Nvidia Turing GeForce 2080 (Ti) architecture review

1st and foremost, TSMC is open to any contract work.

however, as i've said and some people refuse to hear or acknowledge, 7nm production is restricted to those companies that invested in the infrastructure of the new fab. it's really very simple.
the so-called "increased costs" associated with the new node are no different than the costs associated with other new nodes at other times adjusted for inflation.

having something "new" for christmas is never a priority for either AMD or Nvidia as there have been many christmases with nothing new, like last year and the year before.
it is about herding cats (i.e. design validation, fab prep, fab shipping, AIB's getting gpus and shipping AIB's, etc).

when the job is done well (i.e. Pascal), the game developers are under NDA for months with ES cards, the product is widely available upon release (or shortly thereafter), and the final product is an improvement, usually a marked improvement.
that is not Turing.

as for AMD going all in on 7nm, from their position in the market why the hell not? they are not going to catch Nvidia at it's own game, so they are playing another game...called leapfrog. and it's scaring the hell out of Nvidia and Intel.

personally i have no axe to grind, i'm invested in Nvidia, Intel, TSMC, and AMD. so i get copies of the quarterlies that i actually read, i'm in the tech sector myself, and i live in Silicon Valley less than 10 minutes away from AMD, Nvidia, and Intel...and i've been following these companies (or their predecessors) since the 1980's.

 

You understand that what you said means that nvidia chose not to go for 7nm, as this is both a new arch, and a new node.

Your own first paragraph takes back what you said about nvidia not being a first partner for 7nm, as now you basically just said that they could have been.

The fact of the matter seems to be that nothing complicated (either in the CPU or GPU arena) seems to be able to be made in any quantity this year.

Please don't say the 7nm Vega 20, as the production for these will be just for sampling.

We will most likely see Ampere, or however nvidia decides to call the Turing shrink, tape out some time in Q1.

 

that is exactly what i mean.
Nvidia chose to not change nodes and architecture at the same time. the RTX architecture has been in the works for quite some time, and IMHO was originally slated for later release.

Nvidia was caught being complacent...as many people would be when you dominate a market. they were not alone.

Intel is still having problems with 14nm at this minute and their 10nm (7nm equivalent honestly) is vaporware.

everyone except for Apple, AMD,and Qualcomm have been dumbfounded at the progress TSMC has made and are being "left in the dust".

the entire iphone new line/ refresh is using TSMC 7nm
which by itself will outnumber the entire production of Nvidia over two years in a matter of months. AMD will out produce all Nvidia production (counting CPU's) as will Qualcomm (by a factor of at least ten...Android dontcha know...and CPU's).
these are facts.

but i totally agree with you re: Vega 7nm (not 20) consumer models. Navi will be extremely compelling mid-market and make anything below and up to the 2080 non-competitive (without ray tracing).

but $300-$400 for 4k free-sync gameplay is more than compelling

 

Just out of curiosity because so far nobody has mentioned this- is this tech more of a step in the direction of rendering light and shadows appropriately for viewing with a HMD, specifically for stereo-vr where you have to have everything perfect otherwise it will make you hurl. I have been using Nvidia's stereo-3D tech since they introduced their glasses somewhere in 2009 or so and some games definitely do better than others, most issues come from disparities in how the two offset views receive light and shadow info, which meant that most of the time if you turned shadows off and extra light processing like bloom you would get a perfectly rendered 3D scene. It is my hope that raytracing will greatly enhance the scale of what is possible for rendering 3D objects to actually be seen in stereo-3D instead of being "flattened" into 2D, or as I like to call it "pirate vision".

 

Apparently Squire Enix mistakenly posted the benchmark result for Final Fantasy XV including Turing GPU. It is deleted now. But somebody managed to save the data although I am not sure if this is genuine or not (http://blog.livedoor.jp/bluejay01-review/archives/54156401.html#more). The original language is Japanese. But you can understand the figures.

 

Arghhh, these new Turing cards may be EVEN more expensive than we have heard/expected! See here, Trump's 10% tarriffs:
https://www.notebookcheck.net/Nvidi...-increased-MSRP-thanks-to-Trump.332236.0.html

lol, another reason not to buy this architecture!

 

sounds like bs news to me

 

actually, no it's not.
tariffs are bad for business period. arbitrary tariffs (like these) even more so.
most chip fabs aren't in China, but a hell of a lot of AIB's make their cards in China and would be subject to these tariffs.
the worst hit will be Apple and Intel (motherboards)

 

no not that its fake news, its bs to pay more when its too much already
if it was a normal price I could live with it

 

Trumps trade tariffs take effect 24th Sept?

 

Amazing chart @Noisiv !

It's quite amazing how in just 20 years technology evolved from 10-15 million transistors to 12-21 billion transistors !
And yes, while it starts to level out due to approaching atomic scale, I can still wonder about ... how much more powerful computing chips will be in another 20 years.

 

Unless new physics are involved, not by as much as most people think. That's why tricks like DLSS or those Vega primitive shaders are so important.

 

I was looking for this exact chart yesterday, thank you so much!

 

(If you call something "bs", it implies that it's fake. Thought I'd point that out, English is not everyone's first language, so understandable.)

In other news, reviews for the new cards out tomorrow right, tomorrow is end of NDA right?! Looking forward to reading them! What time tomorrow does NDA lift (incl time zone)?

 

ok

 

Ah, if that's the case, then here in the UK the NDA will lift at 2pm, because Finland is two hours ahead of the UK when UK is on British Summer Time (BST). Will be some interesting afternoon reading!

 

To those that think the price is too high,


basically the deal is that gpu designs require exponentially more transistors
and the cost per transistor is not falling like it used to with every node shrink

The cost per transistor has remained the same or increased since 28nm,

An example of the effect this has,

The 980Ti has 8billion transistors , and now the 2080ti has 18.6Billion transistors, it should be no surprise that it costs ALOT more. (even with out yield considerations)

This is going to be the future of GPUs across the board im afraid(amd and nvidia alike), so long as transistor count increases and no major breakthroughs on the physics side of things occur.

Price will increase with performance from now on for the most part, at least it seems that way.

 

That was the Tesla architecture. The first dedicated compute card from NVidia, was also named "Tesla". The Tesla architecture was used for the GF8 series, GF9 series, 200 series and the (OEM only) 300 series cards.

 

Interesting, although we have to remember that the GPU silicon is only part of the cost of the card. You have VRAM/power supply circuitry/board/cooling solution and assembly + research costs as part of the final cost too (as well as other things I've not listed too). So a doubled increased production cost of the GPU core from 980ti to 2080ti won't result in overall production costs being twice as much - it depends on the percentage cost of the GPU core production in relation to the total production cost of the whole GPU card, and I don't know those numbers. (And I'm assuming the graph has been adjusted for inflation.)