Soldato
- Joined
- 11 Mar 2013
- Posts
- 5,502
This 100%^
-
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
This 100%^
If they are mass producing them by the start of the year, I'm sure it wont take long to stick them on the cards.
"The two chip makers are scheduled to supply HBM chips to servers, supercomputers, networks, and high-performance PCs, in addition to the Pascal GPU, starting 2016"
Why would they need to be tested and refined? Nvidia will already have preview/test samples - "after completing pilot production and reliability tests by the end of 2015". Presumably the memory will be mass produced and then just stuck on the GPU's ready for sale.
Last edited:
With GPUs post production refinement tends to turn up as a refresh 6+ months later.
And QC tends to be - oh that sample failed? keep pulling samples until one passes then certify it :S
And QC tends to be - oh that sample failed? keep pulling samples until one passes then certify it :S
I love the enthusiasm in a word: 'Stick/Stuck'It is so simple to put separate next gen parts together these days, isn't it? Just glue them together and the whole package just works. magic
When it comes to engineers, it is or should be that simple. They are skiloed in their field and know what they are doing and would have already experimented with it, even if it was HBM 1.
I'd guess a lot will just depend on what kind of quantities they can deliver to Nvidia and how quickly.
Well, it is unwritten rule in the industry to not do next gen arch, next gen process and next gen memory controller/system at once. Intel does not do that anymore, AMD doe not do that anymore, and nvidia shouldn't be doing this, yet they are.
There are numerous times in history when companies trying to do even two next gen things together fail miserably. Nvidia done it in the past, AMD done it in the past and Intel done it in the past.
With process tech going more complicated and smaller it is more chance something to go wrong with dies itself, not even mentioning new memory tech and architecture.
So no, it is not that simple even with the brightest and richest engineering teams. nVidia would need to pull a major miracle to release all those next gen pieces together and working. And, no, fiddling with couple of samples of HBM in the lab is not experience by any stretch, especially when none of us here know how those couple of samples are behaving, since no one would ever admit that they are failing or working perfectly.
They need to seriously think of a better name for the 1050ti.
So is the pipe cleaner what the 980ti is to a titan x, a slighty less powerful card but a lot cheaper?
No, pipe cleaner is relatively small chip with some new tech on it. Like fiji for AMD
So since AMD has quite a bit of experience with HBM, and they have functioning product on the market with HBM, next year they just need to worry about other things than memory subsystem. While nvidia would need to worry about 3 things going wrong at the same time.
How early was 750ti released in relation to full blown Maxwell? I'm just wondering if 1050ti pipe cleaner for q1 next year would be too late to troubleshoot things for major release later that year.
And I really really doubt that Titan X is so much more expensive than 980ti to produce
Just nvidia can charge anything for Titan X, since there are plenty of people willing to spend such money for the GPU.Pascal isn't a next gen architecture, Maxwell is the next gen architecture that will take Nvidia through to Volta and perhaps beyond. Pascal is Maxwell scaled up and a new memory controller for HBM2.
AMD will have a brand new architecture "beyond GCN" and a new process. So Both AMd and Nvidia will be having 2 new technologies. Neither of them can do much about that, 16nm was delayed.
I hope its within the first 4 months of 2016.
I'm currently sitting on the fence of either grabbing a 980Ti now with a whole new build or waiting for Pascal. I am so indecisive when it comes to upgrades.
Pascal does have quite a lot of changes over Maxwell though - even ignoring stuff like nvlink and HBM - quite a lot of enhancement of the compute capabilities (including mixed precision and context switching, etc.), big changes to how different sub-systems communicate with each other - scheduling, caching and so on.
I will "eat my hat" if they release the "big" pascal (980ti / titan equivalent) in the first 4 months of next year. I expect it will be near the end of 2016, especially the 980ti version, the titan version will probably come first then the 980ti version when they have enough titan rejects. Probably Q3 - Q4 2016 for a pascal 980ti
Yeah, for their top card
It'd be a miracle if the rest weren't be re-brands or just shrinks
Looking forward to the next coming of the 7970 I can't see the yields being good enough to harvest big dies profitably for the consumer market before mid-year. Tesla on the other hand might start early.
If someone wants to do the maths the cost of a wafer is supposedly 8K, 300mm wafer, assume 550mm2 die, yields will prob be around low tens % atm.
If someone wants to do the maths the cost of a wafer is supposedly 8K, 300mm wafer, assume 550mm2 die, yields will prob be around low tens % atm.
Last edited:
That's why I said hope
If its a near end of 2016 for the ti equivalent I might just have to go with a 980ti, I won't upgrade for another 3-4 years after. Running a 670 still.