Will Pascal close the mobile/desktop GPU gap?

What are your thoughts on Nvidia's mobile line of Pascal GPUs? There have been some rumors about a 990m coming out before then but I highly doubt it. Over the years, the gap of performance between mobile and desktop GPUs has gotten smaller and smaller, and currently there is only a 20% difference between the 980m and 980 GTX. So, do you think Pascal will finally close the gap?



Due to it's small size it seems more than capable of being squeezed into a laptop. Will we finally be able to laugh triumphantly at all the "Desktop>>>>laptop" junkies?

 

I think IMHO that the gap will never be totally closed due to physical issues... Desktop GPU's are considerably bigger in size which allows them to comprise more electronic components, hence they carry more "horsepower". We can never forget that as well as mobile tech is improving, so is desktop tech, which is the bigger slice of the gamer market. But well, that's only my opinion

 

Has the gap actually closed in recent years?

580M SLI = 580
680M SLI > 680 <--- only because of the delay of GK110
780M SLI = 780
880M SLI = 780 Ti
980M SLI = Titan X/980 Ti

Seems like it's always taken 2 flagship mobile GPUs in SLI to match the flagship big die desktop card.

 

980M is 22% slower than the 970. 980M SLI is only 24% faster than a single 980. There's still a large gap and that doesn't even take into account overclocking.

The gap will never be closed. Desktops will always outshine mobile GPUs because they have the capacity to dissipate the heat. As dies keep getting smaller, the heat is becoming more of an issue, not less of one. Maxwell is an anomaly... And when Maxwell is running at a steady voltage and clock rate, it loses that artifical thermal advantage.

 

I don't have the time to scale this further back, but if you go on back 10+ years, I know they must be getting much closer in performance... my old ATI 9600 Pro Mobility Radeon was a beast back in the day, but I seem to recall it being way more than 20%+ slower than the desktop king card of the day. This was my first laptop I splurged on a strong GPU (2004ish if I recall), but I can't help but feel the mobile front runners are much closer to the desktop front runners today then they were back then. Before I splurged on the ATI 9600, it was desktop only gaming that was worth a damn... Rendition, Voodoo etc.

We are definitely getting closer, but as others said, there will always be way more room to play in a desktop environment. Pascal won't be the equalizer.

 

The day you see desktops matching laptops means its either an electrical miracle or you know they are holding back on the desktop side (aka what they did with the desktop 980)

 

I actually remember mobile being closer to desktop back then (flagship vs. flagship). It was really G80 (8800 GTX) where Nvidia started investing heavily in die space building these monolithic GPUs on desktop every generation (G80, GT200, GF100, GF110, GK110, GM200, etc.) that mobile had no chance of matching.

 

That's years after when I'm talking about... Nvidia had some crappy "go" card at the time if I recall but it flat sucked.

 

I was talking pre-G80 same as you. If you look at the old GeForce Go or Mobility Radeons from 10 years ago they had fully enabled flagship desktop cores with all the functional units intact, just underclocked. That's not what we have nowadays. Nowadays mobile gets the GXxx4 chips at the high-end while desktop gets the GXxx0 ones.

Ofc it does help that 10 years ago the fastest desktop cards used less than 100W ( Nvidia/ ATi) while nowadays they're like 250W+ and mobile GPUs are 100W+. Again this goes back to my whole point about die size (and power consumption which goes hand-in-hand).

 

As processes become smaller, tablets and mobile gaming will become more popular. Eventually desktops will be a thing of the past. We are at the tipping point now where desktops are becoming more of an "enthusiast class" type of gaming machine, and we all know enthusiasts are not the future. The naive and moronic consumer is what makes companies the most money. In a way, Pascal, Skylake, and whatever else is coming is closing the gap, not because of performance, but because of profitability of mobile computing. One day soon, mobile graphics processors will be equal to the cost of producing desktop processors, but mobile gaming is more preferable and profitable, thus manufacturers will slowly push desktop gaming out.

We see this today with BGA - the first step into this mobile future. It's not necessarily Pascal closing the gap, but the advancement of technology itself erasing the line. Enjoy it while it lasts!

Spoiler: Graph

 

Its not really fair to throw the three together. Advancements in mobile chipsets has exploded compared to PC parts which causes people to constantly upgrade, not to mention that anyone can get a trade in for their used tablets easily without actually having to do the job of selling it. Other than Apple products and the legacy Alienware lines, PCs are just like American cars - they lose a ton of value as soon as you open them.

The biggest thing though is that there hasn't really been any real increase in x86 performance since Sandy so people haven't been upgrading.

 

People are replacing PC's with tablet's, so why shouldn't they be included? Consumers' needs/wants are being met.

By the way, I didn't make that graph. Take it up with Morgan Stanley or whoever made it.

 

A good tablet is awfully useful, no doubt... I grab my Surface Pro for the majority of what I do for both work and pleasure, and haven't bothered with a desktop for 10 years. The lack of future desktop development will be what closes the gap as J.Dre says, I agree. EDIT: If it wasn't for gaming, I wouldn't even bother having a notebook, much less a desktop!

Octiceps, I may absolutely be remembering things wrong, as I didn't bother with laptops until 2000 or so, as 80's and 90's was busy building desktops. But, back then, laptops weren't even an option. Then, around the early 2000's laptops became at least "viable", but I swear I don't remember them being anywhere near as good as full desktops. Yet today, I can game most of my PC games on a tablet! Progress I suppose. Only option was gaming on a desktop, then it became a 12lbs laptop, and now all those games are playable on a tablet or a 4lbs notebook.

EDIT:, see above

 

I use my Nexus 6 more than either of my PCs because of the convenience factor and the silent factor so I get that. People like thin and light devices, its no surprise tablets are taking over, especially given the fact that there are so many incredibly low cost options out there. You can get a Nexus 7 on eBay for what, 100, 120? And it can do most things people use a mainstream PC for which would have cost a lot more.

With Chromecast and Miracast, its going to get worse for the PC market. There's nothing more convenient then projecting whatever you are doing on your phone or tablet directly to your TV. I use Miracast all the time to play something on YouTube from my phone to the TV while I'm doing something else on the phone.

 

The day when we'll see laptops remotely matching desktop performance, will occur when we move to photonics or use graphene instead of silicon in our ICs. Until then, Si is but a semiconductor, which will have a resistance when a p.d. is applied across it, and will therefore dissipate heat. The larger and hence more powerful the chip, the more heat is dissipated. As @Zero989 mentioned, you need an electrical miracle for laptop-desktop parity. Not going to happen, guys.

Pascal will probably make laptops as powerful as today's desktop chips, but then desktop parts will get proportionally more powerful too. It's a never-ending cycle.

 

You hit the nail on the head... Untill we move away from silicon, this gap will be bridged but never eliminated

 

I partly agree with this line of reasoning, but not completely. It's true that having a greater thermal budget means that desktops will always outperform laptops, but there is another component to the discrepancy and once the advantage granted by this component is reversed, the difference will most likely be quite small. Every silicon chip (be it a GPU or a CPU) is aimed at a certain optimal power consumption range and a corresponding performance range. It is not possible to make a chip which scales perfectly from sub-watt fanless devices to 250W monstrosities; the manufacturer must always make a choice.

Up until very recently, GPU architectures have been aimed primarily at desktops and then cut down and down-clocked to fit into laptops which leads to inferior performance than one would expect from a lack of thermal headroom alone. However, it finally looks like this trend is about to reverse; Maxwell is not so much an anomaly as a harbinger of things to come. The TDP it is aimed at is still in the desktop range, but it is significantly lower than the one Kepler was aimed at. If this trend continues, then the tables will turn and we will eventually get to the same situation as CPUs. Compare the Core i7-4970K to the Core i7-4910MQ: the desktop version is certainly better, but by nowhere near what you would expect from having nearly double the TDP (88W vs. 47W).

 

Maxwell isn't all its cracked up to be in the TDP department. It uses clever voltage adjustment to try to keep the TDP down. If you hit it with a full heavy load of AA, it falls flat on its face in the efficiency department and both the power draw and the temperature go back to Kepler levels. If you flash a custom vbios that kills boost and the dynamic voltage behavior, Maxwell is not far from Kepler TDP and can easily surpass it.

On the Intel CPU front... You stick a 4790k in a desktop and you get turbo bins 24/7 unless you have garbage cooling. The 4910MQ, on the other hand, rarely hits its stock turbo bins out of the box. In some cases I've seen stock 4910s actually drop to base clock when hit with a wPrime load. Even my 4940MX fluctuates between 3.4 and 3.5GHz out of the box and it has a 10W higher TDP than the MQ does. Trust me, you don't want GPUs to go the direction CPUs did. At least my 980Ms will hold their 1126MHz boost clock most of the time, regardless of the load. My 4940MX needs 70W to do 4GHz 24/7 and the heat with a conventional thermal paste puts it right on the edge of hitting that magical 95C throttle zone.

 

This is not uncommon for products that are the first to go in a new direction. Yes, it has its drawbacks, but in most scenarios, the tricks it uses work. The next generation will refine them and by the third one, it will no longer be possible to notice a difference without deliberately creating pathologies that never occur in ordinary use. Think about SpeedStep/PowerNow -- nobody even comments on them anymore.

In the absolute worst case scenario for the 4910MQ, it will always run at the base frequency (2.9 GHz) while the 4790K always runs at maximum turbo (4.4 GHz) which means that the desktop has a 50% performance advantage. More realistically, it's probably going to be half to two-thirds of that. This is not trivial, but nor is it the overwhelming advantage desktop GPUs currently have over laptops.

 

I did not 980M SLI is equal to Titan X in terms of power. I am impressed.

 

But I prefer mobile system because yes mauby difference in perfomance two times but that is mobile and I can carry everywhere and walking around with 15kg is not wise I guess