ivy bridge engineering sample igpu benchmarks

[nategator]

Assuming the benchmarks are accurate, then the article below is interesting as it provides actual gaming benchmark results. A couple of minor quibbles:

1. It would have been nice to see the fps for the more demanding games at low settings, just to understand the tradeoffs.

2. I with they would have also benchmarked the ib i5 with a llano chip instead of just the 2500k and a low end discrete.

My takeaways are the Ib igpu performs a little less than llano and can handle most pre 2010 games if not too demanding or maybe on low settings. Not too shabby for an Intel igpu

http://en.expreview.com/2012/02/19/ivy-bridge-core-i5-3570k-engineering-sample-test/21214.html

 

[gmaster456]

Still won't come close to llano for most games.

 

[nategator]

Well just looking at farcry 2 from this article it looks like the A8 puts out 34.96 fps at around the same settings. The ivy igpu is putting out 33.47 fps. I would say a 1.5 fps differential is close, no?

Apologies in advance if I am missing something.

http://hothardware.com/Reviews/AMD-Fusion-A83500M-ASeries-Llano-APU-Review/?page=10

 

[exar333]

Still won't come close to llano for most games.

Bias much?

If IB is around 65% faster than SB, it is in 'spitting distance' of the existing Llano. Based on AT's mobile review last year http://www.anandtech.com/show/4444/amd-llano-notebook-review-a-series-fusion-apu-a8-3500m/11, they put HD3000 at about 70-75% slower on medium settings than A8 Llano. The difference averaged-out to similar #'s on low-quality as well, but with more variability. Medium settings are probably the most realistic IMHO.

Keep in mind that an IB quad will be more between 50-100% faster in CPU tasks than the Llano as well, and be more power efficient. Thats a pretty good case for IB myself.

 

[shabby]

And in dirt3 it seems to get around 4fps more than llano, so it seems to be pretty even.

 

[exar333]

Well just looking at farcry 2 from this article it looks like the A8 puts out 34.96 fps at around the same settings. The ivy igpu is putting out 33.47 fps. I would say a 1.5 fps differential is close, no?

Apologies in advance if I am missing something.

http://hothardware.com/Reviews/AMD-Fusion-A83500M-ASeries-Llano-APU-Review/?page=10

Good link. Keep in mind that Llano also needs more $$$ memory vs. SB/IB to keep performance high.

 

[gmaster456]

depends on the llano cpu were talking about. Some have better gpu's than others. I was referring to the the the A8-3850 and 3870

 

[Idontcare]

Just goes to show that AMD can't be resting on their laurels with this HSA/Fusion stuff like they did when they had an upper-hand over Intel with their Athlon X2 vs. Prescott P4.

If AMD doesn't get its butt in gear and put itself on a more aggressive timetable for advancing its iGPU SKU's then it is on a path now in which Intel could well eclipse them in a matter of 3-4 yrs.

 

[waffleironhead]

Thats a pretty impressive gain by intel imo. Has anyone done an image quality comparison between nvidia/amd and intels gpus recently?

 

[nategator]

depends on the llano cpu were talking about. Some have better gpu's than others. I was referring to the the the A8-3850 and 3870

Fair enough, I could not find any 3850 3870 benchmarks running the same games at 720 p so I guess that is an open question.

For me this is all interesting as I was debating whether to try building an IB computer in april, burning through some older games and then upgrading to discrete in 6 months time or go with a 2500k build if IB IGPU was goingto suck for gaming almost as badly as SB. It looks like I may wait for Ivy unless I get a killer SB + discrete offer.

 

[nategator]

Just goes to show that AMD can't be resting on their laurels with this HSA/Fusion stuff like they did when they had an upper-hand over Intel with their Athlon X2 vs. Prescott P4.

If AMD doesn't get its butt in gear and put itself on a more aggressive timetable for advancing its iGPU SKU's then it is on a path now in which Intel could well eclipse them in a matter of 3-4 yrs.

Truth. It puts AMD's announcement that they may be going with arm as an indication of eventaul capitulation of x86 rather than a strategic shift for new markets.

 

[AtenRa]

Nice performance upgrade over SB but it still loosing from 2 and half year old low end GPU like the GT240.

 

[Lepton87]

AMD can't keep their performance advantage for long because they are ultimately limited by memory bandwidth, even now they are memory bound so a better iGPU won't do much. SBs gpu wasn't limited by memory bandwidth that's why it was relatively easy for Intel to increase its performance.

 

[exar333]

Nice performance upgrade over SB but it still loosing from 2 and half year old low end GPU like the GT240.

True, but the performance delta between the Intel IGP 2 years ago and a low-end option was HUGE. There is a ways to go, for sure, but it's come a long way from the IGPs of yesteryear.

 

[sawtx]

Well just looking at farcry 2 from this article it looks like the A8 puts out 34.96 fps at around the same settings. The ivy igpu is putting out 33.47 fps. I would say a 1.5 fps differential is close, no?

Apologies in advance if I am missing something.

http://hothardware.com/Reviews/AMD-Fusion-A83500M-ASeries-Llano-APU-Review/?page=10

Isn't that comparing the mobile Llano with a desktop Ivy Bridge?

 

[Vesku]

DDR4 should help keep more powerful IGPs fed, going to be a rough transition though as estimated 2013 DDR4 supply is supposedly ~5% of DDR3 volume.

So you are right that the APUs immediately after Trinity and IB have a good chance of being hobbled by designing for DDR3. Would explain the rumored heavy interest by Intel in the graphics side of Haswell, it's a great chance to match or beat the AMD APU competitor.

 

[nategator]

Isn't that comparing the mobile Llano with a desktop Ivy Bridge?

That´s a fair criticism. It was the best comparison between the two I could find based on game resolution and settings since the above benchmark never actually compared Llano to Ivy Bridge.

The point I was trying to figure out is wheter IB integrated are coming "close" to the performance of Llano´s APU, based on the benchmarks that are availble. Looking around, I could see a larger differential between the desktop Llano´s on the highest quality A8s for Starcraft II, which is why I believe people are saying that the IB IGs look more comprable to an A6 than an A8 on the desktop front.

A big caveat though is that until we have benchmark comparisons from a reputible source like Anandtech using as similar as possible rigs, all of this is speculative. While I think the evidence tilts more towards IB getting closer to Llano´s APU we´ll probably not know definitively until later.

Close is important because for a mainstream user choosing between the two within a tight budget, it may make more sense to buy the IB as it has a ton of headroom for a GPU upgrade in a couple of years rather than Llano, if you were looking to get into a system without a dedicated GPU. The laptop side is trickier since you have no easy upgrades but power consumption concerns are very important where IB should have an edge. This seems to be AMD vs. Intel´s battleground -- for the average user that wants to save as much money as possible but still have some light, Xbox 360-console-level gaming.

But who knows? Trinity may have a 65% improvement over Llano as well, returning us to the same cost/benefit calculus as SB vs. Llano. Guess we´ll find out :-D

 

[frozentundra123456]

AMD can't keep their performance advantage for long because they are ultimately limited by memory bandwidth, even now they are memory bound so a better iGPU won't do much. SBs gpu wasn't limited by memory bandwidth that's why it was relatively easy for Intel to increase its performance.

Why is Llano limited by memory bandwidth and Intel integrated graphics are not.
I thought they both had integrated memory controllers now. Guess I dont really understand what causes the limitation and why Intel is less memory constrained.

As far as IVB goes, I am glad to see the IGP improvement, although I wish the CPU was more improved also as regards to performance. The problem I see for intel is that as far as integrated graphics are concerned, they are still a generation behind, even best case if IVB can match Llano. But soon they will have to compete with Trinity. It will be interesting how Trinity competes with the new CPU architecture, and GPU wise as well. But I only see these APUs as useful in laptops. For a desktop, I would still go for an Intel CPU and a dedicated card, as even a low/midrange card will beat either AMDs or Intels integrated.

 

[IntelUser2000]

The point I was trying to figure out is wheter IB integrated are coming "close" to the performance of Llano´s APU, based on the benchmarks that are availble.

The early leaks are suggesting mobile Ivy Bridge parts(at least the high end quad core ones) perform ~15% better on graphics than desktop Ivy Bridge parts. Meanwhile, desktop Llanos perform 20% better than mobile Llanos.

In general, I think Ivy Bridge will do relatively better in mobile than desktops. Specifically, the current HD 3000 graphics don't perform as well in CPU intensive games like Starcraft and Civ V(they are also GPU demanding). Llano doesn't fare well in Starcraft either, but does relatively better on Civ V.

There's also a chance that Ivy Bridge will fare a bit better on weaker points, because the 2598 driver revision tested by Expreview is version 15.22, and official Ivy Bridge supports start at 15.26.

(Test with improved Sandy Bridge driver as an example: http://ht4u.net/reviews/2011/treibervergleich_intel_2361/index3.php)

Why is Llano limited by memory bandwidth and Intel integrated graphics are not.

That's because Sandy Bridge's iGPU is at a lower performance level, plus there's that L3 cache sharing(though the gains aren't massive, 15-20%).

 

[Idontcare]

Why is Llano limited by memory bandwidth and Intel integrated graphics are not.
I thought they both had integrated memory controllers now. Guess I dont really understand what causes the limitation and why Intel is less memory constrained.

I think it has something to do with the efficiency of the memory controllers and their ability to access a greater portion of the same theoretical max bandwidth provided by the dram operating at a given speed.

DDR3-1600 provides the same theoretical max bandwidth on both AMD and Intel, but on Intel they extract much more of that bandwidth for actual benefit of the program of interest if I remember correctly.

This makes the AMD need to use much higher bandwidth ram in order to obtain the same effective bandwidth for the integrated GPU.

 

[IntelUser2000]

I think it has something to do with the efficiency of the memory controllers and their ability to access a greater portion of the same theoretical max bandwidth provided by the dram operating at a given speed.

The thing I have not been able to find out is if that is true in GPU accesses as well. Memory bandwidth benchmarks are all about CPU related tests. AMD may be behind in CPU memory controllers, but they have loads of experience in graphics memory controllers.

 

[gevorg]

Very nice step forward in IGP performance, hopefully Haswell will have similar improvements. Clock is ticking for mobile GPUs.

 

[exar333]

I think it has something to do with the efficiency of the memory controllers and their ability to access a greater portion of the same theoretical max bandwidth provided by the dram operating at a given speed.

DDR3-1600 provides the same theoretical max bandwidth on both AMD and Intel, but on Intel they extract much more of that bandwidth for actual benefit of the program of interest if I remember correctly.

This makes the AMD need to use much higher bandwidth ram in order to obtain the same effective bandwidth for the integrated GPU.

I have seen a number of reviews that show the differences, but not the root cause here. Would definitely like to know more...

 

[bunnyfubbles]

this is where I think AMD might have made the mistake of going all in with BD. Had they just stayed course and made even only slight improvements to CPU performance, they could keep packing on the iGPU until we eventually got a CPU that is "good enough" for most people (certainly for gamers) and an iGPU that would finally be good enough for entry or even mid level gaming. Heck, Intel's own strategy doesn't have the mainstream moving beyond quadcores until sometime after Haswell...

granted, I think pretty significant problem to that solution is that we simply do not have enough apps that are GPU compute friendly, so outside of gaming there really isn't enough of a reason for an iGPU to be so powerful. Kind of a damned if you do, damned if you don't scenario for AMD :\

 

[AtenRa]

In SB, both CPU and iGPU cores share the L3 cache using a 256-bit/cycle ring bus.

http://www.youtube.com/watch?v=vliLjL4SSsA&feature=player_embedded
First minute of the video is about the integration of graphics on the CPU die and the LL (Last Level) Cache and the Ring bus.

The following slide is from A8-3850 review, Llano's iGPU is much faster in Street Fighter II than IVs at 1280x720. This is with 1600MHz memory.
http://translate.google.gr/translat...%87%CE%AF%CE%B4%CE%B1-%CF%84%CF%89%CE%BD-apus

This game is playable even at 1080p with Llano.

DDR-3 memory prices will continue to fall and higher frequency dimms will become cheaper in the near future. Trinity is supposed to be able to use 2133MHz memory vs 1866MHz in Llano.