• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

The first "proper" Kepler news Fri 17th Feb?

drunkenmaster said:
Again I'll say it, the SPEED of the memory is rated at 6Gbps PER PIN, gddr5 has 32 pins, that makes the speed 192Gb/s. That gives a chip on 32bit bus 24GB/s, on a 256bus, that gives 192GB/s.
Click to expand...

True enough - and I'll admit I clearly had the wrong end of the stick here. The 680GTX using 6Gbps DDR5 does fall in line with what could be expected, and adds weight to the validity of the slide. Given this, I withdraw my previous claim that the slide is "clearly" fake.



However, falling back to the semantics for a second - whatever way you dress it, a rating of "Gb/s", or anything which is "units of data per unit time" is most certainly a bandwidth (see the definition of bandwidth):

wiki said:
bandwidth,[1] network bandwidth,[2] data bandwidth,[3] or digital bandwidth[4][5] is a measure of available or consumed data communication resources expressed in bits/second or multiples of it (kilobits/s, megabits/s etc.)
Click to expand...

So, while you may colloquially refer to the bandwidth as the "speed" rating of the chip, it's still a measure of bandwidth. In the same way, a memory "speed" in terms of Mhz is a measure of frequency, even though it is often referred to colloquially as the "speed" of the memory.
 
xsistor said:
... In other news http://www.legitreviews.com/news/12616/
Click to expand...

Interesting...

So, if this is correct, Nvidia have managed a 107% increase in transistor density going from 40nm to 28nm (GTX580 -> GTX680). This is largely in line with the 105% expected increase in packing density from the process change.

By contrast, AMD achieved 'only' a 74% increase in transistor density going from 6970 -> 7970.
 
Duff-Man said:
Interesting...

So, if this is correct, Nvidia have managed a 107% increase in transistor density going from 40nm to 28nm (GTX580 -> GTX680). This is largely in line with the 105% expected increase in packing density from the process change.

By contrast, AMD achieved 'only' a 74% increase in transistor density going from 6970 -> 7970.
Click to expand...

It's not surprising IF Nvidia gained, which, we've still got to see a real die size and specs. GF104 was relatively speaking, rushed as GF100 was struggling so badly, it also had a not wildly different core but slightly different shader setup, so it wasn't so much a tweaked/slimline part made from a working top end core. Basically you expect Kepler to be(without major screwups) better all around than Fermi. It's still a slimline somewhat midrange part, and even by those guessed diesizes, puts it behind PitCarn by a noticeable amount.

We really have to see what it actually ends up being, a DP monster, or a DP neutered card, which would explain a lot in terms of lower bandwidth/size. You also have different parts of a core, different types of logic which have differing transistor densities. Does the 384bit bus and pinout both increase die size and has a lower transistor density, while using more power....... but then I won't be remotely surprised to see surround gaming res with Tahiti spanking the hell out of GK104. It's a cost and isn't very apparent at 1080p, but 3x1080p could/will be another story.

Pitcarn has, basically gone 100% transistor density increase, gone silly low power, gone faster and smaller die all in one package, thats comparing fairly like for like in terms of a cut down not very compute part.
 
kitch9 said:
The 680 smacks of a mid-range part that is clocked to within an inch of its life to compete with the 7970 to me.

I might be wrong.
Click to expand...
Do we know the final GPU clocks yet for GK104? Most of what I have read indicates 700-800MHz, which seems very low for 28nm. 800MHz is not even high for the old 40nm process, where even my old GTX480 could exceed that.

It seems likely that GK104 will have lower clocks, lower memory bandwidth, plus lower transistor counts, yet it will still compete favorably against the 7970. This means NVidia's core architecture is likely superior to Tahiti's. Perhaps GCN has scaling limitations, because AMD's own 7870 comes close to the 7900 series when overclocked, even with significantly less shaders and memory bandwidth.

We will find out soon.
 
Last edited:
Duff-Man said:
Interesting...

So, if this is correct, Nvidia have managed a 107% increase in transistor density going from 40nm to 28nm (GTX580 -> GTX680). This is largely in line with the 105% expected increase in packing density from the process change.

By contrast, AMD achieved 'only' a 74% increase in transistor density going from 6970 -> 7970.
Click to expand...

It could be one major contributor to why we're hearing good things about GK104. The design sounds efficient from all I've heard.

Architecture will affect transistor density. I was convinced that was the case with previous NVIDIA GPUs if anyone cares to look that far back. Compute-heavy architectures appear to be less heavy on transistor density and that seems reasonable. For compute you'll have local memories and such which will impact this.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom