GTX 880 and 870 to release in November 2014

[jmaster299]

Nvidia and their low memory buses... disappointed.

Which does absolutely nothing to hurt the performance of their cards because of their higher memory clocks.

[ACatWithThumbs]

Which do absolutely nothing to hurt the performance of their cards because of their higher memory clocks.

They could never make up with the clock for the slow 256bit bus and they are for sure not making a 10Ghz clock on that bus.

That's why I call this hole 256bit bus on GTX870/880 BS.

The 780 Ti has the most memory bandwidth of all GPUs on the market already so why should they downgrade it and cripple the performance of the new cards?

[RelentlessAF]

I'm upgrading from a 660 so yeah it will be a huge jump to me regardless of the low memory bus.

 

EDIT: Plus I still play at 1080p...

[ACatWithThumbs]

I'm upgrading from a 660 so yeah it will be a huge jump to me regardless of the low memory bus.

 

EDIT: Plus I still play at 1080p...

It's fake the card wouldn't even get 780/ti performance with such a memory bus.

 

[jmaster299]

They could never make up with the clock for the slow 256bit bus.

That's why I call this hole 256bit bus on GTX880/870 BS.

The 780 Ti has the fastest memory bus of all GPUs on the market so why should the downgrade it and cripple the performance of the new cards?

 

Yes, they can. A 770 4GB can already use 3.8GB of it's 4GB of VRAM on it's 256-bit bus. No, I'm not going to go dig up the links because this issue has been asked and answered a thousand times on this forum. Both the math and real world testing show the 770 using 3.8GB of VRAM. For any single 870 or 880 card, a useable 3.8GB of VRAM is way more than enough, since in a single card setup you will hit the processing power limit before you run into situation where the game will need that much VRAM. There are also no games on the market that use that much VRAM, even at 4K, so there is no risk of running out of VRAM even in a SLI set up.

Also, there reason why they are not using the 780 Ti platform is because the new cards are going to be based on the GK104, not the GK110. The new GM204 is supposed to offer a 35% performance increase over the GK104. That will allow Nvidia to take a full utilized chip, like what is currently used in the 770, slap an 880 label on it, and call it a day. That same chip will be used for all the cards down to the 860, just with some of it's cores disabled. That means the performance jump from a 780 to an 880 will be minimal, but the 880 will provide that performance while using less power than the 780, producing less heat than the 780 and will be sold at a price point cheaper than a 780.

But the 800 series being nothing more than a refresh is intentional. Within 6-9 months we should be getting Nvidia cards using 16nm chips, giving us a massive performance increase.

For the top end cards with in the refresh series, the next Titan card will be using what they are calling the GM200 chip, and will feature 4000 CUDA cores and a 512-bit memory bus. If Nvidia follows along with what they've been doing recently, that means we will see an 880 Ti with the same specs, but likely with less total VRAM and no floating point calculations for rendering. That card will put every single other single GPU on the market to shame.

So stop spreading the bullshit about the 256-bit bus some how limiting performance. It's 100% not true. Dig through my posts in various threads about the 770 if you want to see the math and the benchmarks.

[RelentlessAF]

It's fake the card wouldn't even get 780/ti performance with such a memory bus.

 

I'd figure that, especially with how the 750ti showed the advancements of Maxwell, with less power consumption they should at least be able to pump out the same performance as the 700 series.

[ACatWithThumbs]

Yes, they can. A 770 4GB can already use 3.8GB of it's 4GB of VRAM on it's 256-bit bus. No, I'm not going to go dig up the links because this issue has been asked and answered a thousand times on this forum. Both the math and real world testing show the 770 using 3.8GB of VRAM. For any single 870 or 880 card, a useable 3.8GB of VRAM is way more than enough, since in a single card setup you will hit the processing power limit before you run into situation where the game will need that much VRAM. There are also no games on the market that use that much VRAM, even at 4K, so there is no risk of running out of VRAM even in a SLI set up.

Also, there reason why they are not using the 780 Ti platform is because the new cards are going to be based on the GK104, not the GK110. The new GM204 is supposed to offer a 35% performance increase over the GK104. That will allow Nvidia to take a full utilized chip, like what is currently used in the 770, slap an 880 label on it, and call it a day. That same chip will be used for all the cards down to the 860, just with some of it's cores disabled. That means the performance jump from a 780 to an 880 will be minimal, but the 880 will provide that performance while using less power than the 780, producing less heat than the 780 and will be sold at a price point cheaper than a 780.

But the 800 series being nothing more than a refresh is intentional. Within 6-9 months we should be getting Nvidia cards using 16nm chips, giving us a massive performance increase.

For the top end cards with in the refresh series, the next Titan card will be using what they are calling the GM200 chip, and will feature 4000 CUDA cores and a 512-bit memory bus. If Nvidia follows along with what they've been doing recently, that means we will see an 880 Ti with the same specs, but likely with less total VRAM and no floating point calculations for rendering. That card will put every single other single GPU on the market to shame.

So stop spreading the bullshit about the 256-bit bus some how limiting performance. It's 100% not true. Dig through my posts in various threads about the 770 if you want to see the math and the benchmarks.

You are completely missing my point.

I'm totally aware that a lower memory bus with higher clock performs better than a big bus with low clock.(As you can see with the 780Ti compared to the R9-290X)

But to get the memory bandwidth of at least a 780 they would need a 8-9GHz clock just to not lose performance.

That's like running the 780 with a 4,5GHz memory clock.

We are talking here over 100GB/s difference which makes about 10-20Fps difference in real world performance.

If this is true then it's not a GTX870/880 it would be more like a GTX860/ti.

 

[flibberdipper]

Just give us the damn things already.

[jmaster299]

You are completely missing my point.

I'm totally aware that a lower memory bus with higher clock performs better than a big bus with low clock.(As you can see with the 780Ti compared to the R9-290X)

But to get the memory bandwidth of at least a 780 they would need a 8-9GHz clock just to not lose performance.

That's like running the 780 with a 4,5GHz memory clock.

We are talking here over 100GB/s difference which makes about 10-20Fps difference in real world performance.

 

You are completely missing the point. The card in no way requires the same memory bandwidth because the difference in memory bandwidth between a 770 4GB and a 780 3GB has absolutely no impact at all on performance. For both cards you will reach the processing power limitation of the card before you reach the memory bandwidth limitation. Even games at 4K are incapable of exceeding the memory bandwidth of a 4GB 770.

[ACatWithThumbs]

You are completely missing the point. The card in no way requires the same memory bandwidth because the difference in memory bandwidth between a 770 4GB and a 780 3GB has absolutely no impact at all on performance. For both cards you will reach the processing power limitation of the card before you reach the memory bandwidth limitation. Even games at 4K are incapable of exceeding the memory bandwidth of a 4GB 770.

I'm not talking about the bandwidth effecting the amount of ram you can use.

I'm talking about memory bandwidth directly effecting GPU performance.

Higher memory bandwidth has a direct effect on your GPU performance.

Overclocking my 256bit bus from 6Ghz to 7Ghz gave me an 4Fps increase on my GTX670 and it just increased my memory bandwidth by 20GB/s.

With this we are talking more than 100GB/s this would have a huuuuge effect on GPU performance.

 

[jmaster299]

I'm not talking about the bandwidth effecting the amount of ram you can use.

I'm talking about memory bandwidth directly effecting GPU performance.

Higher memory bandwidth has a direct effect on your GPU performance.

Overclocking my 256bit bus from 6Ghz to 7Ghz gave me an 4Fps increase on my GTX670 and it just increased my memory bandwidth by 20GB/s.

With this we are talking more than 100GB/s this would have a huuuuge effect on GPU performance.

 

We are talking about the same thing, yet you are still missing the point. Memory bandwidth is just one small factor in GPU performance. The performance gap between the 780 and the 880 will be made up by the more powerful GM204 chip, and with that will likely come a higher memory clock as well. But the 880 in no way requires the same memory bandwidth as a 780 in or to be able to meet or beat the 780's performance. That's because, again, it will be getting it's performance from the more powerful graphics chip.

You have to remember that the 880 is not an upgrade for a 780, it's a replacement that Nvidia will be able to offer at a cheap price than the 780. 

[ACatWithThumbs]

We are talking about the same thing, yet you are still missing the point. Memory bandwidth is just one small factor in GPU performance. The performance gap between the 780 and the 880 will be made up by the more powerful GM204 chip, and with that will likely come a higher memory clock as well. But the 880 in no way requires the same memory bandwidth as a 780 in or to be able to meet or beat the 780's performance. That's because, again, it will be getting it's performance from the more powerful graphics chip.

You have to remember that the 880 is not an upgrade for a 780, it's a replacement that Nvidia will be able to offer at a cheap price than the 780. 

I'm totally aware of this but if you look at what we know about 28nm Maxwell it won't make up for it unless it has even more Cuda Core which is highly unlikely.

Maxwell has less CUDA Cores per SMM unit while they do perform better you would need more to make up for the lack of bandwidth and in 28nm there is no chance for this to happen.

3000 Cuda Core Maxwell is just not going to happen on 28nm.

And this would also make no sense from business perspective.

Why should Nvidia change to 28nm Maxwell if Kepler is doing perfectly fine there is simply no need for 700 series replacements unless it has a huge performance increase.(They barely released even any 700 series cards )

And making a larger GPU die is more expensive then making a bigger memory bus so this logic also wouldn't work especially since they already have 386bit/7ghz in mass production.

Also where do all this rumors about 20nm not being ready come from TSMC has it production ready for months now and they're pretty open about it.

 

[Commander Llama]

This could be why the GTX780Ti I just ordered keeps price dropping?

[jmaster299]

I'm totally aware of this but if you look at what we know about 28nm Maxwell it won't make up for it unless it has even more Cuda Core which is highly unlikely.

Maxwell has less CUDA Cores per SMM unit while they do perform better you would need more to make up for the lack of bandwidth and in 28nm there is no chance for this to happen.

3000 Cuda Core Maxwell is just not going to happen on 28nm.

And this would also make no sense from business perspective.

Why should Nvidia change to 28nm Maxwell if Kepler is doing perfectly fine there is simply no need for 700 series replacements unless it has a huge performance increase.(They barely released even any 700 series cards )

And making a larger GPU die is more expensive then making a bigger memory bus so this logic also wouldn't work especially since they already have 386bit/7ghz in mass production.

Also where do all this rumors about 20nm being not ready come from TSMC has it production ready for months now and they're pretty open about it.

 

There are multiple different pieces of information floating around, most of it has been posted on these forums. Take the time to actually look it up and read it like the rest of us have before making your assumptions. The refreshed 28nm chips are supposed to offer a 35% gain in performance over their Kepler counterparts. As in a 1536 CUDA core GM204 will offer about a 35% performance gain over a 1536 CUDA core GK104. What we don't know at this point is the number of CUDA cores that Nvidia will be packing on each chip. What will more than likely happen is the GM204 for things like the 870 and 860 will run the same number of cores as their 700 series counterparts with the 880 running something along the lines of 1800-2000 CUDA cores. Add the performance gain per core over the GK chips, and that will meet or beat the performance of the current 780.

The only chip that would possibly need a larger GPU die is the 880, but the die will still be smaller than the massive die that the current 780 has with it's huge GK110 chip. It will be cheaper to produce a slightly larger GM204 chip for the 880 and all the standard GM204 chips for the 870 and 860 will be cheaper than their 700 series counterparts. That's just a normal part of production, manufacturing costs naturally decrease over time.

As far as the issue with the 20nm chips, yes TSMC is ready to ship them now, but it's too late for them to make any sense for either Nvidia or AMD to use. The delay has already done it's damage. TSMC has also moved on to perfecting their 16nm fabrication process, which is supposed to be ready for mass production by the first couple of months of 2015. Logically and financially it makes more sense for Nvidia and AMD to simply skip the 20nm chips, since the 16nm chips are so close to being ready. This move is also backed up by the fact that Nvidia's deal to get first delivery of 20nm chips has disappeared and TSMC delivered the first 20nm chips to Apple instead.

If Nvidia is going to be able to GPUs on the market with 16nm chips with in the next year, it makes no sense for them to fill that 12 month gap with 20nm chips. Their profit margin on the refreshed 28nm chips will be much higher than selling more expensive 20nm chips, but their customers who want to upgrade from Kepler can get a reasonable gain in performance from the 28nm Maxwell chips, and get that upgrade at a lower price than the current 700 series cards. 12 months down the road they can release their 16nm chips to the market and charge a premium price for them because they should offer a very significant performance gain over both Kepler and Maxwell 28nm chips.

[Altecice]

If this is true, and they are cheaper and more powerful than the 700 series is. Oh my god.

 

This is Nvidia we are talking about... not a chance in hell.

[BarisERY]

Well, NVidia has a countdown and it says ''THE ULTİMATE İS COMİNG'' ;D

[dmegatool]

Well, NVidia has a countdown and it says ''THE ULTİMATE İS COMİNG'' ;D

Could be totally something else. Like a Shield tablet or whatever... 

 

Nvidia countdown

[TheProfosist]

Meh I'd rather go with better power consumption IMO. We need more single slots.

I NEED SINGLE SLOTS. WHERE THE HELL ARE MY SINGLE SLOTS.

Dude I totally agree. There isn't on single slot 750 or 750 to, WTF.

This is Nvidia we are talking about... not a chance in hell.

That's what supposed to happen every generation. The prices are basically set in stone and you get a better card for the same money.

[jmaster299]

Dude I totally agree. There isn't on single slot 750 or 750 to, WTF.

That's what supposed to happen every generation. The prices are basically set in stone and you get a better card for the same money.

You mean like the one Galaxy has announced?

http://www.techpowerup.com/201646/galaxy-gtx-750-ti-darbee-edition-graphics-card-pictured.html

[TheProfosist]

You mean like the one Galaxy has announced?

http://www.techpowerup.com/201646/galaxy-gtx-750-ti-darbee-edition-graphics-card-pictured.html

Interesting i didnt see that one out of the many they had. I am already super interested in the Darbee Edition one.

[BenD]

i'll probarly skip the 800 series to, my 680's are still doing a descend job

[Hidden Orange]

Everything... Everything happens during November... Everything supersedes my birthday...

[AlwaysFSX]

This is Nvidia we are talking about... not a chance in hell.

You don't know that... Maybe they're nice this time around.

[Hidden Orange]

You don't know that... Maybe they're nice this time around.

[Deletive]

lol, it's either 2560 cores or 2688 cores. 35% better than a titan non-black or 30% better than a titan non black if it has 2560 cores.

 

The gtx 880 will knock the prices of GPU's down; althought they know people will buy 800$ gpu's most people will still choose a 780ti for like lets say 150$ less than something that is 20% better than a 780ti. Ir's not practical and sorta stupid when they can price is at 550$ and drop prices of AMD cards.

 

anyone saying 256bit won't be enough. If you look at 1440-1600p reviews of the 750ti. with a well rounded engine it can pull ahead. infact it mostly pulled ahead of the 650ti at higher resolutions and sometimes even beat out cards with double the bus rate of 256bit; thats at 86gb/s.

 

256 bit will be a lot more than 256 bit for kepler. 256 bit will probably have 224.32gb/s through the bus thats 56gb/s less than a 780. that mean seem like a lot but let me put it in nvidia's terms:

 

". In addition, the large 2MB L2 cache configuration (larger than any previous GPU design) is highly effective at reducing memory bandwidth demand and ensuring that DRAM bandwidth is not a bottleneck."

 

so basically I can see 30gb/s of that be not much of an issue.