Sandy Bridge preview

Don't the two sentences contradict each other? They have upped the performance by 4x only increasing the EUs by 50%.
Yes, this is true, but at some point you will reach the limits of the shader alus - it seems unlikely that if you want to double the performance of the igp, you'd suddenly go back to half the shader units and compensate that loss with improvements in other parts. The architecture is still very similar, after all.

So notebook quad-cores will ony have 6 EUs?
I'm still really curious how many different dies these chips will actually have, but I guess the notebook quad-cores only having 6 EUs wouldn't be a big loss. Those quad-cores have a bigger power budget anyway, hence nothing for thin&light and could be easily paired with a discrete gpu. I'm still sceptical about this 1/2 core gpu thing, but from that point of view it would make sense.
 
So this means the bottom line finally got raised a smidge, right?

Hmm yeah well I guess most devs that don't make Farmville, World of Warcraft or Windows Solitaire used to just ignore GMAs completely. This might just be enough for them to actually start to target them. Ah well, who needs DX11, an AA setting in an engine, or any of that fancy stuff anyway..
 
Ah well, who needs DX11, an AA setting in an engine, or any of that fancy stuff anyway..
I wouldn't be totally surprised if intel added AA capability. At the very least, according to the open source driver, the ROP partition of the chip looks like seriously redesigned, could as well be completely new (I guess it needed changes anyway due to the different memory integration, i.e. L3 cache), but in any case, while I don't see any new msaa bits (note surface state already had multisample bits since ages but needed to be zero) there are definitely new bits for at least coverage to alpha.
I can't see any significant changes to the EUs, though - of course that doesn't mean they didn't change internally but I think they generally remained the same. Well, one change there seems to be the addition of a per-instruction accumulator write control bit - it always looked to me like it's very, very hard to use the accumulator efficiently in common shader code (that is, using it for MADs translated as MAC for instance - the open source driver never even attempted it), that should make it more feasible I think (note that it couldn't do accum write followed by read directly without a thread stall at least up to g45, but if you only have global accum write control it's a bit difficult to schedule other instructions in-between...)
 
Last edited by a moderator:
Twice the units at a twice lower frequency can run at a lower voltage and hence have lower power consumption.

Does anyone even believe this makes a hint of sense? I WAG Anand himself doesn't know enough yet. I mean he admitted that he's waiting for Sandy Bridge details. Intel is keeping the details really tight.

Early Huron River slides indicated the difference will be only 2x. With the lack of such details on desktop I wouldn't be surprised if the "2 core" part exists.

The biggest question is, when they mean "core" does it mean the EUs only or is it true multi-GPU style, or even a combination of both?
 
Just bought an i5-760 like a month ago. Congratulations to me or to Intel?
Damn I hate the PC market, devaluation is just so fast... If it wasn't for Starcraft I wouldn't touch it with a stick ;)
 
It doesn't support OpenCL. That is one thing. ( Or Intel simply decide not to work on OpenCL Drivers )
 
It doesn't support OpenCL. That is one thing. ( Or Intel simply decide not to work on OpenCL Drivers )
Why would they waste time on that? A quad-core Sandy Bridge CPU can do 200 SP GFLOPS or 100 DP GFLOPS. The GPU can likely do only 100 SP GFLOPS. Add to this the ability to run fully independent threads and low latency, and the CPU is vastly more interesting for generic computing than the GPU.

Also, no developer is going to invest in porting his code to OpenCL unless it is universally supported and reliable. That's still half a decade away. They can rely on SSE now and take advantage of AVX with minimal effort.
 
Does anyone even believe this makes a hint of sense? I WAG Anand himself doesn't know enough yet. I mean he admitted that he's waiting for Sandy Bridge details. Intel is keeping the details really tight.

Early Huron River slides indicated the difference will be only 2x. With the lack of such details on desktop I wouldn't be surprised if the "2 core" part exists.

The biggest question is, when they mean "core" does it mean the EUs only or is it true multi-GPU style, or even a combination of both?

Even i suspect the part he tested had full 12EU GPU capability. The indication from intel was a 2X increase in graphics performance over Arrandale, which is what Anand achieved with his sample. Now there is a possibility they decided to go 4X to counter Llano but i doubt it

I doubt they're gonna have multiple dies for the 6EU/12EU parts. They probably have one die for dual core and one die for quad core and they'll disable cache/EU's/HT for the various SKU's. And i believe its not Multi-GPU style, its apparently just like 40SP's vs 80 SP's for AMD or 16 CUDA cores vs 32 for NV
 
Just bought an i5-760 like a month ago. Congratulations to me or to Intel?
Damn I hate the PC market, devaluation is just so fast... If it wasn't for Starcraft I wouldn't touch it with a stick ;)

That processor should last you years. And if it ever starts feeling slow you can give it a nice overclock.
 
AFAICS, all that has been updated is that he is suspecting that it might have been a 12 eu part. There has been no confirmation so far.
Maybe Dave knows it's 12 EUs but can't tell :).
I still can't make much sense out of why some products would have 6 EUs, some 12. Sure if the EU count is tied to different dies (like dual/quadcore), but if all mobile products have 12 that can't really be. Otherwise, 6 EUs is going to be a tiny part of the die, so it makes not much sense to disable half of it for salvage parts. And with all the turbo stuff it can't be for power reasons neither. So why would intel make it slower for some products - after all intel doesn't want to sell you low-end discrete cards...
 
Maybe Dave knows it's 12 EUs but can't tell :).
I still can't make much sense out of why some products would have 6 EUs, some 12. Sure if the EU count is tied to different dies (like dual/quadcore), but if all mobile products have 12 that can't really be. Otherwise, 6 EUs is going to be a tiny part of the die, so it makes not much sense to disable half of it for salvage parts. And with all the turbo stuff it can't be for power reasons neither. So why would intel make it slower for some products - after all intel doesn't want to sell you low-end discrete cards...

Mczak, there's a new Anandtech article which says GT1=1 core and GT2=2 core. There are references to GT2 out there. Therefore we can't count the possibility out. 1 core vs 2 core seems more than just EUs though.
 
Just bought an i5-760 like a month ago. Congratulations to me or to Intel?
Damn I hate the PC market, devaluation is just so fast... If it wasn't for Starcraft I wouldn't touch it with a stick ;)

if its for a laptop then its congrats to intel. If its for a desktop then you should be fine. I'd think adding a faster video card would do more than a faster cpu with what amounts to $50 graphics card performance.

I never recomend laptops for gaming. They are passed by way to quickly
 
Why would they waste time on that? A quad-core Sandy Bridge CPU can do 200 SP GFLOPS or 100 DP GFLOPS. The GPU can likely do only 100 SP GFLOPS. Add to this the ability to run fully independent threads and low latency, and the CPU is vastly more interesting for generic computing than the GPU.

Also, no developer is going to invest in porting his code to OpenCL unless it is universally supported and reliable. That's still half a decade away. They can rely on SSE now and take advantage of AVX with minimal effort.

OpenCL is supported by every graphics core AMD and nvidia have sold in last 2-3 years. And there are just 3 big players on graphcis market.

So you say intel GPU's should not support openCL because it's not currently supported by intel's GPU's? quite a circular logic.


Quad-core Sandy Bridge may theoretically do 200 SP gflops when running at >3 GHz but most laptops will be having dual-core chips running at 2.5 GHz. We are talking about 80 theoretical SP gflops then.

Nvidia's new integrated chip(found in 13" macbooks and mac mini) can do about 150 gflops, and does it with much lower power consumption. And that's still based on last-generation tech, not fermi.


And when the brute-force-dsp-calculations are done GPU, CPU can do others things at the same time.


The major reason Intel does not do openCL drivers is because Intel don't want openCL to succeed; As long as most software cannot execute their dsp-style code in GPU, people need intel's bigger, more powerful, more expensive CPU's to run their video and image processing software.
 
Last edited by a moderator:
The missing pin is just so Intel get to sell everybody new motherboards all over again. They have a hard-on for changing the pinout slightly just so you HAVE to buy new everything from them one more time. They've done it a million times before in the past.
If you want all the new features of Sandy Bridge, you need to get a new mobo with the new southbridge. No doubt about it, you dont want to run SNB with older motherboards.
 
If you want all the new features of Sandy Bridge, you need to get a new mobo with the new southbridge. No doubt about it, you dont want to run SNB with older motherboards.

Its not anyone wouldn't want to, simply that you cant run SB processors on the older motherboards. Intel's moved the clock generator on chip with SB which requires a new motherboard design. Intel sure is making a killing on the chipsets though. They're selling southbridges which earlier used to cost $5 for the price of a northbridge(ie $40).
 
Back
Top