only for gaming which is the source of the memory starvation (textures), so if they use it for calculations the 400sp might have a benifit. How many SP are needed to support all the video features from AMD?

 

[Arkadrel]

I am under the impression that the IGPs on current products can not be used for GPGPU computing. I asked this question when SB first came out.

Hopefully, this will change in the not so dustant future.

Thats only true for Intel IGPs.... AMDs APU is all about the GPGPU, thats why they put the GPU in with the CPU (they claim they removed a bottleneck) and allow for better GPGPU computeing this way.

Hopefully, this will change in the not so dustant future.

Yeap... Intel is bound to follow suit.

 

[Phynaz]

And AMD already has plans for Llanos replacment with BD cores as well.

I'm not aware of Bulldozer being a replacement for Llano. Did I miss something?

 

[Edrick]

Thats only true for Intel IGPs.... AMDs APU is all about the GPGPU, thats why they put the GPU in with the CPU (they claim they removed a bottleneck) and allow for better GPGPU computeing this way.

Interesting. I was not aware. This has me excited about the direction they are headed.

 

[piesquared]

http://www.xbitlabs.com/news/video/...ics_Performance_than_Core_i_Sandy_Bridge.html

this is comparing a low freq / cut down Sandybridge 6EU with LLano
boost the Intel IGP to 2600k specs and LLano is perhaps 50% faster than SB
but IB is coming 6 months after LLano, it has 20% increase due to process, just add an additional 4 EU (from 12 to 16 - just a wild guess) would make it equivalent to LLano for graphics. Add any more EU and it is beyond LLano.

How about an apples to apples comparison at equal image quality, which most reviewers curiously ignore....
The most glaring fail is that sandy bridge doesn't support dx11, making comparisons at the premium level impossible. Getting ivy bridge to the same speed, at the same image quality, with the same driver support, with the same standards and cutting edge techniques, is a pipe dream for anyone that thinks it'll compete at the same level AMD and NV do.

 

[alexruiz]

When the Fusion concept started, it was rumored AMD sets target at Radeon HD4650 / 4670 levels. Then it was mentioned 5570 (still 4670 level). Only a few more weeks to find out

I know one of the desktops will be upgraded. I will be in the market for a new laptop, but what I don't know yet (and will know once llano is out) is if I'll get portability and battery life with a bobcat, or graphics power with llano. Just a few more weeks to find out.

 

[Abwx]

I'm not aware of Bulldozer being a replacement for Llano. Did I miss something?

http://www.techspot.com/news/41076-amd-unveils-2011-2012-roadmap-for-servers-desktops-laptops.html

 

[Vette73]

I'm not aware of Bulldozer being a replacement for Llano. Did I miss something?

Not replacing llano, but the BD cores will be put into the next gen llano called trinity.
The current llano has the k10 cores.

 

[Arkadrel]

Interesting. I was not aware. This has me excited about the direction they are headed.

^this

it would be fun if people that benchmark the Llano, tried out software that uses GPGPU as well, and see if its faster than the SandyBridge in those.

Id be disappointed if reviewers didnt, I mean its a big apart of the feature of the APU.

I also hope reviewers mention 3D bluray playback, with the Llano... vs SandyBridge (not sure if it can do this).

 

[Chiropteran]

Intel has a lot of work to do in the driver department before I can take their IGP seriously. Even if it offers more raw performance, driver issues make it undesirable.

 

[formulav8]

I can see no way IB will match AMD's latest and greatest igp being released shortly. Last I heard IB is only up to 30%-35% faster than SB. Even with Intel mooching off nvidia's ip (Intel wouldn't stand any kind of chance otherwise), they will still be quite a ways behind the curve when it comes to raw gpu performance, GPGPU performance, driver quality.

Also I didn't see anything saying IB will match llano. Did I look over it or something I guess?

 

[iCyborg]

^this

it would be fun if people that benchmark the Llano, tried out software that uses GPGPU as well, and see if its faster than the SandyBridge in those.

Id be disappointed if reviewers didnt, I mean its a big apart of the feature of the APU.

I also hope reviewers mention 3D bluray playback, with the Llano... vs SandyBridge (not sure if it can do this).

Yes, this is something AMD will be pushing aggressively, I've seen some marketing slides regarding this. PCI-e transfer to and back from GPU is a major bottleneck that limits GPGPU to very specific applications, and having CPU and GPU sharing and accessing the same memory will open the door to more uses. They need to leverage from this since they're in a good position here with Intel still lagging behind AMD/nVidia in performance, drivers, OpenCl support etc, while nVidia only has discrete (they have Tegra 2 for ARM though). At least right now, Intel sure will be looking to catch up.

 

[busydude]

Ivy Bridge in 1H- 2012 according to this leaked slide.

 

[jones377]

It's all about how many transistors Intel wants to dedicate to the graphics part of IB. For example if they stay at 4 CPU cores, they could put all the extra transistors afforded by the 32nm->22nm shrink into the graphics EUs. With that they could easily quadrouple it's size. The successor to Llano will still be on 32nm so that will limit the amount of improvements for the graphics part. AMD could use VLIW4 shaders but those aren't really better perf/mm2 which is what the Intel 22nm shrink will enable.

Of course other considerations come into play, such as memory bandwidth. Intel would have the same problems as AMD Llano with only 2 channels of DDR3. Then there is that rumor of Intel going to use stacked DRAM on the processor package... That could be a game-changer entirely

 

[RobertPters77]

I hope that Ivy is compatible with Sandy. 36 eu =/= 300~ radeon cores.

 

[podspi]

Intel's IGP does seem to be pretty small... Compare the number of transistors between the HD3000 and the 5450, it's 114 million to 292 million respectively.

I'm not sure if this is because of lower IQ, lack of compute power, the result of Intel's IGP being clocked higher, or some combination of all of that.


I don't think AMD needs to play with stacked memory yet... If need be they can add it on package much easier, I think.

 

[Vette73]

^

The 5450 has its own memory controller and cache. The intel one shares all that. So you can't compare on a 1 to 1 design like that.

 

[nonameo]

Intel's IGP does seem to be pretty small... Compare the number of transistors between the HD3000 and the 5450, it's 114 million to 292 million respectively.

I'm not sure if this is because of lower IQ, lack of compute power, the result of Intel's IGP being clocked higher, or some combination of all of that.


I don't think AMD needs to play with stacked memory yet... If need be they can add it on package much easier, I think.

Does that number include the memory controller? SB shares a mem controller with the CPU and GPU right?

 

[jimbo75]

Intel's IGP does seem to be pretty small... Compare the number of transistors between the HD3000 and the 5450, it's 114 million to 292 million respectively.

I'm not sure if this is because of lower IQ, lack of compute power, the result of Intel's IGP being clocked higher, or some combination of all of that.


I don't think AMD needs to play with stacked memory yet... If need be they can add it on package much easier, I think.

115 million? That seems high to me, where did you see that figure?

 

[Arkaign]

If the thing can't match the performance of a GTS450 or 5770, it's kind of worthless just like existing onboard video is. Useful only for *very* light 3d gaming, but mainly just good for general desktop use and hd video.

This is kind of a moot point imo. Those that care about decent video will just get something with a real credible gpu, and those same people who care will probably want the faster cpu to begin with, whether it be SB, IB, or BD/etc.

Those that don't care (or know) all that much about performance, just play farmville, watch netflix and whatever, it won't really matter to them.

 

[podspi]

The 5450 has its own memory controller and cache. The intel one shares all that. So you can't compare on a 1 to 1 design like that.

Whoops, forgot about that! I guess I should be looking at Bobcat...

115 million? That seems high to me, where did you see that figure?

From Anandtech's Sandy Bridge Review.

 

[IntelUser2000]

Not sure how easy that will be for Intel. They've certainly been able to get better performance, but they're not doing it very efficiently.

I've looked at few reviews: http://www.legionhardware.com/artic..._2500k_and_core_i7_2600k_sandy_bridge,14.html
http://www.silentpcreview.com/article1139-page4.html
http://www.xbitlabs.com/articles/video/display/intel-hd-graphics-2000-3000_10.html

Something seems wrong. The other reviews are showing far lower usage.

-3D Blu Ray is supported but they still need the ecosystem going. I think they are still trying to get the glasses part working. For now it works with passive, later active.
-Some GPGPU compute is supported on the HD Graphics in Sandy Bridge. Like DX11 compute.

 

[jimbo75]

I've looked at few reviews: http://www.legionhardware.com/artic..._2500k_and_core_i7_2600k_sandy_bridge,14.html
http://www.silentpcreview.com/article1139-page4.html
http://www.xbitlabs.com/articles/video/display/intel-hd-graphics-2000-3000_10.html

Something seems wrong. The other reviews are showing far lower usage.

Not really. The first one shows the 2600K pulling 57 extra Watts, the 2500K pulling 49 extra. That's compared to 58 Watts extra in the techreport review, which is at the same kind of level.

Whatever way you look at it, the HD 3000 is capable of pulling 60 Watts+ probably, while giving out less than 20 Watts gpu performance.