Nvidia Turing Product Reviews and Previews: (Super, TI, 2080, 2070, 2060, 1660, etc)

I can't help but give nvidia a (tiny little) bit of comprehension here.
Unlike the Pascal line, these are huge chips that cost a lot of money and probably don't have the greatest yields. Even the smallest TU106 from the 2070 is almost as big as the GP102 from the 1080 Ti / Titan Xp.
Of course nvidia is still making increasingly larger profits out of each chip, but I don't think it's as much as people are making it to be simply by looking at the SKU names.


As I mentioned before, I do think the GDDR5X products will cease production in the short term because GDDR6 will probably be a lot cheaper (the 3 major RAM makers will produce it for many products).
This means the 2070 and 2080 will replace the 1080 and 1080 Ti, respectively. They'll probably keep the 1070 Ti and below in production until the 7nm chips arrive, but the G5X products will stop production well before that.


I get that, but all in all, they can "allow" themselves to do that because of non highend competition imo.
 
So now that Anandtech has posted their review, I though I would update the chart. Because game number 4, (Wolf2) has a much higher delta than everything else, I had Excel spit out a random number which ended up representing Total War and grabbed those results as well.

upload_2018-9-19_17-50-55.png

So.... unless you are huge Wolfenstein II fan, as far as games available NOW are concerned, 2080 TI is by far the worst of these transitions. Using the (Performance Δ - Power Δ - Price Δ)*100 for a composite score, 980 scores 50, 1080 39 and 2080 is ...0
 
I like that Anand was the only site that emulated ref clocks on the FE cards. This kind of standards are becoming rare in the media, and I honestly expected none other than Anand to adhere to them.
we've gone ahead and emulated the true reference specifications with a 90MHz downclock and lowering the TDP by roughly 10W. This is to keep comparisons standardized and apples-to-apples, as we always look at reference-to-reference results.
In the end Anand has the 2080Ti 32% faster than 1080Ti ref to ref. And has it 37% faster with the default 90MHz OC.

https://www.anandtech.com/show/1334...tx-2080-ti-and-2080-founders-edition-review/5
 
So now that Anandtech has posted their review, I though I would update the chart. Because game number 4, (Wolf2) has a much higher delta than everything else, I had Excel spit out a random number which ended up representing Total War and grabbed those results as well.

View attachment 2659

So.... unless you are huge Wolfenstein II fan, as far as games available NOW are concerned, 2080 TI is by far the worst of these transitions. Using the (Performance Δ - Power Δ - Price Δ)*100 for a composite score, 980 scores 50, 1080 39 and 2080 is ...0
Why wouldn't you just include all the games?
 
So now that Anandtech has posted their review, I though I would update the chart. Because game number 4, (Wolf2) has a much higher delta than everything else, I had Excel spit out a random number which ended up representing Total War and grabbed those results as well.

View attachment 2659

So.... unless you are huge Wolfenstein II fan, as far as games available NOW are concerned, 2080 TI is by far the worst of these transitions. Using the (Performance Δ - Power Δ - Price Δ)*100 for a composite score, 980 scores 50, 1080 39 and 2080 is ...0

That's interesting. This 20xx series reminds me of the jump from the 7xx to 9xx series with the 970's and 980's performance straddling the 780ti. Pascal's jump in performance looks to have been something special. Great node shrink 28nm to 16FF as well as some good architecture optimization.

I think the only thing about the RTX launch that is bad is the prices. Given the same node and what Nvidia chose to do with the chips (RT enhancements) the performance gains seem ok, it's just that the price is so out of whack. I bet once Pascal is gone, they'll drop the price by a $100 bucks or so. I don't think we're going to be going back to $500 x80 series given how large these dies are and I doubt that 7nm in the future will allow for that as well even if the chip sizes shrink for the follow up series.
 
I don't think we're going to be going back to $500 x80 series given how large these dies are and I doubt that 7nm in the future will allow for that as well even if the chip sizes shrink for the follow up series.

A ~500mm2, 7 nm big Turing2 GPU would not be unreasonable.
On the then brand new 16 nm process, GP102 was 471mm2 (coming from 28nm for Maxwell)
(7nm SoCs like the Kirin 980 have 7B transistors on ~100mm2)
That could be ~35B transistors for TU2, twice that of the TU102.
A lot of guessing here, but 7 nm should bring a huge jump in performance, we will see soon enough with the new NV HPC GPU.
 
A ~500mm2, 7 nm big Turing2 GPU would not be unreasonable.
On the then brand new 16 nm process, GP102 was 471mm2 (coming from 28nm for Maxwell)
(7nm SoCs like the Kirin 980 have 7B transistors on ~100mm2)
That could be ~35B transistors for TU2, twice that of the TU102.
A lot of guessing here, but 7 nm should bring a huge jump in performance, we will see soon enough with the new NV HPC GPU.

I agree perf wise, but my point was more about price. I don’t really see prices going back down with 7nm for such large chips.
 
Back
Top