Ivybridge should match LLano in graphics

[IntelUser2000]

The test in Metro is at 1024X768 at minimal settings...
At slightly higher resolution or settings, the HD3000
would collapse more than a 5450, wich by the way
has only 80 stream processors compared to as much
as 400 in Llano or in a 5570.....

So IB will be about three times less performant
than Llano even if Intel manage 50% perfs increase,
wich would be equivalent to 120 SPs.....

How do you get the idea that 120SPs would be 50% faster than 80SP HD5450 when 400SP 5570 is only 2.5-3x faster than HD5450?

http://www.anandtech.com/show/4263/amds-radeon-hd-6450-uvd3-meets-htpc/1

If we look at HD 6450, while it achieves 2x the performance over HD 5450 in some games, about half the games achieve 50-60% gain and it needed 2x the shaders, 15% increase in core clock, and 2.25x the memory bandwidth to achieve that.

Plus: http://www.anandtech.com/show/4084/intels-sandy-bridge-
upheaval-in-the-mobile-landscape/5

7% advantage over HD5450 in low, 13.9% advantage in medium.

 

[Abwx]

How do you get the idea that 120SPs would be 50% faster than 80SP HD5450 when 400SP 5570 is only 2.5-3x faster than HD5450?

http://www.anandtech.com/show/4263/amds-radeon-hd-6450-uvd3-meets-htpc/1

If we look at HD 6450, while it achieves 2x the performance over HD 5450 in some games, about half the games achieve 50-60% gain and it needed 2x the shaders, 15% increase in core clock, and 2.25x the memory bandwidth to achieve that.

Plus: http://www.anandtech.com/show/4084/intels-sandy-bridge-
upheaval-in-the-mobile-landscape/5

7% advantage over HD5450 in low, 13.9% advantage in medium.

This doesnt unfortunately change the whole picture...
Thank you for the links , wich show by the way that
a HD3000 is slightly better than a 80sp 5450 but still
below a 160 SPs 6450.

Given that Intel s increasing the EU count from 12 to 16,
plus a higher frequency, we can safely assumethat IB graphics
will be at best at a 160 SPs 6450 level.

True that it s significantly more than the 120 i first
estimated...🙂

 

[IntelUser2000]

Given that Intel s increasing the EU count from 12 to 16,
plus a higher frequency, we can safely assumethat IB graphics
will be at best at a 160 SPs 6450 level.

True that it s significantly more than the 120 i first
estimated...🙂

Now notice nowhere in this thread I've speculated how IVB's graphics will perform. In a general sense, you will probably turn out to be right. IVB's graphics might not beat Llano. But to see how it will really perform, we need to look at the details.

-Ivy Bridge's GPU being 30% faster than Sandy Bridge's graphics is very likely not an Intel statement, but a speculation from the source that leaked such info. 16 is approximately 30% greater than 12, so people who seen that must have concluded it would be 30% faster.

Unfortunately, the leakers do not have a brain capable about knowing something called a diminishing return. 33% increase in shaders alone have never resulted in 33% increase in performance, or even close to that. By the increase in EUs alone, I'm thinking we'll only see 20%. Now that's not a conclusion of how I think IVB's graphics will perform.

-Llano's graphics. Are you sure it'll perform like the 5570? How do you know it won't perform 10% less because of its peculiarities(eg. sharing the memory with the CPU)? Will it perform like the 5570 on Phenom II, or 5570 on Core i7? Overclocked best case scenario 5570, or reference 5570? History of integrated graphics suggests that the same graphics core using system memory resulted in 10-15% less performance than a dedicated equivalent. That's with all the fancy marketing acronyms added making it sound like the integrated version has the technology to perform even better than the dedicated memory version.

-Performance with specific games. You can have two products trading equal punches to each other but depending on the preference of the reader they will point out on benchmark that performs especially worse with one product and say whatever they endorse wins. Now with such a different architecture and direction we will have such case with Ivy Bridge and Llano's GPU.

 

[ydnas7]

since when has the performance of Intels IGP been primarily connected to either process or EU count increase? these may be the key drivers for ATI and Nvidia, but for the last 3 or so generations (1 per year) Intel's IGP performance has increased way beyond what the EU count or process technology would indicate. Intels IGP did come from a very low base, but they are still develooping - the addition of post processing on IB could provide free AA to the IGP (think Intel encoder/decoder quicksync)

 

[Vesku]

Keep in mind that the 400 shader Llano is the top tier part, Ivybridge should be much closer to the low-mid tier Llano's in graphics while clearly beating them in cpu strength.

Regardless of which approach is best, I really welcome this push for better IGP by the manufacturers. As has been stated, if we to wait on business demand to force better IGP we'd have to wait probably another 5-10 years for a Llano or Ivybridge.

 

[Dark Shroud]

This might sound simple but I just like the fact that IGPs are now upgradeable. You buy a higher end mobo that you’re going to stick with for a few years you don’t have to worry about the IGP just sitting there on the board not getting upgraded with the rest of the system.

 

[Abwx]

IVB's graphics might not beat Llano. But to see how it will really perform, we need to look at the details.

-Ivy Bridge's GPU being 30% faster than Sandy Bridge's graphics is very likely not an Intel statement, but a speculation from the source that leaked such info. 16 is approximately 30% greater than 12, so people who seen that must have concluded it would be 30% faster.

Unfortunately, the leakers do not have a brain capable about knowing something called a diminishing return. 33% increase in shaders alone have never resulted in 33% increase in performance, or even close to that. By the increase in EUs alone, I'm thinking we'll only see 20%. Now that's not a conclusion of how I think IVB's graphics will perform..

Intel only stated 30% more EU, without any perfs expectations.
We can assume that with a 20/25% higher frequency; IB graphics
will be close to 50% better perfs than SB Gfx.


Btw, not sure , but seems theses plateforms
will not be compatibles with IB :

- Intel Z68 Express
- Intel P67 Express
- Intel H67 Express
- Intel H61 Express
http://www.hardware.fr/news/lire/14-04-2011/

-Llano's graphics. Are you sure it'll perform like the 5570? How do you know it won't perform 10% less because of its peculiarities(eg. sharing the memory with the CPU)? Will it perform like the 5570 on Phenom II, or 5570 on Core i7? Overclocked best case scenario 5570, or reference 5570? History of integrated graphics suggests that the same graphics core using system memory resulted in 10-15% less performance than a dedicated equivalent. That's with all the fancy marketing acronyms added making it sound like the integrated version has the technology to perform even better than the dedicated memory version..

According to a slide from AMD :

"Even the same sized GPU is substancially more effective
in this configuration" (edit : Integrated in Llano).

Traditional chipset integrated and discrete graphic
cards graphics have :

"3 times less bandwith from main memory to GPU.
and serious latency due to chip crossings"

So what is lost at a end is gained at the other one, and overall,
Llano GPU should be on par with its discrete siblings.

 

[Tuna-Fish]

So what is lost at a end is gained at the other one, and overall,
Llano GPU should be on par with its discrete siblings.

Perhaps in AMD's marketing. Being closer to the CPU helps make state transitions and shuffling vertex arrays around cheaper, but most of the memory bandwidth that a scene needs is texture fetches and rendering. Reducing the distance to the CPU doesn't make those any cheaper. The highest-end Llanos will likely be very bandwidth-limited, and very considerably slower than the same amount of units running at the same speed with a nice dedicated memory bus.

It will of course still be much faster than anything Intel can muster.

 

[podspi]

Perhaps in AMD's marketing. Being closer to the CPU helps make state transitions and shuffling vertex arrays around cheaper, but most of the memory bandwidth that a scene needs is texture fetches and rendering. Reducing the distance to the CPU doesn't make those any cheaper. The highest-end Llanos will likely be very bandwidth-limited, and very considerably slower than the same amount of units running at the same speed with a nice dedicated memory bus.

It will of course still be much faster than anything Intel can muster.

For rendering perhaps, but I think for GPGPU Llano will be faster than the GPU component would suggest. I think this is what AMD was referring to. If they are talking about rendering (not sure which document/slide we're talking about) then you have to assume that there is some benchmark they are talking about 😀

 

[Abwx]

Perhaps in AMD's marketing. Being closer to the CPU helps make state transitions and shuffling vertex arrays around cheaper, but most of the memory bandwidth that a scene needs is texture fetches and rendering. Reducing the distance to the CPU doesn't make those any cheaper. The highest-end Llanos will likely be very bandwidth-limited, and very considerably slower than the same amount of units running at the same speed with a nice dedicated memory bus.

It will of course still be much faster than anything Intel can muster.

http://forum.hardware.fr/hfr/Hardwa...rocesseurs-llano-sujet_868949_34.htm#t7822246

 

[grimpr]

For rendering perhaps, but I think for GPGPU Llano will be faster than the GPU component would suggest. I think this is what AMD was referring to. If they are talking about rendering (not sure which document/slide we're talking about) then you have to assume that there is some benchmark they are talking about 😀

Wont be long for the Folding@Home ATI OpenCL Client...when its out and works as good it will be interesting to watch out some FAH Llano setups with discreet gpus at the cost of a 2600k.

 

[daveybrat]

I personally don't care how fast Intel's onboard graphics are, i would rather have AMD's superior drivers.

In games, AMD is more compatible and tweaked than Intel's graphics drivers are. Intel has gotten better, but i don't expect AMD or Nvidia like performance nor have i ever expected it.

 

[grimpr]

I personally don't care how fast Intel's onboard graphics are, i would rather have AMD's superior drivers.

In games, AMD is more compatible and tweaked than Intel's graphics drivers are. Intel has gotten better, but i don't expect AMD or Nvidia like performance nor have i ever expected it.

I suspect the multithreaded Catalysts will be out on Bulldozer launch, 11.6 or something.

 

[SickBeast]

AMD's Fusion is a huge disappointment in my books. Seeing the Zacate cores getting trashed by the new i3 is really bad IMO. AMD bought an entire company so they could do this stuff, and Intel just hires a bunch of engineers and does it better. It's got to be demoralizing for them.

 

[grimpr]

Indeed, i just wish xbitlabs makes a miniatx am3+ Bulldozer vs Atom review too or a Llano against Atom, that would make a great comparison too using 1200W psus.

 

[podspi]

AMD's Fusion is a huge disappointment in my books. Seeing the Zacate cores getting trashed by the new i3 is really bad IMO. AMD bought an entire company so they could do this stuff, and Intel just hires a bunch of engineers and does it better. It's got to be demoralizing for them.

Zacate's x86 cores were never meant to compete with anything Sandy Bridge-derived. They were intended to compete with Atom, and they do that very well. Zacate's IGP is not terribly impressive, but this is more AMD's cleverness than anything else. Given the CPU performance Zacate is capable of, making the IGP any more powerful (especially given the amount of pixels needed to be pushed on smaller screens) would have been a waste of die space.


At the same time, the i7 2600K's (a $300 CPU, I can get a Brazos system for that!) IGP can't do DX11 or OpenCL. So I would say that Fusion hasn't been a failure. Now, if Llano can't beat the 2600K in gaming (IGP only of course) then I will be the first to agree with you.

Of course, if Ivy Bridge's IGP turns out to be faster than Trinity's (I'm still not convinced IB's IGP will equal Llano) then that was certainly $5b wasted.

 

[dma0991]

I don't see how the test done by X-bit labs is a fair comparison considering a search from Newegg for the Core i3 2100T + Zotac H67ITX would result in a total of $295 just for a HTPC setup while the AMD side with MSI being the cheapest and most power efficient E-350 around is only $115-140.

That is less than 50% of the price of the SB HTPC and I am not surprised if the performance would even double on the SB setup compared to the E-350. They should have used products from a similar price point to compare.

 

[Ovven]

Xbitlabs review is quite amusing, seeing how it compares a 18w zacate to 2x more wattage part, which besides consuming a lot more power is a lot more expensive as well. Way to fail AMD 🙂

 

[Accord99]

I don't see how the test done by X-bit labs is a fair comparison considering a search from Newegg for the Core i3 2100T + Zotac H67ITX would result in a total of $295 just for a HTPC setup while the AMD side with MSI being the cheapest and most power efficient E-350 around is only $115-140.

This isn't meant to be a fair comparison. It's more of a demonstration of what kind of CPU performance is available at or near Atom or Fusion levels of power use.

That is less than 50% of the price of the SB HTPC and I am not surprised if the performance would even double on the SB setup compared to the E-350. They should have used products from a similar price point to compare.

CPU difference is more like 3X-4X so the 2100T has a performance/$ advantage.

 

[Ovven]

It's more of a demonstration of what kind of CPU performance is available at or near Atom or Fusion levels of power use.

Except that intel cpu used 2x more power in that review (don't take xbitlabs' graphs at straight value, because they used an inefficient gigabyte motherboard over MSI, which increased over power consumption by 5-6wats. Xbit acknowledged that MSI is a lot more efficient than gigabyte but didn't include those results in a review titled "Every Watt Counts" 😀 )

 

[dma0991]

This isn't meant to be a fair comparison. It's more of a demonstration of what kind of CPU performance is available at or near Atom or Fusion levels of power use.
CPU difference is more like 3X-4X so the 2100T has a performance/$ advantage.

Indeed the power consumption was near that of a Zacate but the SB is on a 32nm process while the Zacate is on the 40nm. As of now the Zacate is still an immature product and Bobcat cores are not something to shout about and without the GPU it is not even what it is now. I'll wait for the 28nm Krishna and see how it goes.

The performance on the Core i3 2100T is 3-4X but considering that this is a HTPC that does simple tasks I don't see the need for the extra power. I'd rather see the 2100T coupled up with a discrete GPU to make a gaming miniITX rather than a HTPC.

 

[IntelUser2000]

Indeed, i just wish xbitlabs makes a miniatx am3+ Bulldozer vs Atom review too or a Llano against Atom, that would make a great comparison too using 1200W psus.

This is very true. You are way better off using a low power voltage version of a regular desktop CPU than a Netbook/Tablet CPU just popped into a desktop motherboard. The latter uses much more power everywhere so the point of using an ultra low power CPU is not there.

BTW, the 880W PS is what they use as their bench platform. True, they made a fault there.

 

[grimpr]

Speak of the devil.

 

[Idontcare]

Speak of the devil.

I am so ready for Llano laptops...please come soon!

 

[podspi]

^ It's the hybrid crossfire that really excites me. I wish we could get a release date...