Would not the 2MB of on-chip cache be helpful in requiring less bandwidth from the external memory?
GM108 may not have the same 2 MB of cache as GM107. But even if it does, I doubt the on chip cache would be enough to make up for such a narrow bus. Heck..there are some SoC's with higher memory b/w than GM108
With Maxwell yes, with Kepler I doubt and that's why I was asking for comparisons.
I could be wrong but my impression was that while GCN and Kepler were very close, Kepler was still a bit ahead. But even if GCN was ahead, why would Nvidia advertise this?
The potential software advantage on Nvidia side is an argument only in case you care, since I don't care and guess there are others who don't either, I will try to stick with AMD products only...
Guess you are one of the few who don't, most people do care.
I am not a NV fanboy or anything btw, I've had a number of AMD cards and am currently running a Radeon HD4850. I'm just calling a spade a spade.
And it is a shame that NV doesn't even bother to mention their true competition in their marketing slides. So, don't they consider them competitive anymore?
I guess you dont understand the concept of marketing...
edit: I have read the whole Anandtech's article on the matter and even there I sense some type of hubris, pride, airiness against AMD's products, as it is really some weird type of propaganda without any hard numbers to support their claims.
Propaganda against AMD? I guess you missed the "AMD Centre" section on Anandtech...
Anyway, Anandtech didn't outright call Enduro bad, they generally gave reasonably positive reviews on Enduro, even though it was broken.
Given this is gddr5 if you want comparable numbers multiply that by 2 as this isn't "data rate clock".
With these numbers it would actually be less than one fifth, though it's possible the ddr3 is 1000 (2000) Mhz instead which is then exactly one fifth...
Hmmm I got the info from the Table from Anandtech here -
http://www.anandtech.com/show/7834/nvidia-geforce-800m-lineup-battery-boost
From that I got the impression that it means 2.5 ghz effective clock...because since GDDR5 is quad pumped, if it was 5.0 ghz, the only other number that could be listed is 1.25 ghz. 2.5 ghz does not make any sense.
But on checking some other sources..it does indeed seem to be 5 Ghz effective. And the GDDR3 does appear to be at 2000 mhz so yeah, then it would be exactly one-fifth.
Interestingly enough, I decided to check NV's own website as well to see if they had the specs. They list the memory clock for GTX860M as "up to 2500 mhz", however the memory b/w is listed as 64.0 GB/s, which can only be had with a 4 Ghz clock. Man are those guys confused
http://www.geforce.com/hardware/notebook-gpus/geforce-gtx-860m/specifications
Most likely GM108 is 64bit only (just like gk208). But the half-bandwidth-than-IGPs if you pair it with ddr3 definitely isn't going to help.
Oh yeah, you're right. I forgot about GK208, it probably is 64 bit only.
DDR4 cant come soon enough. Assuming they could do 3.2 ghz, would be a nice boost for these crippled cards.
That was my idea as well, but Dave corrected me earlier in this thread saying there was a large market below the gm107 class. I assume that AMD plays with APUs there, but Nvidia probably wants its share of the pie as well.
Interesting..didn't think the market was that large. Then again..I do know people who've bought 2 GB Geforce 610/620M GPU's
AMD would compete with APU's or APU+GPU combinations but I guess there are enough people who would want an Intel CPU and/or a Nvidia GPU that it makes sense for NV to have a separate part to address this market.
That said, given the large gap between 840M and 850M, I wouldn't be surprised if Nvidia launches a 845M or something at some point in the future which is based on a cut-down GM107 with 128 bit GDDR3.
