GPU prices will probably drop off a bit once 10nm lands - for various reasons 16/14nm will be costly for both nVidia and AMD (not just in terms of the basic core production cost).
-
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
GPU prices will probably drop off a bit once 10nm lands - for various reasons 16/14nm will be costly for both nVidia and AMD (not just in terms of the basic core production cost).
Soldato
- Joined
- 30 Sep 2005
- Posts
- 16,610
This guy does have a point
Lots of ocuk members class themselves not as gamers, but hardware enthusiasts. They buy Titan SLi and never play a single game.
Seriously, there are
There are some people like that - I've built a couple of rigs for people who aren't even hardware enthusiasts who rarely do anything more than dabble with stuff like league of legends but have to have the latest and "best" hardware regardless :S
Rroff don't bother mate. You are trying to explain to some that don't understand technology and when someone states double the transistors = double the performance without understanding how that will = said performance.
I can easily invalidate that statement with stating look at CPUS , double the transistors didn't = double the performance....
As Roff said it depends how the transistors are being used and Pascal is a double precision compute device as was Kepler and Fermi. You can't compare Maxwell to Pascal in that way. Maxwell was built as a single precision device and that's also what games use.
Also some need to go have a read about "Amdahl's law" ...
Anyways we will all find out soon what the score is with Pascal... for all we know it's a big fail for gaming.
I can easily invalidate that statement with stating look at CPUS , double the transistors didn't = double the performance....
As Roff said it depends how the transistors are being used and Pascal is a double precision compute device as was Kepler and Fermi. You can't compare Maxwell to Pascal in that way. Maxwell was built as a single precision device and that's also what games use.
Also some need to go have a read about "Amdahl's law" ...
Anyways we will all find out soon what the score is with Pascal... for all we know it's a big fail for gaming.
AS Roff pointed out, it's a lot more complex than Transistor count. More isn't always better.
Well, if some people on this forum are shills, then they are getting paid to promote a product, not sure you can call that that mindless. Annoying? hell yeah, but, hardly mindless.
And I am sure everybody has hobbies that they spend too much on. There are people who buy for the e-peen, but those guys exist in every area.
His comment doesn't equate to the intelligence he talked about having.
Yes that probably sounds like me, though I kind of wait for the "Ti" version of the cards. Full power for FFIX @4k res, I need to push the Afterburner OC profile for it
I would like to know when a shrink hasn't given at least 50% more performance, yes I know costs are higher and the amount of good dies can be a lot lower than a mature process but you will get a lot more dies per wafer.
I know if AMD target power saving on mobile up to a £250 we could have a problem where Nvidia come in with a more performance gpu which scarifies power for performance they could have a higher price hike.
I know if AMD target power saving on mobile up to a £250 we could have a problem where Nvidia come in with a more performance gpu which scarifies power for performance they could have a higher price hike.
When nvidia is reducing amount of cuda cores compared to maxwell, half the die size and add quite a bit of FP64 functionality in the same amount of transistors as titan x
That's when you don't get such a huge jump in performance.Remember, Maxwell was very lean chip. No FP64, half of the dx12 functionality missing.
Now there are some reports that 1080 will have less cuda cores than 980ti with similar amount of transistors as maxwell top end. I know the clocks can go up, but will it offset the loss of cuda cores(which in its own will be enhanced a bit if any).
Soldato
- Joined
- 30 Sep 2005
- Posts
- 16,610
No there are people who buy £800 graphics cards to run benchmarks, they don't promote, review or anything. It's a hobby.
The cards rumored for this year from both AMD and Nvidia seem like a bore-fest to me, mid tier parts (Not the full fat chips) using GDDR5 or GDDR5X.
The full fat cards with HBM 2.0 aren't rumored to come until next year. Wake me up when they arrive I guess...
The full fat cards with HBM 2.0 aren't rumored to come until next year. Wake me up when they arrive I guess...
I've got a factory overclocked 970. It replaced a factory overclocked 780Ti. Same CPU/mem/mobo/SSD. Performance difference between the two on my system is negligible. That's playing Arma III/DayZ/ED @ 3440x1440.
What do I expect from the next gen? Stock 1070 being the equivalent of stock 980Ti wouldn't be unreasonable IMO.
What do I expect from the next gen? Stock 1070 being the equivalent of stock 980Ti wouldn't be unreasonable IMO.
When nvidia is reducing amount of cuda cores compared to maxwell, half the die size and add quite a bit of FP64 functionality in the same amount of transistors as titan x
Remember, Maxwell was very lean chip. No FP64, half of the dx12 functionality missing.
Now there are some reports that 1080 will have less cuda cores than 980ti with similar amount of transistors as maxwell top end. I know the clocks can go up, but will it offset the loss of cuda cores(which in its own will be enhanced a bit if any).
I know but we get a good increase with a shrink never mind what gn4 and pascal brings over from the current gen without a shrink.
Soldato
So that means a GTX1060TI will be the GPU at £300 to £350 and a GTX1060 at £200 to £250??
Edit!!
Think I might hold onto my GTX960 until the next year at this rate. I don't want to be spending £200 to £250 on a part with a 128 bit memory controller.
Edit!!
Think I might hold onto my GTX960 until the next year at this rate. I don't want to be spending £200 to £250 on a part with a 128 bit memory controller.
Last edited:
Increase of what? The only thing that can increase considerably is clock speeds, the rest of it is either going down (cuda cores) or stays the same (transistors).
Soldato
Looking at the piece you quoted, I see that you decided to focus on the worst price in the article rather than the better ones.
Sounds to me like they converted the prices incorrectly, if you look and see that they also say its the same prices that the existing cards are at.
Still seems a bit high to me though and if they do come in at 980 and 980ti prices I for one will be disappointed. Of course if this ties in with the three card launch rumour, then this could be the GP104 400(1080ti) and GP104 200 (1080) with a GP104 150 (1070) to follow latter.
At the end of the day, who knows, it is yet another leak, which means something is definitely getting closer. Not long to go now, well only a few more weeks anyway.
you missed the important part although you highlighted it - the 1070 whilst replacing the 970 - is priced as the 980.
NVidia are price hiking again .
You are getting double the transistors, I thought fp64 added about 10/15% to die size so with a shrink you still should get a lot smaller die at the same performance yes they cost more to make on a new process and you get a lot less good dies and the percentage will go down at a larger size but that is why you see mostly smaller gpu's at first but i still think they will be faster at each price point or what is the point of upgrading.
That's what I said? That we all have hobbies we spend too much on.
The number of cores is irrelevant when the cores aren't equal. Why do you think Nvidia is cutting the number of Cores Because the cores are much more powerful and to make a more powerful card whit a better balance between all subsystem less CUDA cores are needed.More is not always better, look how many more cores Fiji has over Hawaii yet the performance difference is minimal because the architecture is bottle necked in other places.
On a new process with twice the density. Let alone the die area wont actually be half.
There is no guarantee that the consumer GP104 parts have FP64 added in. Even the, that only accounts for perhaps a 10% transistor budget. Amd also have to add FP64 back in to Fiji if they want to regain DP performance. Fiji was cut down much more from Hawiaii than Maxwell was cut down from Kepler.
Do we know the transistor budget for GP104? has Nvidia made an official announcement?in the same amount of transistors as titan x
We will see when its released.
Just like Fiji. No Guarantee the GP104 will get the FP64 functionality added in, or if it does to what extent What will be more useful for gaming parts is if they add FP16 support, which looks likely.
Maxwell V2 supports DX 12 feature level 12_1 with Conservative rasterization and Rasterizer Ordered Views. Fiji only supports feature level 12_0 in comparison.
Irrelevant.
You just wait and see.
There is absolutely zero reason for Nvidia to release a GPU with similar performance to the current 980Ti on a much more expensive process. It would be one of the biggest failures in GPU history like the AMD2900XT, Nvidia FX5800 etc.
Soldato
It makes me wonder whether they are doing it since they are sure they will thrash AMD,or whether there won't be many cards at launch.