![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
This was back in 2011
I hope you are correct, I have been wanting to upgrade for a long time.
That was part of my point regarding yields, but that is not the only reason they target HPC business.
I'm not meaning to pick on you, but what is this based on?
Yeah feel the same, balance between yield, supporting all aspects of the HPC-professional market including cloud and research solutions, and demand in both that-more standard professional workstation-consumers.
Well how you can you be so critical with the lack of consumer Pascal news and yet base that on nothing when all the events to date are more professional/research/CUDA/etc based?
Also just remembered,
That gate is 6+ months away, still surprising...?
It's in volume production NOW, last week nobody would have predicted that anyone(other than Intel) would have a 600+mm^2 finfet chip currently in volume production.
2 x E5 2698 v3 = 2 x $3,000 = $6,000
16 x 32GB LRDIM DDR4 = 16 x $200 = $3,200
4 x Samsung PM863 1.92TB = 4 x $1,000 = $4,000
2 x Infiniband EDR Dual Port Cards = 2 x $1,000 = $2,000
1 x Mobo/dual 10gbe/NVLink Carrier = 1 x $1,000 = $1,000
1 x Case/PSU's = 1 x $2,500 = $2,500
TOTAL = $18,700
Retail System Price = $129,000
Minus base system cost ($18,700) = $110,300
Divide by 8 = $13,787.50
"Small Supercomputer in-a-box" volumes are not the same as "Consumer" volumes
The price will remain astronomical until (1) those Supercomputer needs are satisfied and (2) the yields go up to the point where they can enable the whole chip.
And while the split in REVENUE is 3 to 1 in favor of consumer business, you can bet the PROFITS are not. Those profits are the reason Nvidia can AFFORD to build a monster like this.
That said, while the announcement of GP100 before GP104 is unexpected, it's not unheard of. It just depends on how far behind they get with the big chip. I'm sure GP104 was made in-parallel.
not so sure, i dont expect gp104 to outperform a 980ti at 1500
Shoot for the moon, even if you miss you'll land among the stars.
AMD makes gains across the board at higher resolutions, even if you compare to Nvidia GPUs with similar memory bandwidth.
Do Kepler and GCN 1.0 support that? And if so would it make up that kind of performance margin?
I think we're getting off-topic, the original claim was that the Fury X excels at 4K due to HBM:It's simple, it's because you fail at math
At 1080p, the limiting factor is usually texture or shader throughput.
At 4k, the memory bandwidth is getting hit 4x as hard (because each pixel you draw consumes more bandwidth), making you more bandwidth-limited.
Kepler GTX 780 Ti has more shader and texture throughput than the 280X. But that have about the same memory bandwidth. 1080p = the 780 Ti shows it's shader and texture superiority (shader programs are small, and textures are easier to cache), since the ROP writes are 1/4 the rate they are at 4k.
If you are surprised that the relative performance of the 780 Ti drops off at 4k compared to 1080p, then there's nothing I can do for you.
And yet again this DOES NOT explain the performance margins shrinking at 4K. You can't say "it's bandwidth-limited" when both cards have 'the same' bandwidth.
Any idea where I can download that benchmark? Been looking for it for ages.It's simple, it's because you fail at math
At 1080p, the limiting factor is usually texture or shader throughput.
At 4k, the memory bandwidth is getting hit 4x as hard (because each pixel you draw consumes more bandwidth), making you more bandwidth-limited.
Kepler GTX 780 Ti has more shader and texture throughput than the 280X. But that have about the same memory bandwidth. 1080p = the 780 Ti shows it's shader and texture superiority (shader programs are small, and textures are easier to cache), since the ROP writes are 1/4 the rate they are at 4k.
If you are surprised that the relative performance of the 780 Ti drops off at 4k compared to 1080p, then there's nothing I can do for you.
EDIT: Kepler has 1st-generation pixel compression. It was introduced with Fermi. See here where there's a small imkprovement with the solid texture on the GTX 780 Ti:
AMD's Radeon R9 Fury X graphics card reviewed