Slashdot Mirror
Early Ivy Bridge Benchmark: Graphics Performance Greatly Improved
The folks over at Anandtech managed to spend some time with early Ivy bridge production samples and perform a few benchmarks. The skinny: CPU performance is mildly increased as expected, but the GPU is 20-50% faster than the Sandy Bridge GPU. Power consumption is also down about 30W under full load. The graphics, however, are still slower than AMD's Llano (but the Ivy Bridge CPU beats the pants off of the Fusion's). Is the tradeoff worth it?
It isn't meant to be powerful graphics. It isn't a "tradeoff". Intel's HD graphics are meant to be very low power, but competent enough to run basics, shiny OS features at least. That they do, and it sounds like IB is even better at that. But it isn't a "tradeoff" to get a good CPU with basic graphics that is called "normal". If you need good graphics discrete is still the way to go and there are plenty of reasonable options.
From the look of it, Ivy Bridge is quite a win. Sandy Bridge, but a bit better. Nothing not to like there.
The main reason that integrated GPU performance matters(aside from the fact that it is all the GPU you get in any too-cheap or too-skinny device that doesn't have a discrete option) is that it defines the (overwhelmingly common) baseline for what 'PC graphics' means. If that situation is uniformly awful, GPU intensive stuff will continue to be fairly niche, which leads to a chicken-and-egg issue: if integrated graphics suck, the market for GPU intensive stuff will be constrained, which will reduce the incentive to improve GPU performance, and so it goes...
The beauty of having an on-chip GPU is that you don't have to move data around to do computations with OpenCL. It's something that kills the benefits of using a dedicated graphics cards for almost every GPGPU application. The 10-100x speed-ups are a lie.
As soon as ARM tries to catch up to the performance of x86 (and x64) it no longer has the lower power consumption.
I'm starting to think GNU is the problem with "GNU/Linux" these days.
I would like to see the ability to use the integrated GPU, even if not for graphics. The traditional CPU is good for sequential logic. But for pattern recognition, physics simulations (which is basically what 3d graphics is), encoding, or code-cracking (e.g. bitcoin), the highly parallel structure of the GPU is better. Now you might argue, my offboard GPU is still the same thing, but better. OK. But these are inherently parallel tasks, so if you could use the one built-in AND the add-on, you wouldn't be wasting anything.