Graphics Pascal and Polaris Discussion Thread

By mobile I assume you mean smartphone? Tegra is too hot and too much power consumption to go in mobile. Pixel c will likely see some sales.

 

As well as the Pixel C, Nvidia are doing prety good business with their own device lines (Shield Tablet, Shield Console).

Then I suspect these people are going to be very disappointed.

 

Really, I get the impression that they've not done very well at all. Have we got any stats?

 

I'm guessing we'll get no official word on Pascal until GDC in mid-March.

 

Perceived because of the world-leading Nvidia marketing machine. Reality is they are quite small fry.

Pixel C at $500 should see more sales than their own stuff, but Chrome-ware has been pitched at low-end so even 500 is 'difficult' when compared to entry NBs. This is why partners still stick to the cheapest stuff and let Google do higher tiers.

 

I went to double-check Nvidia's Quarterly earnings calls, and it looks like Tegra revenue is fairly stable other than a spike in Q3 FY2015. So doing well for itself earning about 10% of the discrete GPU arm, which isn't half bad (Nvidia don't break out their GPU revenue by sector, so no idea how much of that is OEM sales, HPC, etc).

 

https://benchlife.info/gp104-aka-nvidia-geforce-gtx-1080-will-ship-in-may-and-no-hbm2-031112016/

PINCH OF SALT WARNING:
GP104,
~May
8GB GDDR5
Single 8-pin power

 

if hats correct then NO HBM2 other than titan (and 1080Ti which would be the same thing anyway)

going by previous cores and product placement - GK104 was GTX 670/680/760/770 , GM 204 being GTX 970/980 (980Ti uses the titan core GM200)

 

Translated for the hard of Chinese

 

So these are going to be the ones that are as fast as a Titan X/980ti but are mid ranged and cost less, right?

Not really interested. I would need the full fat version.

 

If they are priced around £300 then a 980ti performance for that cash most would say yes please.

Really does depend on pricing if it's £500 for 980ti performance no one will buy it. If they are priced like the current 970 but effectively offer double performance then that's a good deal for many.

 

980 Ti replacement should be late q4.

From NV reps I've spoke to they pretty much confirmed no hbm2 outside titan. Q3 to q4 were expected volumes and I expect early qty was going to enterprise cards.

 

AMD need to hurry along with Polaris then really. They really need a head start on Nvidia this time.

 

Well gives me some time to save. See whats best bang for the buck.

Is it real 16nm its using or is it the samsung one that isnt quite 16nm ?

Same q with polaris . Be nice to see amd be the front runner for gpus again. If they can keep performance high this time and pull that power back and keep the price low. Recipe for success ?

 

AMD have had nothing to compete with Gx200 for a few generations though. Not sure at all what AMD will do.

HBM2 availability will likely be so tight this year neither will really 'win' this segment even if they make competing cards at Titan levels. The price will probably be even more as we expect the performance jump to be quite large, making demand persistently higher than supply.

 

AMD are desperate for cash the mythical thing of it been faster and cheaper than a Titan is just that a myth.

If they release a card that out performs the next Titan it will be priced at a similar price point. AMDs gpus have never been significantly cheaper than nvidia.

When they topped the CPU market performance wise they demanded top money.

 

Yeah they have. Back when the 9800 Pro and 5950 Ultra were king the AMD cost far less. The 4850 and 4870 were very keenly priced also, as was the entire 9000 range compared to the Nvidia cards.

Even the 7970 was relatively cheap when you consider that it was quite a chunk faster than a GTX 580 3GB yet AMD brought it to market at the same price of £439.

It's only really been since Titan that AMD are trying to charge Nvidia prices for Nvidia performance. And that's why they're so screwed.

After running two 5770s in Crossfire X and knowing something extremely iffy was going on (BTW the 5870 was also far cheaper than the 480...) and then having it proven a few years later I swore blind I would never touch AMD again nor their drivers. However, when they did sort the drivers again I decided to try a 7990, which I got for less than £400 after I sold the PSU that came with it. And it wasn't bad. Still a few hiccups but yeah, without the runt and dropped frames it was far better.

Then I had a piece of good fortune and extreme luck and had £3500 to spend on a PC. I returned to Nvidia, going for two Titan Black with a 4k Gsync monitor. Nothing but bloody issues, mostly because of the combination of Gsync and SLI. Then The Witcher 3 came and Nvidia derped the drivers making me even more angry and the poor performance coupled with the worst flickering I have ever seen (thanks to the SLI/Gsync problem) and I wasn't happy at all. However, I was still poised to go with two 980ti.

Then Nvidia "Did the 970" and their excuse and lack of empathy put me off of them for life.

So now I got two Fury X and the experience has actually been pretty damn awesome. A few games came along that did not play ball with Crossfire and AMD really stepped up and sorted most of them out (barring JC3 which is hateful toward multiple GPUs).

I really can't grumble at all (apart from the 4gb HBM not really being enough for console slop @ 4k).

Funny really. When I look back both of them have annoyed me equally.

I doubt I will play a part in the next gen at all. I think it's time for me to basically set my rig adrift and just leave it to go to pasture. I've done about £2500 in graphics cards in the last 18 months or so. God, how stupid :S

What will probably happen is I will reach a point where Fury X CFX is no longer sufficient for "up to" 60 FPS and will just get a cheap 1440p monitor. That will provide me with a nice FPS boost and hopefully keep me going for some time to come.

I'm sick of paying out the ahole for stuff that makes me think I am going to see it used properly, before realising that I can only really play what I already have, only slightly faster with slightly improved graphics.

 

Isn't that the way of gaming in general ?

The list of unique games released in the last few years are few and far between. Outside of Minecraft I'd struggle to list a title that is unique compared to what's been released already.

PC gaming has always been a costly business especially if you must have the best. 4K gaming is even above VR costs.

Next gen cards will be another level cost wise for early adaptors with the 980ti equivelent at least a year away. Whilst the next Titan will be out by end of June if rumours are true. Wonder if it will still offer 0-5% performance extra for £350 extra cost.

 

Titan now will be 16GB (4x4GB HBM2). With 8GB modules HBM2 out in Q4 (roadmap ETA) - this could mean Titan becomes 32GB (because there's enough willy waving people with money) and Ti becomes 16GB, with 1080s assuming the 8-12GB space. A 6 month refresh aligned to the launch of the mid-range is very good to boost sales through Black Friday/CyberMonday/Christmas/Chinese New Year season.

BUT if Nvidia decide HBM memory offers waaaaay more bandwidth than they 'need' Titan could get 4 stacks and Ti could have one/two less. That lowers manufacturing costs a big chunk - because interposer - and increases yields with two less parts to go wrong.

 

From NV and AMD now - HBM2 is really a late 2016 into 2017 product so no very high-end replacement for most of 2016? If Nvidia drop a new Titan, they can charge $3000 a pop and still sell out.



Unless Hynix worked out how to get bigger than 1GB stacks of HBM1, who-knows what Polaris will feature. So Fire Pro will be AMD's Titan for this year? (I know this is a Pascal thread, but it's implications for the whole ecosystem)

4GB cards are now entry-level. So maybe they have multi-GPU without mirrored memory working efficiently enough (it's been in the works already).

This idea was proposed before but the technology (interposers) and economics weren't forcing it. In future because the cost of manufacturing is increasing not decreasing with performance, it will be required.