Soldato
-
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 Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **
- Thread starter Boomstick777
- Start date
Why does everyone keep saying Fermi was a fail ? It wasn't at all it was a great architecture for compute and for gaming when they finally fixed their mess with the 480 that was just badly built silicon, the 580 was a fantastic card.
The only fail was the 480 because of the heat it produced, but again it performed very well for that time.
Fermi was a great compute card and that's why they held their value in that segment very well. GPU technology is not just for gaming...
Also all kicking off at Pascal need to understand Nvidia really needed a compute card again to compete with Intel Knights Landing in that segment. We as consumers of gaming mainstream cards need Nvidia to do that so they can keep the mainstream prices at sensible levels or we will end up paying £1K for a mainstream mid range card if they don't.. R&D costs big money.
Also don't we all want self driving cars, smart technology and medical advancements (these may one day save your life or a loved ones) ? Compute devices will bring us science fiction to real world.
Totally agree..
Tired of the Fanboy mentality, you are not a technology enthusiast if you are fanboy.. real technology enthusiasts love to see what each company can do and how they do it. Brand loyalty is just dumb, be loyal to a company that is making advancements as they are all doing and enjoy what they make and pick the one that fits your needs at the time.
I think AMD and Nvidia are both doing great in their own market segments. Just AMD needs to step up its CPU line, but hopefully Zen will do that for them. AMD GPU's are doing great in their own way and they will bounce back in that area soon enough and claim back a bit more market share again from Nvidia.
It would also be nice to have some more companies in these markets as we had in the past, these smaller companies at the time made some massive advancements in certain areas and their technologies live on today in the AMD,Intel and Nvidia products.
In short.. being a fanboy makes you NOT a technology enthusiast.. Cheer for them all to make advancements that we want.
Soldato
Why does everyone keep saying Fermi was a fail ? It wasn't at all it was a great architecture for compute and for gaming when they finally fixed their mess with the 480 that was just badly built silicon, the 580 was a fantastic card.
The only fail was the 480 because of the heat it produced, but again it performed very well for that time.
Fermi was a great compute card and that's why they held their value in that segment very well. GPU technology is not just for gaming...
Indeed, I remember when I witched from 5850 xfire to GTX 480 SLI it was some night and day when it came to performance and number of issues I was having with drivers and games, especially around Skyrim which was what I was playing at the time.
Yep they where hot cards and sucked up a lot of power (still ran on my corsair 620w/Seasonic 650w in sli) and the ref cooler was pants (I WC mine) but the performance was there and scaled better than the 5850/5970 when overclocking I found.
Of course the fan boys ripped on the power/heat/cooler issues (and rightly so to a point) but as soon as AMD released a very hot power hungry card with a Poo cooler suddenly (the 290x eps when compared to the 780ti) it was "who cares about power consumption" "AIB cards will sort the heat and noise" issue"
As did most owners.Of course the fan boys ripped on the power/heat/cooler issues (and rightly so to a point) but as soon as AMD released a very hot power hungry card with a Poo cooler suddenly it was "who cares about power consumption" "AIB cards will sort the heat and noise" issue"
You're clearly looking at it from one side there. The people who smashed the 2900XT all of a sudden loved the 480. It happens every time one camp puts out a product with vectors for fanboy attacks.
Anyway back to Pascal, I have long suspected these first finfet cards will be disappointing. So this isn't a massive shock. I don't think it will affect market share however as I think AMD are being very careful to dance around the shortcomings of their impending offerings. So this protracted dull period will last well into 2017 still.
I totally agree, posters on both red and green teams, were quite happy to pile criticism on 5900 2900 480 290, oh look it seems to be NVidia's turn again
Maybe AMD's time building Fiji taught them enough about HBM and interposers to steer clear of them until 16/14nm matures a bit, who knows. But I do agree that this dull period as you nicely put it will probably last till the end of the year at least.
Caporegime
No GeForce Cards Based On GP100 ?
http://videocardz.com/58865/nvidia-1st-generation-pascal-speculation
http://videocardz.com/58865/nvidia-1st-generation-pascal-speculation
Last edited:
Soldato
It makes perfect sense for them to not bring GP100 to the desktop.
All of the desktop parts will more than likely just have SP SM's. they will also then introduce a new top end part which will probably be called GP102. which will have many more SP's than GP100 but be much smaller since 1/3 of the die is not being taken up by bulky DP units. Considering each DP unit is larger than an SP unit.
All of the desktop parts will more than likely just have SP SM's. they will also then introduce a new top end part which will probably be called GP102. which will have many more SP's than GP100 but be much smaller since 1/3 of the die is not being taken up by bulky DP units. Considering each DP unit is larger than an SP unit.
Great find there and a couple of pics for those who might not have clicked on that link.
Goes to show that they have actual chips in house and the DGX-1 systems built up for testing at the very least.
And here is a possible layout for the GP104 which I thought was interesting.
bru, in future could you please resize any images wider than 1280 pixels, or place them in spoiler tags as i have done this time. Thank you, setter.
Last edited by a moderator:
Soldato
Yeah, i think in the GP106, GP104 and possible GP102 parts they will not have the DP units. so i think the core layouts will be a bit different, having instead something like 48 single precision SP's per SM instead of 32 SP units and 16 DP units.
Soldato
Well the DP unit will be larger than an SP unit as it has to be able to process data with a 64bit word size. Therefore everything in it will be doubled up to accommodate said data. such as L1 cache etc.
But with the way they have made their DP units, they can only process 64bit word data only. Since if their 64bit units could accommodate 32 and 16bit word lengths then the entire card would be made out of them. Especially if it directly scales, so if 1 DP unit could process 2 x32 float or 4 x16 ints.
Volta will more than likely change this and become more GCN like where the architecture can be adjusted for different parts to add 64bit support circuits which allow SP's to fuse to work on a 64bit word length instruciton.
Something we agree on.
I think it was inevitable that nvidia would have to split their Compute parts form the gaming parts to outstrip Intel. 600mm^2 for first round on new node squarely means they don't give a damn about yields, as does the 300W, and so much of the design.
I wouldn't say the GP100 want appear as a consumer product though, it might exist like the original Titan for high end gaming and HPC use.