A little disappointed in Nvidia

[Muppet]

Originally Posted by Johnny C

20%~30%?

I'm sure they'd be happy with 15%......

I already thought the average is 10-15% in games. Some it is more and some the 5870 is faster. At least that is the impression I got from reading the majority of reviews.

[shadow001]

Originally Posted by Muppet

I already thought the average is 10-15% in games. Some it is more and some the 5870 is faster. At least that is the impression I got from reading the majority of reviews.

I meant an extra 20~30% faster than the current GTX480,by enabling the other 32 shaders and 4 texture units,clock the overall processor nearer to 800Mhz core and use the fastest GDDR 5 memory on the market,clocked at 1200Mhz.


Just on the memory bandwith alone,it would go from 171GB/sec on the current GTX480 to about 205GB/sec,it would also get decent increases in raw shader math,more tesselation power,more raw fillrate and texturing speed would see a boost as well.


The trick is pulling the above while staying within the same power consumption limits and thermals of the current version,with better yeilds,and that might only be doable once the 40nm process matures some more and the physical layout of the chip is improved further still(IE new masks).

[Viral]

Not sure why everyone thinks they have to wait for 32/28nm to be able to do a 512 core part, a B1 revision on 40nm if done properly could do it with the same or even lower power draw. It's a big die, but that doesn't automatically mean it must run hot.

[Muppet]

Originally Posted by shadow001

I meant an extra 20~30% faster than the current GTX480,by enabling the other 32 shaders and 4 texture units,clock the overall processor nearer to 800Mhz core and use the fastest GDDR 5 memory on the market,clocked at 1200Mhz.


Just on the memory bandwith alone,it would go from 171GB/sec on the current GTX480 to about 205GB/sec,it would also get decent increases in raw shader math,more tesselation power,more raw fillrate and texturing speed would see a boost as well.


The trick is pulling the above while staying within the same power consumption limits and thermals of the current version,with better yeilds,and that might only be doable once the 40nm process matures some more and the physical layout of the chip is improved further still(IE new masks).

Understood. That would be very impressive if they could pull that one off. I would have thought only another revision would get that sort of performance increase though. That is if they do make a 485 and it is a new revision, rather than just a new model with more shaders and texture units unlocked.

[shadow001]

Originally Posted by Muppet

Understood. That would be very impressive if they could pull that one off. I would have thought only another revision would get that sort of performance increase though. That is if they do make a 485 and it is a new revision, rather than just a new model with more shaders and texture units unlocked.

I don't think it's a matter of if they do it and more of a matter of when they do,as ATI will no doubt have refresh parts out in the next couple of months,so it's easy to assume they'll aim to take back the single GPU speed crown from the GTX480,so a revised B1 stepping of fermi is the shortest and fastest way to counter that,and there's is more speed to be extracted from fermi once all 512 shaders are enabled,running at higher core speeds and packing the fastest GDDR 5 ram afterall,but they do need to keep the overall power consumption and heat output the same or lower as the current model with 480 shaders and lower core clocks.


The 28nm process is no where near ready for mass production and is more likely something that will be a factor next year,and when using that process,both GPU makers can further increase the transistor budgets another 50% and keep the production costs the same as long as yeilds are good,as well as keeping thermal and power consumption within limits.....


That's when i expect the next generation that can double the performance of current cards to really begin,so until that point,it's most likely relatively smaller enhancements to both the current Fermi and Cypress architectures,while still using the 40nm process,at least until the end of the year.

[Muppet]

Originally Posted by shadow001

I don't think it's a matter of if they do it and more of a matter of when they do,as ATI will no doubt have refresh parts out in the next couple of months,so it's easy to assume they'll aim to take back the single GPU speed crown from the GTX480,so a revised B1 stepping of Fermi is the shortest and fastest way to counter that,and there's is more speed to be extracted from Fermi once all 512 shaders are enabled,running at higher core speeds and packing the fastest GDDR 5 ram after all,but they do need to keep the overall power consumption and heat output the same or lower as the current model with 480 shaders and lower core clocks.


The 28nm process is no where near ready for mass production and is more likely something that will be a factor next year,and when using that process,both GPU makers can further increase the transistor budgets another 50% and keep the production costs the same as long as yields are good,as well as keeping thermal and power consumption within limits.....


That's when i expect the next generation that can double the performance of current cards to really begin,so until that point,it's most likely relatively smaller enhancements to both the current Fermi and Cypress architectures,while still using the 40nm process,at least until the end of the year.

I do agree. There really isn't any other choice for either Nvidia or ATI. Mostly because of all the issues with the 32nm process if it is true. If they are then I don't think we will see any 28nm chips until well into next year. It looks like Intel is the only one to really get the 32nm process up and running.

[Ninja Prime]

Originally Posted by Viral

Not sure why everyone thinks they have to wait for 32/28nm to be able to do a 512 core part, a B1 revision on 40nm if done properly could do it with the same or even lower power draw. It's a big die, but that doesn't automatically mean it must run hot.

B1 will make some difference, but not a lot. It might just get them up to 750 mhz 512 shaders like they planned, but then thats only going to be ~10% faster than the GTX 480 is now. It will still be hot as it is now, if not hotter. 32nm was canned btw, have to wait for 28nm, which is a long way out, 2011 at the earliest looks like now.

So basically, if you're waiting for a real NV refresh, you're looking at about a year.

[shadow001]

Originally Posted by Muppet

I do agree. There really isn't any other choice for either Nvidia or ATI. Mostly because of all the issues with the 32nm process if it is true. If they are then I don't think we will see any 28nm chips until well into next year. It looks like Intel is the only one to really get the 32nm process up and running.

Actually,there's a difference between the processes used for CPU's and the ones used for GPU's,in that the version for CPU production is a high performance variation that is partially responsable for allowing CPU's to hit the clocks they do while keeping the power and thermal issues under control,but it's also one of the reasons why CPU's cost more in retail...Anywhere from 100$ per chip to over a 1000$ for the very highest end versions.


The fabrication processes used in GPU's are called "Bulk" variants,so they're not quite as good as the high performance variants for CPU's,but much cheaper and GPU's don't depend too much on the actual clock speeds to hit a certain performance,but mostly on the architecture itself and how it's set up in terms of parallism(more ROP's,more texture units,more shaders,wider busses,etc..),and as we've seen for a while now,clock speeds for GPU cores haven't gone up that much in the past 4~5 years,staying in that 600 to 800 Mhz range for quite a while now.


Generally speaking,it's been ATI using higher clocks,but smaller transistor budgets,and Nvidia doing the reverse,with GPU's using a higher transistor budget,but lower clocks in general.


The 32nm high performance fabrication processes are very much a go for CPU production even at Global foundries,just that it's not suited for GPU use,which is what both global foundries and TSMC cancelled....The "bulk" variation at those same 32nm sizes,supposedly from lack of interest from customers,so they're shifting to the half node,28nm "Bulk" variation instead.

[john19055]

I not really disappointed in nvidia ,because I done seen it comeing,because of the lateness and the hush hush on the benchmarks.I knew it would be hot and draw a lot of power and I knew they would need a big heat sink to keep it cool,so I knew it would be loud and because of not so many made the price would be high.I keep hearing $79 more then a HD5870,I don't know where he is getting that from but because of lack of product the GTX 480 will go up in price to around $600 or more just like the HD5970 has went up. .It is a fast card,if you can get pass all the other things,but I am tired of a loud computer.My three BFG GTX 280 OCX was fast but it did get loud when the fans went past 50%.I know the refrence design of the HD5870 can get louder then I wanted ,that why I went with the Gigabytes heatpipe and dual fan solution because it is quite and hardly ever hits 50c when stressing the card,and the fan just went to 23%.But i would'nt call the GTX 480 a flop because it is faster in most all games compared to a HD5870.I would have been consider a nvidia fan,that all I have really owned since the 9800pro.even tho I have bought a few ATI products when building computers for friends and family.The 9500pro ,2900XT,1950XT,HD4850 .I just think ATI has the best part out as of right now price wise,because you can shop around and useally find the HD5850 for $280 and two of them would be $560 and they are fast and overclock great.To me the HD5970 is to high,when two HD5850 will do around the same performance.I just don't look for nvidia to make a lot of the GTX 480 unless they do a respin and get it more effecient.Even the GTX 285 is getting harder to find and price has went up.It is a good time to sell your GTX 280 or GTX 285 ,I sold two of my GTX 280's and only lost $15 from what I paid for them.

[shadow001]

Here it is:


http://vr-zone.com/forums/606629/glo...k-process.html

As had been though for some time now, due to a complete lack of roadmap updates, Globalfoundries today confirmed that they have ditched their 32-nm bulk HKMG process. Instead, Globalfoundries will move straight to their 28-nm HKMG process, which apparently will still enter volume production in 2H 2010. Globalfoundries does have a 32-nm process, but of the SOI+HKMG variety, that will be used for AMD's upcoming CPUs and APUs (CPU+GPU combination); the 32-nm SOI+HKMG process will enter volume production in 1H 2011.

The high performance 32nm SOI+HKMG process is still a go at global foundries.....It's the process that uses copper interconnects between transistors,hence the need for the SOI part(silicon on insulator),as copper isn't compatible with silicon without using an insulator in between.



The bulk 32 nm version,and the same goes for the 28nm version,still uses aluminum interconnects between the transistors which is compatible with silicon and doesn't require an insulator at all,and something that both Intel and AMD stopped using in their CPU's years ago.....Intel uses it's own variation of SOI with copper interconnects,while the SOI version that AMD uses was actually developed by IBM originally.