[THG]Core i7-4770K: Haswell's Performance-Previewed

[lamedude]

It took until Windows8 for the Windows Photo Viewer to use SSE2 and if Intel keeps playing their games I wouldn't be surprised if took another 12 years for it use AVX2.

 

[ShintaiDK]

Regardless that was my question.
For the internal code each APP obviously needs to be compiled with AVX2 flags in mind.


...but the general overhead of the OS\Driver layer does play in effect.
(Hello GPU's magicly increasing performance by drivers?)

Was just wondering - if AVX2 might cause some root execution of kernelcode\comdrivers better in a way that youd feel or syntheticly see greater performance with AVX2\Haswell in a IPC alike case on Windows.

(no just AVX2 - but the extra port as well, Haswell in a nutshell).

The GPU ecosystem works differently.

You got OpenGL/DX code that goes to the driver and gets translated there into whatever language they use.

Applications goes directly to the hardware so to say. Their translation happend during compile.

 

[NTMBK]

Regardless that was my question.
For the internal code each APP obviously needs to be compiled with AVX2 flags in mind.


...but the general overhead of the OS\Driver layer does play in effect.
(Hello GPU's magicly increasing performance by drivers?)

Was just wondering - if AVX2 might cause some root execution of kernelcode\comdrivers better in a way that youd feel or syntheticly see greater performance with AVX2\Haswell in a IPC alike case on Windows.

(no just AVX2 - but the extra port as well, Haswell in a nutshell).

As ShintaiDK said, GPU code is compiled every time the game is loaded up. (This is how code written once in a DirectX shader can run on literally hundreds of wildly different GPUs.) The only way you could see similar magical performance boosts in CPU code is something like Java or .NET, where the code is JIT-compiled.

 

[mikk]

Highend GT3 specs.

http://chinese.vr-zone.com/56926/in...th-gt2-gt2-gt2-package-cache-memory-03252013/

 

[Idontcare]

Highend GT3 specs.

http://chinese.vr-zone.com/56926/in...th-gt2-gt2-gt2-package-cache-memory-03252013/

I see they dropped TJmax to 100C from the 105C that it is for IB. That helps to get that TDP down below 50W.

Anyone know if they did the same for the 4770k?

 

[jpiniero]

Interesting, they are offering GT3 with and without Crystalwell. And those quad core models are the only ones getting it. But with the turbo all the way to 1300 it should be close to double GT2.

I'll guess that the GT3 without Crystalwell is just to 800 like the original rumors.

 

[Arachnotronic]

Will be pretty exciting to see how GT3 and GT3e do!

Unfortunately for AMD, this is just yet another example of huge R&D + finally aiming at the right target from Intel :/

 

[Idontcare]

Will be pretty exciting to see how GT3 and GT3e do!

Unfortunately for AMD, this is just yet another example of huge R&D + finally aiming at the right target from Intel :/

With the on-package VRMs as well as the on-package eDRAM, I get the feeling that performance/watt for mobile applications dealing with graphics (gaming) is going to get pretty darn impressive.

Performance/dollar might not be all that great though.

 

[inf64]

AMD has no 47W mobile part that is comparable to these parts. They do have upcoming Richland 35W parts but price is a big factor too and I expect that Richland will hold a big advantage there. Also it will be able to do hybrid CF with latest SI discrete parts so if AMD offers discount for all-in one purchases to OEM partners(APU+board+SI GPU that does Hybrid CF) they might come out as both cheaper and faster (in graphics workloads). CPU parts would not be comparable of course since Haswell is in another perf. (and price) league.

 

[mikk]

You get huge micro stuttering with crossfire. Not really comparable to a single gpu solution.

 

[inf64]

No, you get that with high end GPUs and only in some titles. With mid range/low end stuff you just get no scaling in few games. Microstuttering is more of a 2009 thing with drivers and game profile support getting better and better with each new drive release.
Also comparing AMD Radeon image quality,driver capabilities and game support to HDwhatever is like comparing Mercedes to Dacia. Both are cars,both can get you from point A to point B but Mercedes is another class.

 

[Rvenger]

No, you get that with high end GPUs and only in some titles. With mid range/low end stuff you just get no scaling in few games. Microstuttering is more of a 2009 thing with drivers and game profile support getting better and better with each new drive release.
Also comparing AMD Radeon image quality,driver capabilities and game support to HDwhatever is like comparing Mercedes to Dacia. Both are cars,both can get you from point A to point B but Mercedes is another class.

I had a Llano A6-3650 coupled with an HD 6570 and it was Microstutter heaven. .. or hell i guess.

 

[Homeles]

Microstutter is still a real problem -- very much so.

 

[mikk]

And it's tendentially much worse on Crossfire than on Sli. Crossfire fps and single GPU fps are not comparable.

 

[inf64]

Those graphs show high end cards in high resolution modes running high end games. I'm talking about mainstream/low-end stuff here. Also using frametime gathered by Fraps,as explained in latest AT article, is flawed. It does show an issue but it's nowhere near accurate at measuring the real microstuttering in high end configurations (by serious tools like GPUview).

 

[SPBHM]

CF only makes sense if you need to go higher than the fastest single GPU... apart from that, I don't see any sense, and I think DG shouldn't even be mentioned as something positive...

you can CF richland only with the old 480sps VLIW5 GPU or the newer 384sps GCN GPU with dedicated memory, both will perform slightly different than the IGP, and this is certainly not going to make stuttering better than what already is bad with two identical GPUs I think.

Those graphs show high end cards in high resolution modes running high end games. I'm talking about mainstream/low-end stuff here. Also using frametime gathered by Fraps,as explained in latest AT article, is flawed. It does show an issue but it's nowhere near accurate at measuring the real microstuttering in high end configurations (by serious tools like GPUview).

they are not using fraps, this is video capture.
http://www.pcper.com/reviews/Graphi...ils-Capture-based-Graphics-Performance-Testin

 

[Homeles]

Those graphs show high end cards in high resolution modes running high end games. I'm talking about mainstream/low-end stuff here.

Lower end hardware has even larger stuttering issues.

Also using frametime gathered by Fraps,as explained in latest AT article, is flawed. It does show an issue but it's nowhere near accurate at measuring the real microstuttering in high end configurations (by serious tools like GPUview).

Read the full review: http://www.pcper.com/reviews/Graphi...ils-Capture-based-Graphics-Performance-Testin

It's just as bad when using hardware video capture. Dropped frames and stutter are everywhere.

 

[inf64]

I don't see any lower end GPUs tested in that pcper article.

 

[Phynaz]

No, you get that with high end GPUs and only in some titles. With mid range/low end stuff you just get no scaling in few games. Microstuttering is more of a 2009 thing with drivers and game profile support getting better and better with each new drive release.
Also comparing AMD Radeon image quality,driver capabilities and game support to HDwhatever is like comparing Mercedes to Dacia. Both are cars,both can get you from point A to point B but Mercedes is another class.

Did you bother to read the front page article about stutter? Multi-GPU stutter happens always. And crossfire a higher performing GPU with integrated graphics? Ugh, what a mess for the consumer that would be.

What major vendor / retailer is going to want to foist crossfire onto Joe Public and incur the support costs?

 

[inf64]

I just said it's an option. Richland is going to be way more cheaper than those Haswell parts so in the end they don't need to make such an offer to OEMs. It's a matter of price and performance.

 

[mikk]

Low end and high end are based on the same multi-gpu method, technically there is no difference. If a high-end suffers from stuttering then also lower end gpus. The stuttering is worse on lower frame rates, since you have a better chance reaching higher fps on fast multi-gpu configurations it's more the opposite that slower multi-gpu configurations doing worse overall.

 

[mikk]

http://chinese.vr-zone.com/58507/in...70r-and-i5-4570r-with-graphics-5200-04052013/


Looks like BGA GT3e for desktop is coming.

 

[Homeles]

http://chinese.vr-zone.com/58507/in...70r-and-i5-4570r-with-graphics-5200-04052013/


Looks like BGA GT3e for desktop is coming.

Not all too unsurprising. Guess who that's going to be for, though? Definitely not your average consumer. If I had to guess, some fruity company would be interested in that.

 

[AtenRa]

I see they dropped TJmax to 100C from the 105C that it is for IB. That helps to get that TDP down below 50W.

Anyone know if they did the same for the 4770k?

IDC, Intel Ivy Core i7 3740QM (base clock 2.7GHz, Max Turbo 3.7GHz) is at 45W TDP not 50W.

According to the slides, Haswell 4950HQ (base clock 2.4GHz, max turbo 3.6GHz) is at 47W TDP. So it seems that new Haswell mobile parts featuring GT3 graphics, will not provide more CPU performance vs Ivy parts. It seems to me that Haswell design was focused on the iGPU part and we will see this trend continue with Broadwell.

It will become increasingly difficult for Intel to continue doubling the iGPU logic and at the same time keeping the same CPU performance while lowering the power consumption.

Also, with the GT3 graphics the die size should be close to 180-200mm2(from 160mm2 of Ivy). That is a huge die size for an expensive 22nm FinFet node and prices will not be cheap.

 

[mikk]

Finfet isn't expensive.