If this holds true then I will be getting a Maxwell. And then see what happens with CPU's.
-
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.
Nvidia Geforce 'Maxwell' Thread
- Thread starter Boomstick777
- Start date
If this holds true then I will be getting a Maxwell. And then see what happens with CPU's.
Need a date currently using HD4600 and a 120hz VA panel incoming. I need a beast and i need one soon. But there is not a hope in hell they are getting anything more than a 750ti out of me because 780ti is a joke for 120hz or 4K. You need two, £1000 for GPU's madness!
Last edited:
apologies people. but I must type this in response to drunkenmasters die foundry business advice. After this post I willl not mention anymore on this maxwell thread.
Whilst I agree Amd played the safe part by going as big as they felt was balanced enough for the 7970,, My idea is not silly , I'm not saying you make a 7970 faster because you will it so. I'm saying if you had a full fat cayman on 32nm then if silicon allowed would we have seen a slightly more beefy 7970 if it was truely meant to be a high end gpu. neither is my idea based on this post
My first post #630, clearly demonstrates my views on the 7970 and the 680, my viewpoint isn't so much obsessed by technical die and foundry engineering business practices. But your stating I'm silly for claiming 32nm cayman would have magically made 7970/680 faster defying laws of physics, isn't fair and is incorrect because you are going off on a tangent of what i was trying to get across when replying to Broomsticks original post where he mentioned he was impressed by the progress in gpus over the last few years..
In post 636 I've explained why I feel gcn 7970 doesn't look much to me like a high end architecture in comparison to the previous high end which was crippled from its true potential (cayman 6970). But then its funny how the 290xt although rumoured is exactly x2 the scale up of cayman the 6970. Now I agree theres a limit to what you can do with the silicon you work with. My feelings in post 630 are in reflection of post 636 to demonstrate why I feel that progress has been slow for full fat gpus to have materialised.
In your post #643 your last paragraph you demonstrate you fixation on scaling of architecture via manufacturing, the bit about 32nm cayman would have made no difference, not even one transistor difference to 28nm. My response to that is if the 7970 was truely a high end gpu but a safe one with less cu's for high yields(higher midrange ), then if the 6970 had been made to it's true potential then I believe amd would have have refined the 7970 architecture and core config to eek out more performance to widen the gap from a previous gen high end but still mantain a healthy yield rate . But no they didn't need to as the 6970 was so poor anyway and they didn't because as you like to remind me so many times, Amd/Nvidia knew making a big shader core gpu was hard to do during the early days of the 28nm. On top of this Amd knew Nvidia would struggle to make a gk110/gf110 so the 7970 was first marketed with intentional lowered clocks but with headroom for higher clocks. Allowing the 680 mid range gpu to be marketed with the 7970 as high end gpus, until they worked out a way to make bigger cores.
So why does the 680 get faster because of the 6970? Well if we ignore the yield rate must be drunkenmaster certified
Then if the 7970 had been set from day 1 as a 2048 core 1050-1100 mhz card, or a beefier 7970 with slightly more cores and high clocks, then I'm sure Nvidia would have scaled their 680 up during the design stage/comparison of the rival stage too. But all this is my theory and so of course holds no ground, or relevance on this maxwell thread.
Whilst I agree Amd played the safe part by going as big as they felt was balanced enough for the 7970,, My idea is not silly , I'm not saying you make a 7970 faster because you will it so. I'm saying if you had a full fat cayman on 32nm then if silicon allowed would we have seen a slightly more beefy 7970 if it was truely meant to be a high end gpu. neither is my idea based on this post
My first post #630, clearly demonstrates my views on the 7970 and the 680, my viewpoint isn't so much obsessed by technical die and foundry engineering business practices. But your stating I'm silly for claiming 32nm cayman would have magically made 7970/680 faster defying laws of physics, isn't fair and is incorrect because you are going off on a tangent of what i was trying to get across when replying to Broomsticks original post where he mentioned he was impressed by the progress in gpus over the last few years..
In post 636 I've explained why I feel gcn 7970 doesn't look much to me like a high end architecture in comparison to the previous high end which was crippled from its true potential (cayman 6970). But then its funny how the 290xt although rumoured is exactly x2 the scale up of cayman the 6970. Now I agree theres a limit to what you can do with the silicon you work with. My feelings in post 630 are in reflection of post 636 to demonstrate why I feel that progress has been slow for full fat gpus to have materialised.
In your post #643 your last paragraph you demonstrate you fixation on scaling of architecture via manufacturing, the bit about 32nm cayman would have made no difference, not even one transistor difference to 28nm. My response to that is if the 7970 was truely a high end gpu but a safe one with less cu's for high yields(higher midrange ), then if the 6970 had been made to it's true potential then I believe amd would have have refined the 7970 architecture and core config to eek out more performance to widen the gap from a previous gen high end but still mantain a healthy yield rate . But no they didn't need to as the 6970 was so poor anyway and they didn't because as you like to remind me so many times, Amd/Nvidia knew making a big shader core gpu was hard to do during the early days of the 28nm. On top of this Amd knew Nvidia would struggle to make a gk110/gf110 so the 7970 was first marketed with intentional lowered clocks but with headroom for higher clocks. Allowing the 680 mid range gpu to be marketed with the 7970 as high end gpus, until they worked out a way to make bigger cores.
So why does the 680 get faster because of the 6970? Well if we ignore the yield rate must be drunkenmaster certified
Then if the 7970 had been set from day 1 as a 2048 core 1050-1100 mhz card, or a beefier 7970 with slightly more cores and high clocks, then I'm sure Nvidia would have scaled their 680 up during the design stage/comparison of the rival stage too. But all this is my theory and so of course holds no ground, or relevance on this maxwell thread.
Last edited:
Woh, woh woh dude, hold on a sec !
The first mistake you made was calling me Broomstick.. This isn't the nursery playground and only LTMatt gets that privilege. You are not LTMatt.
The second mistake you made was taking my comment out of context. I said that GPU progress was still good whilst in comparison to CPU progress, which is on AMD's side virtually non existent, and on Intel's side generally 5% per year. The 7970 and GTX 680 were improved over their predecessors. This is what I said and that was the context I used it in. Get it right next time boy
Now we have set this straight you and DM can continue with your argument. Or maybe do it via PM so this thread doesn't get do messy with personal comments etc. (Just a suggestion).
Peace.
Woh, woh woh dude, hold on a sec !
The first mistake you made was calling me Broomstick.. This isn't the nursery playground and only LTMatt gets that privilege. You are not LTMatt.
sorry old boy
The second mistake you made was taking my comment out of context. I said that GPU progress was still good whilst in comparison to CPU progress, which is on AMD's side virtually non existent, and on Intel's side generally 5% per year. The 7970 and GTX 680 were improved over their predecessors. This is what I said and that was the context I used it in. Get it right next time boy
Apologies are indeed in order to you. You have indeed made a comparison between gpu's and cpu's progress. However you told fibbies and your comparisons differ between your first post and this recent one, in your first post you stated Amd cpus went backwards, which I feel is more accurate than your recent one!!
Now we have set this straight you and DM can continue with your argument. Or maybe do it via PM so this thread doesn't get do messy with personal comments etc. (Just a suggestion).
Peace.
Peace xx
Soldato
But you can't make bigger transistor oxide smaller with smaller transistors.
And other fascinating theories. Bad enough having dribble in the AMD threads without it spilling over into NV ones
And other fascinating theories. Bad enough having dribble in the AMD threads without it spilling over into NV ones
Associate
Ladies.I want to hear about sexy Maxwell specs not the specs of your handbags
Man of Honour
Ladies.I want to hear about sexy Maxwell specs not the specs of your handbags
Yes I have to agree with the above tbh.. debate and argue by all means but don't make it personal.
Associate
Indeed.. came in here expecting big news after all the new posts!
+1
Been doing my best to try and steer the thread back onto topic lol.
-------------------------------------------------------------------------
Nvidia may Skip 20nm and go straight to 16nm finFET
http://www.overclock3d.net/articles...ay_skip_20nm_and_go_straight_to_16nm_finfet/1
Am I correct on this, feel free to correct me. A die shrink situation (A64 90nm to 65nm) replicating the same layout but transposing onto the new half node , then you are scaling everything down in size? The gap between the source and drain of the transistor is determined by the foundry node process and the architect/engineer works with the spec. in caymans case everything was designed for 32nm which would have been a full node coming from the 40nm that they were on with barts. 32mm was cancelled so AMD had to rescale cayman back to 40nm, effectively compromising the performance by removing simds and other stuff so to make it work on 40nm. this is my point not the dms take with 28nm and 32nm.
More rumours here. http://www.kitguru.net/components/g...0nm-process-technology-jump-straight-to-16nm/
if the foundry can provide a half/node, providing the cost is good, and power savings and yield rate outweighs the final cost of manufacture, then taking advantage of the node makes sense. In the case of 20nm it probably doesn't if you intend to make mid range gpu's first. Thats why the 750 was made on the 28nm node. I'm very interested in Nvidas maxwell it seems more efficient than AMD, just they require a node that'll let it all happen. Funny really as both companies are stick in a similar situation to the 40nm to 32nm
Last edited:
Associate
Dosent the 780ti have like 2800 shaders ? ( can't remember ) so why would the 880ti have 2500 ?
Maxwell shaders are rumored to have more performance so 2500 in theory could match '2880' 780 Ti / Titan Black performance couple that with less VRAM Bandwidth but extra cache to make up for it means the new cards would be using less power on a very mature 28nm with a lot less cost to manufacturer for Nvidia.
Looks like the real high end stuff (Significant jump over current cards) will come later..
Last edited:
Well a 512 core Maxwell 750 beats a 768 core Kepler 650ti. So even with less cores it could still be a good bit faster.
Ok the 750 is clocked 100 Mhz or so faster but it does beat the 650ti by a good margin. both are on a 128 bit bus.
If Maxwell isn't around by September you could always go with one of the upcoming 'new' AMD
Apparently Charlie wrote in his secret club behind the paywall that 16nm is coming Q1 and the 880/ti/70/60 will be launched in ~September & then "subtly" shrunk to 16nm (20 with fins, remember) keeping the same names.
So if this is true (back up the salt truck), it could be a minefield trying to buy a 800-series come the spring. Hopefully in this scenario the product code or some other identifier will change.
TSMC have said they are going to be also offering an enhanced 16nm process later in 2015 called 16nm FinFet Plus, which does offer density improvements (as well as power & performance) so in the space of a year we could go from what we have now, old moldy 28nm to "true" 16nm. If this does transpire I will believe their claims of 10nm end of 2015 after all!
Who knows where this leaves GloFo, they have cut a deal with Samsung for a low-power 14nm process but it's unclear (to me) when this would come online and if a performance variety can be offered. If AMD really are going to be exclusive to GF then they could be missing out big time.
So if this is true (back up the salt truck), it could be a minefield trying to buy a 800-series come the spring. Hopefully in this scenario the product code or some other identifier will change.
TSMC have said they are going to be also offering an enhanced 16nm process later in 2015 called 16nm FinFet Plus, which does offer density improvements (as well as power & performance) so in the space of a year we could go from what we have now, old moldy 28nm to "true" 16nm. If this does transpire I will believe their claims of 10nm end of 2015 after all!
Who knows where this leaves GloFo, they have cut a deal with Samsung for a low-power 14nm process but it's unclear (to me) when this would come online and if a performance variety can be offered. If AMD really are going to be exclusive to GF then they could be missing out big time.
Last edited: