-
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
Soldato
http://www.anandtech.com/show/10193/micron-begins-to-sample-gddr5x-memory
With an and saying they sampled 12 days ago with vole starting in the summer I really cannot see retail gtx 1080 with gddr5x before September. If production starts in July they need time to create invantory
With an and saying they sampled 12 days ago with vole starting in the summer I really cannot see retail gtx 1080 with gddr5x before September. If production starts in July they need time to create invantory
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 30,323
I think we're all looking forward to Computex, a bit of solid info to chew over!! I'm obviously guessing, but if the 1070/1080 shader count is correct I'd probably bet on it!
They do still have a bit of competition in AMD still, they ain't dead yet!!
Thought so
Love this place.
One of the sources for a lot of the rumours was claiming 6144 shaders on GP100 even right before GTC
I didn't write the article bud, no idea. Just gave a link to the latest gossip. Write to the editor if you are concerned.
Man of Honour
Does this mean that we can expect a Titan X replacement later on in the year?
Soldato
Yes, No, Maybe.
I don't know. Can you repeat the question?
Hopefully it will be better than that. We had the same if not better with Maxwell without a die shrink. Core count seems low on those specs for the 1080 and 1070. Surely it is more likely that the 1080 will have around 3000 cores.
The 980 had 2048 cores compared to full maxwell having 3072 so why would the 1080 have 2048 cores compared to big pascal having 3840 cores?
You would have thought the 1080 would have around 2600-2800 cores.
Definitely I hope it is better than that. To me that would be a fail, but judging from majority of posters here, they would be lapping it up.
980Ti beater that is cooler, quieter, uses less power for under £500... I can hear them panting from here!
Soldato
One of the sources for a lot of the rumours was claiming 6144 shaders on GP100 even right before GTC
Well it does have 5760 shaders, but 1/3 of them are larger DP shaders.
Associate
It really doesn't matter if any of these leaked rumors are true or not.....they are designed to build anticipation of what may (or may not) be coming just around the corner and if they can get into your head that they have something juicy just around that corner then it will make you wait for it.
Sony did exactly the same with the PS2 when Sega had released the Dreamcast. The Dreamcast was a good console, but Sony had drip fed "Rumors" to the public which made them hold out for the PS2....and it worked.
What with all the banter about AMD being first out of the gate with Polaris, Nvidia are probably leaking bits and pieces to feed the faithful. It's a great marketing strategy which is proven to work.
Sony did exactly the same with the PS2 when Sega had released the Dreamcast. The Dreamcast was a good console, but Sony had drip fed "Rumors" to the public which made them hold out for the PS2....and it worked.
What with all the banter about AMD being first out of the gate with Polaris, Nvidia are probably leaking bits and pieces to feed the faithful. It's a great marketing strategy which is proven to work.
Permabanned
- Joined
- 8 Dec 2015
- Posts
- 1,485
Sony did exactly the same with the PS2 when Sega had released the Dreamcast. The Dreamcast was a good console, but Sony had drip fed "Rumors" to the public which made them hold out for the PS2....and it worked.
Sega messed up with the Stepping Stone that was the 32x, they also messed up with the Saturn which got Blown away by the PS1. By the time the Dreamcast came even the most loyal of Sega fans had had enough. They deserted Sega forever never to return and Sega have not been the same since, they failed to adapt to a more mature audience which Sony was well aware of.
The PS1 was so cool i ordered mine straight from Japan for a cool £900 back in 1994 with Ridge Racer and Tekken. People in the UK hadnt even heard of one for atleast 12 months after (no internet back then only Grey import) and it made it even more desirable to have one. Not to mention the PAL UK ones ran at an awful stutterfesting 50hz instead of NTSC much smoother 60hz. Same as the N64.
Associate
I still have my PS1 and it still works. Last time we got it out to have a look at it, the graphics looked rather poor. Mind you it was amazing in it's day (Just over 20 years ago now, how time flies).
Last edited:
Soldato
NVIDIA Readies Three Pascal GP104 GPUs To Replace GeForce GTX 980 Ti, GTX 980 and GTX 970 in June – GP104 SKUs Leaked
Read more: http://wccftech.com/nvidia-pascal-gp104-gpu-skus-june/#ixzz45c7EK8TJ
Very interesting. That probably explained Nvidia stopped GTX 980 Ti production a while ago as there will be GTX 980 Ti Pascal successor use part GP104-400-A1 which used 8GB GDDR5X so GP104-200-A1 with 8GB GDDR5 wont be reserved for GTX 1070 but only for 1080. I guess GP104-150-A1 could use either 4GB or 7.5GB GDDR5 memory.
If 980 Ti replacement will be called either 1080 Ti then it will be the first time it not used the big pascal chip but instead use mid range chip. If not then 1080 Ti will use big pascal chip and it would make sense to call 3rd GP104 400 SKU as GTX 1080X?
Read more: http://wccftech.com/nvidia-pascal-gp104-gpu-skus-june/#ixzz45c7EK8TJ
Very interesting. That probably explained Nvidia stopped GTX 980 Ti production a while ago as there will be GTX 980 Ti Pascal successor use part GP104-400-A1 which used 8GB GDDR5X so GP104-200-A1 with 8GB GDDR5 wont be reserved for GTX 1070 but only for 1080. I guess GP104-150-A1 could use either 4GB or 7.5GB GDDR5 memory.
If 980 Ti replacement will be called either 1080 Ti then it will be the first time it not used the big pascal chip but instead use mid range chip. If not then 1080 Ti will use big pascal chip and it would make sense to call 3rd GP104 400 SKU as GTX 1080X?
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 30,323
If that's the case then hopefully they'll at least be keeping the same pricing structure. I also wouldn't be expecting a great jump in performance (with the Ti aiming to edge out the top Polaris 10), more of a functionality update and less heat
Soldato
GDDR5X in the 980Ti replacement? Is it me or will those be short lived? Cards using new memory (not GDDR5x which seems like a stop gap) aren't far off i thought?
Soldato
if its got GDDR5X then you wont get 1 before September - micron sampled 2 weeks ago with mass production not due to start till july
remember Samsung sampled HBM2 in jan this year and mass production has started - hence why we saw GP100 already.
really putting money on a die shrunk Maxwell .
remember Samsung sampled HBM2 in jan this year and mass production has started - hence why we saw GP100 already.
really putting money on a die shrunk Maxwell .
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Are people genuinely, honestly unable to see a completely BS story for what it is.
If you design a card to work with lets say a 256bit bus, gddr5 and have lets say 256GB/s of bandwidth... you think there will be a random version with around 400GB/s of bandwidth? If that architecture is designed to need around 256GB/s because it's got an appropriate amount of shaders/rops/tmus to be fed by that bandwidth then 400GB/s will do nothing for it. Likewise if a card is designed to required 400GB/s, then 256GB/s will cripple it and the card would be utterly worthless.
You will not get the same core using completely different types of memory, you can't design a given architecture to work with two completely different levels of bandwidth and gddr5x on the same bus width would use significantly more power than GDDR5. The idea is to enable you to use half the bus width and half the chips at nearly twice the speed. The only way it could work is if they taped out an entirely separate GP104x core with half the bus width of the normal GP104... at which point the end user sees basically no difference at all in performance.
Now what are the chances Nvidia will spend millions and millions to tape out a second version of a core that uses more expensive memory to provide exactly the same performance, absolutely none.
The only time you see different memory used is on the extreme low end where it's more about display outputs than actual performance and the difference is huge. It's also happened for teh 4870/50, the 50 didn't have GDDR5 mostly due to the availability and performance was pretty different at higher resolutions. However that was also on a top end card, like Fiji using HBM and having volume issues. The midrange cards as GP104 is, is the biggest segment and requires big volume. If you make 2 GP104's for the midrange you'd have one completely unwanted card and one wanted but with no volume. You can just about do it at the high end, you do it at the low end for margins/volume/the fact that performance doesn't matter as much, but in the midrange cards it makes absolutely zero sense.
Think about it this way, if Nvidia hopes to sell 20mil GP104 based cards and they all performed as you'd expect them to, people buy them. Now if you split them into a high performance and low performance gddr5/gddr5x versions, one with 20-30% better performance at high resolution(if you did do this the core would be designed to require more bandwidth and so the lesser version would be bandwidth starved at anything but 1080p)... you automatically reduce demand for GP104. If you can only make 3mil gddr5x versions and 17mil gddr5 versions... but you've killed demand for the latter by the former existing, you are going to kill sales as everyone waits for the former.
If you design a card to work with lets say a 256bit bus, gddr5 and have lets say 256GB/s of bandwidth... you think there will be a random version with around 400GB/s of bandwidth? If that architecture is designed to need around 256GB/s because it's got an appropriate amount of shaders/rops/tmus to be fed by that bandwidth then 400GB/s will do nothing for it. Likewise if a card is designed to required 400GB/s, then 256GB/s will cripple it and the card would be utterly worthless.
You will not get the same core using completely different types of memory, you can't design a given architecture to work with two completely different levels of bandwidth and gddr5x on the same bus width would use significantly more power than GDDR5. The idea is to enable you to use half the bus width and half the chips at nearly twice the speed. The only way it could work is if they taped out an entirely separate GP104x core with half the bus width of the normal GP104... at which point the end user sees basically no difference at all in performance.
Now what are the chances Nvidia will spend millions and millions to tape out a second version of a core that uses more expensive memory to provide exactly the same performance, absolutely none.
The only time you see different memory used is on the extreme low end where it's more about display outputs than actual performance and the difference is huge. It's also happened for teh 4870/50, the 50 didn't have GDDR5 mostly due to the availability and performance was pretty different at higher resolutions. However that was also on a top end card, like Fiji using HBM and having volume issues. The midrange cards as GP104 is, is the biggest segment and requires big volume. If you make 2 GP104's for the midrange you'd have one completely unwanted card and one wanted but with no volume. You can just about do it at the high end, you do it at the low end for margins/volume/the fact that performance doesn't matter as much, but in the midrange cards it makes absolutely zero sense.
Think about it this way, if Nvidia hopes to sell 20mil GP104 based cards and they all performed as you'd expect them to, people buy them. Now if you split them into a high performance and low performance gddr5/gddr5x versions, one with 20-30% better performance at high resolution(if you did do this the core would be designed to require more bandwidth and so the lesser version would be bandwidth starved at anything but 1080p)... you automatically reduce demand for GP104. If you can only make 3mil gddr5x versions and 17mil gddr5 versions... but you've killed demand for the latter by the former existing, you are going to kill sales as everyone waits for the former.
Last edited: