What I´m deducing is that buswidth and amount of RAM are related and
allow to
predict the probable amounts of video RAM for a series of cards of the same
generation.
After looking at the 6xx to 9xx range of cards, I conclude that when
going from
256bit in the 6xx card range to 384bit in the 7xx range, there was no
increase
to 512bit in the 9xx range for bus width.
This means, in terms of what to expect in video ram amounts from a 9xx
series,
that there will probably not be an increase to 16 GB of RAM for
something like
a Titan X Black edition type of card, unless this card will already use
a next gen,
e.g. 10xx type card layout with probably a 512bit bus.
Don´t be such a bully, biting my ankles.
I give a rat´s ass about wether bus follows RAM or the other way around,
it´s not the point,
especially when I´m deducing that bus width hasn´t been adjusted to
allow convenient
adressing of RAM and conclude that nvidia probably found this not
neccessary, re-using
the 7xx card layout instead.
I take bus width only as an indicator for what to expect from the 9xx
card generation.
Probably no 16GB card version, unless a new, modified card layout is
introduced as
part of a Titan X Black edition with a possible card layout already
taken from a 10xx
type card generation.
Cheers,
tim
P.S: Regarding Softimage and the 970 cards, I hope it´s just a driver
issue that´ll go away.
Am 03.06.2015 um 23:08 schrieb Raffaele Fragapane:
If they need 512 they can probably do it without issue. What you're
saying is that you would have liked to make cards with more ram, the
bus width is irrelevant, it gets sized as needed, not the other way
around.
On 4 Jun 2015 5:33 am, "Tim Leydecker" <bauero...@gmx.de
<mailto:bauero...@gmx.de>> wrote:
Am 03.06.2015 um 08:32 schrieb Raffaele Fragapane:
Huh?
The width is whatever is required for the controllers to address
the RAM.
If they have 12GB over 6 32bit controllers as that manufacturing
specs max why would they have more than 384?
I was hoping for nVidia to bring the bus width up to 512bit,
making 2,4,8,16,32,64 GB Ram likely because that
would go well together with such a bus width (or even just a
256bit width bus).
Of course, if all you have is 384bit, 12 GB is what is convenient
to connect, not 16GB (as in AMD´s current 512bit bus cards)
My point.
Looking at previous release/development cycles of nVidia, one
could now expect to see a Titan Z Black edition coming
to close off the 9xx series, with some sort of shrunk production
process, more cores or a little bit of higher clocking
but unlikely to have a wider bus to adress video RAM in the 16GB
range.
Such a thing will probably not come before the next generation of
cards, in pseudo naming, the 10xx series.
Not before next year.
This gives AMD 1 year to try and get customers looking for lot´s
of video ram for their editing, comp, etc.
tim
Also, what the architecture and the proposed manufacturing
guidelines allow in terms of addressing width isn't the same as
what's out in the current card of the month.
The 980 is the same in most regards but only has 256bit in
example because al it needs to address is 8GB.
If they need to address more It's very likely the width can be
pushed a good deal further.
The bottleneck isn't currently measured in bus width, the
throughput is an issue, and it's got little to do with the width
of addressing stacks, and it's why things like NVLink and new PCI
bus specs and so on are being looked into.
There are a lot other design issues that are being worked on by
more than just a company, the addressing width across the bus
isn't particularly symptomatic of any of those AFAIK.
On Wed, Jun 3, 2015 at 3:58 PM, Tim Leydecker <bauero...@gmx.de
<mailto:bauero...@gmx.de>> wrote:
The 980ti (starting at EUR 735,-) is a good opportunity
compared to the gtx980 (starting at EUR 500,-)
but it is annoying to know that Video-RAM will soon become a
bottleneck because more and more
applications start to utilize GPU performance to their
benefit, either when caching out like in Nuke for
huge environment images or a GPU renderer like Redshift3D
having to optimize, e.g. limit it´s
cache sizes to fit into a smaller than desireable meomory
footprint.
All that on top of what a 4k display would demand for it´s
share of available video memory to start with.
I think Nvidia missed an opportunity there, not just for
quadro cards.
They are pulling an Intel in terms of price tags but they
didn´t make sure their base is safe for the future.
I had hoped for a wider than 384bit bus, e.g. something more
like a 512bit bandwidth which would
have made power of two steps in video ram more likely, e.g.
cards with 4GB, 8GB, 12GB, 16GB, etc.
To me, it seems the gtx9xx bus width comes directly from the
gtx7xx range, which was already starting
to show limits in buswidth back then.
All that said and taking tax laws and such for wrting off
hardware into account, I´d probably have to go
with a Titan, using it 2-3 years and finding myself wanting
more video ram soon anyway...
Cheers,
tim
