Unlike you it appears, yes. They cleared 1100MHZ, I'd say a 35% increase over what you said is possible, albeit almost a year ago, warrants acknowledgement.
That's exactly what I didn't say. You continue to demonstrate your lack of understanding of context.
No, that was proof that despite your attempts to say otherwise 800MHZ wasn't a limitation or the RAM they were using, obviously we all know that was wrong, that has not only been available in consumer parts for some time, but it has been hitting considerably faster speeds then what you said was possible.
You can always go higher, with more voltage, active cooling, LN2 etc. Does that make it feasible for a consumer tablet/phone
Not once have I stated what speed T4 was going to use for RAM- that is 100% you. You made an absolute claim as to both the bandwidth and clockspeed. I merely pointed out that there are more options available then what you are claiming. Going to be real simple, either every T4 part ships with 800MHZ or slower RAM and you are correct, or some ships faster then that and you are wrong. I made no claim except to point out there are options outside of what you stated was possible.
T4 is limited by its dual 32-bit memory interface. This is the point that was being argued, before you de-railed as usual and used your copy paste article "knowledge". Considering this is for a mobile device, not one hooked up to a 1KW PSU it would make a lot more sense to put down a slower quad channel interface at the cost of a bif of extra die space, and clock in the range of 533 instead of ramping clocks (which is easy if we believe you) to 1GHz and paying every power penalty that comes with it. Also I didn't say every T4 part, I said every T4 launch part. Subtle difference, one again your reading skills fail you.
On T4 it is? I wouldn't say it is or isn't, but I haven't seen anything that backs your claims nor refutes it. I guess the simple solution to that is we shall see.
It's not impossible, of course not. But you'll run into the same physical limitations and power problems desktop parts had. This is why we have multi-channel memory architectures!
That was DDR3L, not LPDDR3- two different things. T4 supports them both.
Sorry, you are right. You linked to DDR3L, which is basically irrelevant in mobile space. Remind me again why people want higher power consuming memory in their mobile devices?
Here is my original quote. Once again you post before you think.
LPDDR3 is in its infancy, do you think we are ready to ramp clocks already?
GLBench is utterly useless, it is a bad sort test, nothing more.
A bit like any non TWIMTBP game on the desktop is irrelevant right?
Prime had one sixth the bandwith that T4 devices are supposed to have according to current whitepapers. It had one sixth the GPU cores. You have repeatedly stated that T4's GPUs were going to be entirely bandwidth limited- I accurately pointed out if that is the case then T3 was also bandwidth limited(which obviously, it wasn't). Six times the computational resources, six times the bandwidth. Where I come from we consider that linear.
For it's screen size it coped, but it was still bandwidth limited under texture fetch, by 35%. Did you read what you linked again? The infinity prime with its much larger screen needed the increased memory bandwidth!
Also don't CPU cores need bandwidth anymore either? Display output, that's 2.25x from 720p to 1080p. You even said it yourself in a post somewhere that larger display resolutions also mean larger renders, larger blits and more data moving around!
With T4's power requirements looking so high it's future in anything under a 1080p screen looks unlikely.
Now that I think a bit more about it though, T4 only has 48 Pixel Shaders, so actually it'll probably be fine on the bandwidth side. Just under powered by 33%. Oh well when they unify they can get a 50% gain and make some fancy marketing slides so say how great unified shaders are.
Facebook
Twitter