![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
ZiggyDeath
Limp Gawd
- Joined
- Nov 3, 2017
- Messages
- 222
Been around a bit longer than that.
I made the following predictions/considerations when I bought a 1080Ti in July:
Pascal is a generational leap; Volta without Tensor is not a generational leap. Everything I've seen indicates that Tensor is not relevant/usable for gaming.
You generally need a generational leap to bring high end down to mid-range, like the 4000s, 6000s, 8000s, 200s, 600s, etc. I draw the line between mid range and high end when the chip is different; historically this is around the 60s/70s (or 600s/700s).
A 1060 6GB barely meets my 1080p requirements today, and because I am committed to moving to 1440p/144, 4k, or VR within 2 years.
My timeframe is 4-5 years, which is when I typically retire machines to secondary duty.
If I believe Ryzen 2/3 is able to extend the lifetime to 6 years or more, I am likely to replace the card in 4 year anyways. (aka stagnated processor development aka 2600k to 7700k)
I also don't have the habit of selling parts, so no salvage cost, and no consideration for interest/inflation.
A fully enabled "Tesla/Titan" Volta right now has ~5400 CUDA cores, a mid range Volta is likely to be in the ~1500 range.
Volta's CUDA cores are, by my estimates at best 7% faster clock for clock, core for core - Volta cores are Pascal cores.
This puts it between a 1060 6GB and 1070.
Let's assume it somehow matches a 1080:
If I buy a 1060 today, and then buy a Volta I'd be buying 2 mid range cards, "1080 price", for 1080 performance only when Volta is released.
What are the chances of a generational change in 2-3 years? In my estimation - very likely.
But if we make a generous assumption that it does happen, and it reaches 1080Ti levels, does it make sense to pay 2/3rd the price of a 1080Ti to get 1080Ti price in 2-3 years? Perhaps. But that's ~2 years sitting at 1060 performance.
Remember, this is being extremely generous; the gap between a 1070 and 1080 is smaller than between a 1080 and 1080Ti.
The 1080Ti is quite literally twice a 1070 for twice the price, which makes it, actually a pretty good value, albeit expensive - something that I can't recall happening at the top of the product stack (discounting SLI on a card).
Did it make sense to buy a 1080? Well, looking at the above predictions, no. The only way a 1080 will retains value is if we assume slow development and no generational changes.
Plus it was obvious from the getgo that a 1070Ti would be released. Although I expected 1-3 SMs to be re-enabled, not 4.
Had I known a 1070Ti would have 4 SMs re-enabled and it was actually available at the initially rumored MRSP, there is a chance I might have gone down that route. Of course the actual price is still completely jacked, and these cost about as much as a 1080, so my 1080Ti choice still stands.
Why didn't I choose a 1070? The incremental benefit was greater than the incremental cost by going to a 1080 (1080 was a better value). I also do not consider the 1070 to be a 1440p/144 card.
TLDR;
1070 too slow
1080 has better value than a 1070
1080Ti is even better value
1060+Future GPU (2 years) is a gamble that needs it to be ~1080Ti to be worth it
What do I think is probably going to happen in 2 years? A fully enabled mid-range card being about as fast as a 1080, maybe slightly faster.
Uh yeah... I need to write less...
For me, I would have been better off going with the FX8150 and then upgrading to an FX8350 a year or two later. (Should have gone with the Asrock 990FX Professional board instead of the Extreme 4 though.) 
