Sandy Bridge architecture overview

[IntelUser2000]

I'm a big proponent of "if its delayed it probably sucks" theory. If any of their graphics performance claims are true, the GPU could be purpose-designed to be only DX10.1. The mass movement from programmable in Larrabee to fixed function units is one indication.

Of course, who's to say if they are hiding their cards again. 🙂

Edit for Tech Report's brief overview: It has different info from Anand's, but there's much less detail. If I was to choose between two to read, it would be Anand's. Of course, I'd read both, because not every article on same subject has same insights.

 

[ydnas7]

does cinebench 11.5 use AVX? because SB's score is tops, even with turbo off
http://www.coolaler.com.tw/coolalercbb//SandyBridge/S45.PNG

 

[Nemesis 1]

Heck, if it does GPGPU well enough to do OpenCL (and also DirectCompute then), they could implement tessellation through a compute shader.
The DXSDK contains some examples of tessellation using DirectCompute. Good fallback for DX10 cards.

You seem to know this Graphics stuff well . I was wondering what is the benefit of transcendental math in the GPU? Terrian and the like could use transcendental math can it not?

 

[Nemesis 1]

does cinebench 11.5 use AVX? because SB's score is tops, even with turbo off

No.

 

[Nemesis 1]

I'm perplexed how it is that they did not plan to support DX11 with a leading edge design.

I wonder if it is really a DX11 design but either the driver support is still lacking or the design itself turned out to be borked so they down-graded its functionality via driver limits to just use the DX10.1 function.

For example say they have Tesselators included but it turned out their design of them limited the clocks to 500MHz max or the output itself was not DX11 compliant?

They had 4yrs to get this right after all, and it is being released a full year and change from the first debut of DX11. Why ensure AMD gets the upper-hand in marketing Llano and eOntario with their DX11 support by making SB a mere DX10.1 part?

We are talking IGPs here. Tess. Will come with 22nm. Lers see how AMD does with tess on their stitched cpu/gpu.

 

[IntelUser2000]

None of the current benchmarked programs use AVX. Heck, their demos showed AVX-optimized programs being over 2x faster than the Westmere comparison.

Two interesting notes from Tech Report's overview:

this is, at long last, the breaking point where one can finally say virtually nothing remains of the P6 (Pentium Pro) roots that have undergirded everything from the Conroe/Merom Core 2 to the Nehalem/Westmere Core i-series processors.

Wondered about this.

Several folks from Intel told us the firm is looking into possible options for making the transcoding hardware available to users of discrete graphics cards, but if that happens it all, it will likely happen some time after the initial Sandy Bridge products reach consumers.

Before, DirectX instructions would break down into two to four internal instructions in the IGP, but in Sandy Bridge, the relationship is generally one-to-one.

I didn't get what they meant on the 2nd one when they mentioned "CISC", but the previous architecture must have been more "RISC-like".

 

[Martimus]

I didn't get what they meant on the 2nd one when they mentioned "CISC", but the previous architecture must have been more "RISC-like".

I didn't see that article, but they likely were talking about CISC since "complex instruction set computing" is often considered synonymous with x86 processors.

Do you have a link to the article? It sounds interesting.

EDIT: I found the link. It was on the previous page.

 

[Scali]

You seem to know this Graphics stuff well . I was wondering what is the benefit of transcendental math in the GPU? Terrian and the like could use transcendental math can it not?

Transcendental functions are used in a lot of places in graphics, both for solving lighting equations and for geometric computations.

 

[Idontcare]

Two interesting notes from Tech Report's overview:

Ugh, that article made for some dry reading D:

Moar purdy pics please! 😛

 

[Martimus]

and a good short article on SB, its like grandpa's axe...
http://techreport.com/articles.x/19670

That was a really good article. Thanks.

 

[ilkhan]

So this is rather a ways out there, but anybody want to predict if *IVY* bridge will have 25W ["normal" voltage] mobile quads? Looks like sandy is going to have the same 45W mobile quads (albeit now with GPUs included in that figure) as lynnfield.

 

[Dark_Archonis]

We are talking IGPs here. Tess. Will come with 22nm. Lers see how AMD does with tess on their stitched cpu/gpu.

The tessellation performance on AMD's 5xxx cards is fairly weak, and Fermi cards easily beat 5xxx cards in tessellation. AMD's Ontario platform is supposed to have a 5xxx-based GPU.

 

[Viditor]

does cinebench 11.5 use AVX? because SB's score is tops, even with turbo off
http://www.coolaler.com.tw/coolalercbb//SandyBridge/S45.PNG

There's a good chance of it...

["Tilo Kühn at Maxon Computer said, “We’ve been enthusiastically using the new version of Intel® C++ Compiler Professional Edition that includes support for Intel® Advanced Vector Extensions (Intel AVX.) Being able to performance tune our software well in advance of processor availability gives us a major development head start to ensure that our Cinebench product will be ready when the first Intel AVX-enabled processor is delivered.”"]

http://software.intel.com/en-us/blogs/tag/avx/feed/

 

[ydnas7]

http://translate.googleusercontent....&twu=1&usg=ALkJrhhH_d4KLSRUdktdOjCMjCpmsQXjKA

coolaler compares SB to Nehalem (i875K), both at same frequency 3Ghz, both with turbo off. same graphics card GeForce CTX 460
cinebench 11.5 SB CPU score is 12% better than Nehalem
SB OpenGL score is 23% better than Nehalem
SB OpenGL score is 0.3% better than a Magny Cours (with ATI Firepro..)

nb this was with the 6mb cache SB, not the 8mb version

oops, thought i was reusing an old comment, but i edited it, so here is the bump

 

[IntelUser2000]

There's a good chance of it...

But I disagree. 🙂

When AMD when to 128-bit FPU on Barcelona, there was no perceivable gain on Cinebench that was from FP.

http://www.tomshardware.com/reviews/amd-phenom-athlon-64-x2,1746-11.html

Sorry Cooaler, the only thing useful about the new results is that it improved from the original scores that were pathetic, confirming suspicions that there was something wrong with the platform. Anand beat you to the punch on the real preview.

 

[Nemesis 1]

IDC , Says he is perplexed. Here what I see IDC. I see EU taking up small space on a large die. Intel does Open GL3. Open CL 1.1 DX 10.1 . I believe Intel can do Direct Compute but will punish Ms tell 22nm. I think intel can do software tess but don't want to waste the budget. Intel said 3 years ago they don't care what DX is used. Now they have gone fixed function and a really good shader model were Driver libriars are minimized Its pretty clear.
Whats perplexing is those dammed (Ya I know its n). ONLOCK KEYS.

 

[Viditor]

But I disagree. 🙂

When AMD when to 128-bit FPU on Barcelona, there was no perceivable gain on Cinebench that was from FP.

http://www.tomshardware.com/reviews/amd-phenom-athlon-64-x2,1746-11.html

Sorry Cooaler, the only thing useful about the new results is that it improved from the original scores that were pathetic, confirming suspicions that there was something wrong with the platform. Anand beat you to the punch on the real preview.

Fair enough mate...but it still leaves open the question of whether Maxon is using AVX...🙂

 

[Nemesis 1]

this is, at long last, the breaking point where one can finally say virtually nothing remains of the P6 (Pentium Pro) roots that have undergirded everything from the Conroe/Merom Core 2 to the Nehalem/Westmere Core i-series processors.

Wondered about this.

Quote:
Several folks from Intel told us the firm is looking into possible options for making the transcoding hardware available to users of discrete graphics cards, but if that happens it all, it will likely happen some time after the initial Sandy Bridge products reach consumers.

I think 1) above is about the ring bus and scaleability

2) I hate to plug an NV product. But they do have that chip that allows gpu switching . Without shutting down or any bad effects.