Sandy Bridge

If this is the right location of the L2 (as it seems), then the SRAM array pretty much sizes for 256KB, as far, as I can tell from the blurry shot:

70642351.png


I somewhat doubt a 512KB chunk can be tuned for a rapid 9-cycle latency access time, as it states. :???:
 
fellix said:
I somewhat doubt a 512KB chunk can be tuned for a rapid 9-cycle latency access time, as it states.
Click to expand...

How about the L3? IIRC Nehalem's L3 latency is 40+ cycles so this is the kind of improvement inline with the "revolutionaly cache architechture" I've heard about. I wonder what it does to the overall performance. Maybe it's time someone pour LN on a Core i7 and overclock the NB :smile:
 
Let's hope for a robust implementation, then! ;)
Sadly, we won't see FMAC SIMD until 2012, for the next "tick".

p.s.:
That 256 Byte (bit ?) ring-bus for the L3 is puzzling me. Does that mean that each CPU core got to access its own chunk of the L3 array that loops the requests with the others?!
Very Itanium-like cache organization, BTW, and if this is to be the case, the GPU core is likely to be part of the L3 "loop" by a similar manner.
 
Last edited by a moderator:
The stated 1.0-1.4GHz excludes in my mind the graphics core being IMG IP. IMG mentions for it's SGX cores up to 200MHz under 65nm and up to 400MHz under 45nm. Besides IMG hasn't even announced yet its D3D11 IP family which they might call Series6.
 
Ailuros said:
The stated 1.0-1.4GHz excludes in my mind the graphics core being IMG IP. IMG mentions for it's SGX cores up to 200MHz under 65nm and up to 400MHz under 45nm. Besides IMG hasn't even announced yet its D3D11 IP family which they might call Series6.
Click to expand...

Odds are quite good that it'll be an evolution of the current architecture, with a few extra EUs thrown in and some internal shuffling done to include what's required for DX11 compliancy(if it is actually a DX11 part).
 
Funny, isn't it, that intel should come out with a hybrid cpu+gpu before AMD while it is AMD which is talking up fusion all over the place.
 
I fail to see how any IGP will ever be exciting simply because they can't give it remotely enough bandwidth compared to discrete vid cards. Maybe it will have some GPGPU purpose, but that will undoubtedly be niche stuff and discrete will be much better again.

Maybe I'm missing something :)

But honestly as long as an IGP can run Aero and playback HDV, it's all I need. Spend $50 on a card if you wanna game.
 
Last edited by a moderator:
swaaye said:
Maybe it will have some GPGPU purpose, but that will undoubtedly be niche stuff and discrete will be much better again.
Click to expand...

You'd be amazed at how IGPs handle certain GPGPU tasks versus their discrete counterparts. Transfers to and from the GPU aren't particularly cheap, and the more you can trim from that the better...IGPs have an advantage there.
 
swaaye said:
I fail to see how any IGP will ever be exciting simply because they can't give it remotely enough bandwidth compared to discrete vid cards. Maybe it will have some GPGPU purpose, but that will undoubtedly be niche stuff and discrete will be much better again.

Maybe I'm missing something :)

But honestly as long as an IGP can run Aero and playback HDV, it's all I need. Spend $50 on a card if you wanna game.
Click to expand...

What you were missing in 1st part of your post is answered in 2nd part, it's true for vast majority of computer users.
 
Kaotik said:
What you were missing in 1st part of your post is answered in 2nd part, it's true for vast majority of computer users.
Click to expand...
Well it seems like this new IGP is rather huge. I can only guess that there's some good reason to spend the budget that way. I just can't see the "good reason" though. :)
 
You must log in or register to reply here.
Back
Top