Margins for automotive are way way better. Nvidia made 400-500 million on Tegra automotive sales because out of the 560 that tegra generates as far as revenue annually, most of it is from automotive. They were included in 2.8 million infotainment centers last year. This means each system generated $142 dollars in revenue which considering the cost to make a tegra chip is a massive margin on such tiny chips.
The problem is Nvidia is spending alot of R and D on the Tegra program which makes it a loss. But gross margins for tegra in cars are fantastic.
If the volume doubles, which could happen in the future, the tegra business could easily become a cash cow like the professional maket.
I'm confused as to why so many people think that this is impossible.
Does no one remember most other die shrinks and hell, even the 780 TI -> 970 which wasn't even a die shrink?
That was a $300 card replacing their own $700 one, without a die shrink.
Yet new tech, with a huge die shrink (28->20missed->16 skipped -> 14!) and people don't think it can compete?
Both AMD and Nvidia have to compete against their old top end with their new mid range if they expect to sell them.
The Polaris architecture precisely combines the latest 14nm FinFET process and AMD's advanced power, gating and clocking technologies for a superior cool and quiet gaming experience.
Powerful async compute and new geometry capabilities enable unique support for DirectX® 12 and Vulkan™ in the best version of Graphics Core Next yet.
Polaris-powered Radeon™ GPUs are engineered to provide premium VR experiences to a wide range of users. The Polaris architecture provides the perfect combination of low latency, smart software and powerful compute that enables a silky smooth virtual reality experience.
but how do you explain this?I'm confused as to why so many people think that this is impossible.
Does no one remember most other die shrinks and hell, even the 780 TI -> 970 which wasn't even a die shrink?
That was a $300 card replacing their own $700 one, without a die shrink.
Yet new tech, with a huge die shrink (28->20missed->16 skipped -> 14!) and people don't think it can compete?
Both AMD and Nvidia have to compete against their old top end with their new mid range if they expect to sell them.
Right.
It has happened at most node transitions, where next-gen mid-range chips match or defeat previous top-end chips. At a fraction of the price. We can go back to the 4870, gtx460 to the recent 7870 and gtx660 and it is 100% historic fact.
If 14/16nm cannot achieve this, it is abnormal and a major letdown.
I saw the wafer cost chart, and currently 14/16nm is ~4K to 4.5k USD per wafer (only ~25% more expensive then 28nm!), not as expensive as I had thought (it dropped fast from 2x the price of 28nm wafers).
These small chips should actually be cheaper to produce than 600mm2 on 28nm, unless 14/16nm yields are horrid. Which at this point, the heavy lifting has already been done by early adopters for over a year already.
Won't be long now, exciting times ahead for PC enthusiasts!
Historic fact or not, fact is also that processes are getting much more expensive and hence a new process will not bring the same price cuts that were possible before. So while it may be mid-range in die size it won't be in price.
Second part you omit is power consumption. If the next gen midrange part focuses on power-savings instead of higher performance, your argument also fails.
The tell-tale sign that P10 will not offer 980 Ti levels of performance is it's memory bandwidth. AMD would need to catch up to NV in term of memory efficiency (color compression) to get 980 level performance with a 256-bit bus not to mention even better performance. Possible but doubtful.
Polaris 10 is a mainstream small chip. On historic grounds, it should NOT match Fury X. Now I am not ruling it out either way, just to get the air clear that expecting a mainstream chip to match a prior gen high-end enthusiast class is too much. This is the same as expecting GP106 = GM200... unrealistic!
AMD's mid-range will be Vega 11, I expect it it to be ~340mm2 or around there. This chip will easily defeat Fury X/Titan X.
Sure, if you start with the premise that AMD's technology can't possibly be as good as Nvidia's, then you'll reach that conclusion. But there is no reason to believe that is true. It's specifically been noted that Polaris will include better memory compression than GCN 1.2.
SilverForce11, are you going to "retire" your 290x for a new Polaris?
To be clear here.@beginner99
Those examples of previous generations, were as fast or faster than prior top-end big GPUs at a fraction of the cost and also TDP.
That's the benefit of a node change. Add uarch improvements on top and it should defeat current high-end easily.
IF IT DOES NOT, that is a failure.
Now, many people are calling Polaris 10 as mid-range and lumping it together with GP104. This is flat wrong, stop doing it. GP104 is a true mid-range with both the normal expected die sizes and rumored TDP.
Polaris 10 is a mainstream small chip. On historic grounds, it should NOT match Fury X. Now I am not ruling it out either way, just to get the air clear that expecting a mainstream chip to match a prior gen high-end enthusiast class is too much. This is the same as expecting GP106 = GM200... unrealistic!
AMD's mid-range will be Vega 11, I expect it it to be ~340mm2 or around there. This chip will easily defeat Fury X/Titan X.
You're literally doing the exact thing by saying Polaris 10 will be slower than gpus it aims to replace utilizing 0 facts or data to arrive at that conclusion.....Yup, I think at most the fastest Polaris 10 Sku will match a 390 in performance. But I don't think we're looking at 980 Ti performance, or anywhere near it. But we'll all have to wait and see. I wish everyone would just let relax and stop hyping polaris/pascal to death and wait for the reviews.
Depends on the performance and price.
If it's $299 and 390X + it's more compelling, but still not enough for me to upgrade. My R290X OC quite nice, so in fact, I already have 390X + performance now. When I bought it new, it was equivalent to ~$250 USD.
In fact, Polaris 10 isn't going to do much for bang for buck for anyone who already owns 290 and 290X for a few years already. It'll be good for new builders though, as Polaris 10 should give more bang for buck than 390/X at the least.
Vega 10 is where it's at for me (4K).
Anyways, it doesn't sound like Polaris 10 is going to be a big shaker in the performance sector. If so, that sucks, since I already got a water cooled 290X that cost be ~$165+$20 for the water kit mount. Still can't wait to see some benches!
If Vega 10 is >$350 (assuming Polaris 10 full comes in at $350) how is that "mid-range?"
Polaris 10 falling into the $250-$350 price points would be by my definition mid-range.
its not silverforce that is saying its amd own slides...I think Silverforce11 is wrong saying constantly that Polaris 10 is just a mainstream chip - IMHO this is a mid-range when Vega 11 would be for high-end and Vega 10 for enthusiasts.
its not silverforce that is saying its amd own slides...
they have polaris 10 on par with fury lines and vega started above fury lines
This already happened last year, the i7 5775c is probably better than any entry level GPU i.e. sub 100$ & if we were to have a (hypothetical) AMD IGP (with GCN4) having HBM2, or eDRAM, they'd destroy the ~150$ GPU range easily.I'm starting to wonder if the terms are either losing meaning or because the iGPU destroyed the bottom, some tiers are meshing.
Vega 10, assuming it only has HBM2 or GDDR5x, will not cost anything less than 400$ at launch. The same goes for Polaris 10 & if the top bin (higher clocks) is paired with GDDR5x then it'd command a premium, over the regular GDDR5 version, & could likely debut at or close to 400$If Vega 10 is >$350 (assuming Polaris 10 full comes in at $350) how is that "mid-range?"
Polaris 10 falling into the $250-$350 price points would be by my definition mid-range.
At this point in time we'd have to define performance level as a criterion, for various tiers of GPU. Now you can easily get a high end i7 that'll match sub 100$ dGPUs & it'd likely (maybe) extend up to 150$ i.e. if AMD ever decides to release such a part. So, basically what I'm saying is that with the bottom tier bottoming out, the rest of the market has moved up a notch in terms of price & hence these price tiers for dGPUs should be adjusted accordingly.Mainstream would have to be Polaris 11 slotting in under <$200.
Anyways, it doesn't sound like Polaris 10 is going to be a big shaker in the performance sector. If so, that sucks, since I already got a water cooled 290X that cost be ~$165+$20 for the water kit mount. Still can't wait to see some benches!
Audi is using it since 2012 down till A3, VW in Golf, Passat, Skoda Octavia and so on since 2013.
For a firestrike Ultra score of 3760, you need a 1521/2065 980. That's not "slight" IMO.