Features - Roy Taylor: AMD Radeon GPUs remain unsurpassed

[KeyboardDemon]

hi I will be pleased to continue to contribute. We love engaging with the community. It's a pleasure.

That' excellent news.

One new tech that everyone will love is Freesync, it is coming soon!

Not soon enough, unfortunately, I think FreeSync compared to G-Sync sounds like a much better proposition as it offers essentially the same benefits but without the costly overheads of additional chips in the monitor or the restricted range of products that support it. I feel that G-Sync is a little too proprietary where an open solution like FreeSync is better for the community at large. However I have recently opted to buy a G-Sync monitor as I had the GPU to support it and having waited this long I didn't have the will to wait any longer for a FreeSync enabled monitor to launch.

[crossy]

hi I will be pleased to continue to contribute. We love engaging with the community. It's a pleasure.

Hmm, nicely put - although it probably helps if the "community" is a mature one (i.e. lack of rabid fanboys). And I still maintain Hexus is pretty darned good in this regard.

One new tech that everyone will love is Freesync, it is coming soon!

Like the idea that Freesync might give me a good excuse to upgrade monitors, but that's about it.

Personally, I'd be more interested in seeing better Linux support - especially since that'd maybe mean that SteamBoxes running AMD APU's would be viable (last time I checked SteamOS was NVidia-only unless you enjoy debugging).. Speaking of APU's - got an A8 based box for one of the kids and was impressed how good it is at games, so SteamBoxes sounds like an ideal place to use an APU, (small and inexpensive).

Also pretty interested in Mantle, but that presupposes that there's more A list titles that use it. Something nice to replace my old hexcore Phenom II would be nice - preferably with a TDP less than 125W.

[edzieba]

I think FreeSync compared to G-Sync sounds like a much better proposition as it offers essentially the same benefits but without the costly overheads of additional chips in the monitor

This isn't quite correct. the hardware required for G-Sync and DP Adaptive Sync are effectively identical. Both require new ASICs (the display controllers) to be fabbed for volume production at reasonable prices. G-Sync demonstrated using the design implemented as an FPGA, and distributed this (at the cost of an FPGA and associated hardware, which isn't cheap) while awaiting the design to be integrated into the next generation of display controllers. AMD demonstrated their Freesync implementation in a laptop, as the eDP panels are semi-controllerless, allowing driving them in odd ways in software.
Once the hardware required for variable refresh is built into mass produced display controllers, supporting both DP Adaptive Sync and G-Sync will be trivial. The main difference will be the license fee for using G-Sync, the connection protocol (DP vs DVI-D), and the timing negotiation protocol.

[KeyboardDemon]

Thanks for the reply, I'll be honest here, a lot of that went over my head.



When you talk about ASICs, are you referring to the G-Sync module that gets installed into the monitor?



If so, I was under the impression that all DisplayPort 1.2a compliant monitors would automatically support FreeSync as that would be a requirement to meet the design specification.



Are you saying that once the hardware required for variable refresh rate is part of mass produced monitors it would be possible to switch from a GPU running one standard to a GPU using the alternative without having to change the monitor?



If so, does that mean I could use a FreeSync from an AMD GPU on a G-Sync screen?

[HalloweenJack]

adaptive v-sync is already in the vesa spec , it requires DP1.2a , which no NVidia card supports (even the latest GTX 9xx series) , yet nv quickly added HDMI 2.0 , so they chose not to add DP 1.2a.

what does that mean? NV are playing the game - you want VESA adaptive sync ? not with us as we can sell you G-Sync!
another attempt to force a propriety hardware standard (see PhysX as another example)

[KeyboardDemon]

what does that mean? NV are playing the game - you want VESA adaptive sync ? not with us as we can sell you G-Sync!
another attempt to force a propriety hardware standard (see PhysX as another example)

It's because of the way nVidia like to adopt proprietary standards that I stopped using their GPUs for such a long time, I can understand why they do it, but I still don't like it.

[edzieba]

When you talk about ASICs, are you referring to the G-Sync module that gets installed into the monitor?

And 'ASIC' (Application Specific Integrated Circuit) is a chip designed to do only the task it was designed to do. A 555 timer, a display controller and a CPU are all ASICs. Because they only need to do the task they were designed for, they can be made to do it very efficiently. The current crop of G-Sync monitors instead run on an FPGA (Field Programmable Gate Array). these massive chips can be 'reprogrammed' to perform different tasks, and are often used to prototype designs before creating an ASIC. FPGAs are not cheap, costing a few hundred quid for a high-perfromance chip (needed to process video at high resolutions). This is why current G-sync monitors are so expensive: the FPGAs are expensive, and there's no way to make them cheaper. What can eb done is to integrate what the FPGAs do into the display controller ASICS, at which point the cost drops to pennies per chip.

The benefit is, people with deep pockets get the tech faster and that means lots of user testing before you have to tape out an ASIC. The downside is the FPGAs don't get any cheaper during this period, so you get a bad rap for being the 'more expensive tech' until they arrive.

If so, I was under the impression that all DisplayPort 1.2a compliant monitors would automatically support FreeSync as that would be a requirement to meet the design specification.

This is not the case. Adaptive sync is an OPTIONAL par tof the DP1.2a spec. There is zero requirement for a manufacturer to support it to be spec compliant.

Are you saying that once the hardware required for variable refresh rate is part of mass produced monitors it would be possible to switch from a GPU running one standard to a GPU using the alternative without having to change the monitor?

Barring any sort of Licensing Sillyness, there's no technical reason why you couldn;t make a monitor that accepted asynchronoys frame updates over both DVI (G-Sync) or DP (Adaptive Sync) with the same display controller driving the panel.

If so, does that mean I could use a FreeSync from an AMD GPU on a G-Sync screen?

Not unless that screen also supported DP Adaptive Sync.


Both technologies achieve the same (or very similar, Freesync seems to do some weirdness with pre-negotiated refresh rate ranges) result, and the actual tricky part of the display controller will be doing the same thing. the difference is in the interface between the GPU and the display controller.

[KeyboardDemon]

Thanks, for a minute there I was starting to think I had made a bad decision when I picked a G-sync monitor, but reading about the FreeSync weirdness has made me feel more comfortable with my decision. The only downside I see is that it looks like I'll be tied down to nVidia GPUs for the foreseeable future so hopefully they will continue to release product at the more reasonable price points that the current 9x0 series of cards are settling at.

[spl]

Thanks, for a minute there I was starting to think I had made a bad decision when I picked a G-sync monitor, but reading about the FreeSync weirdness has made me feel more comfortable with my decision. The only downside I see is that it looks like I'll be tied down to nVidia GPUs for the foreseeable future so hopefully they will continue to release product at the more reasonable price points that the current 9x0 series of cards are settling at.

Most people consider the 970 to be pretty exceptional value given the 780-beating and 290X-beating performance! Even the 980 is very reasonable considering the price the 780 (and -Ti) launched at. I guess the 960 or 960 Ti will be the highlight of the series for price : performance though.

[DanceswithUnix]

This is why current G-sync monitors are so expensive: the FPGAs are expensive, and there's no way to make them cheaper. What can eb done is to integrate what the FPGAs do into the display controller ASICS, at which point the cost drops to pennies per chip.

I expect it is more down to these monitors being a low sales volume product. FPGAs can be fairly cheap, you can get a breakout hobby board with the FPGA, config flash, voltage regulator and crystal all tested and shipped to you for about £15, so the chip itself can't be that much.

Can you make FPGAs cheaper? Yes, like any component you buy them in bulk.

But it took years for Bluetooth to be integrated into devices to the point that it doesn't really add to the cost, eventually I'm sure the same will happen here but I am not holding my breath.

[watercooled]

And 'ASIC' (Application Specific Integrated Circuit) is a chip designed to do only the task it was designed to do. A 555 timer, a display controller and a CPU are all ASICs.

Being pedantic for the sake of being pedantic here, but by its very definition a CPU is not an ASIC. ASICs can contain CPU cores, along with other logic to create application-specific chips (hence the name), but a CPU alone is pretty much the opposite of one.

And it isn't always cheaper to go from FPGA to ASIC - you often need massive sales volumes for it to make sense to begin the ASIC design process, and considering they've chosen a fairly big FPGA I doubt it's just going to be a case of adding an insignificant amount of logic to the display controller. As an example, a lot of high-end oscilloscopes use eye-wateringly expensive FPGAs for signal processing, the sales volume just isn't there for an expensive ASIC design to offset the cost.

Plus as KeyboardDemon says there are plenty of political rather than technical barriers for G-Sync, it currently seems to be limited to a couple TN panels, it's as proprietary as can get (Nvidia after all, who have effectively strangled and limited the use of PhysX by hoarding it), and there's an open alternative. And AMD have demonstrated FreeSync on desktop monitors, in fact the only laptop demo I remember was when it was first announced, and non-final hardware for demos is nothing unusual.

Thunderbolt vs USB3 is another wonderful example of restrictive licensing and approval vs FRAND. Amongst other things, Thunderbolt could have allowed external GPUs, was proved by modders to be workable, and a few companies announced such products which were never granted approval. But approving such devices would have meant you didn't have to buy the monstrously expensive Iris graphics on your new laptop. The problem with one company with a vested interest in hoarding the technology holding the licensing rights is it tends to lead to said technology being strangled.

[KeyboardDemon]

The problem with one company with a vested interest in hoarding the technology holding the licensing rights is it tends to lead to said technology being strangled.

Thanks, great explanation and very helpful, I particular like your closing remark.

As much as I like nVidia products I really hate the way they use proprietary standards and what's worse is that it now appears that AMD have started to use the same tactics with Mantle for example. I know Mantle is supposed to be an 'open' standard but I think it already looks like there is a split forming between Mantle, DX11/12 and Gameworks developers and I am starting to wonder how long it will be before I find myself in a position where I can't play certain games because my PC doesn't have the right GPU, once again.

[watercooled]

I agree; while there's nothing stopping Intel or Nvidia from utilising Mantle, we know that's not terribly likely regardless of how well-intentioned it might be, or how much it might provide mutual benefit. However it's arguable Mantle has already largely achieved what it set out to do in picking the low-hanging fruit of reducing overheads as the newer versions of DX supposedly achieve much the same thing.

It will be interesting to see how the two compare, and I don't think DX being chosen over Mantle is even a remotely bad thing for AMD going forward if it's all its made out to be. However new DX means new Windows, and Mantle is here now with a huge compatible market.

I'm not yet sure how bad Gameworks is likely to be for the gaming market, I've heard rumours that a lot of the packages aren't locked to Nvidia hardware as they use DirectCompute, but nothing solid yet. As for GW in general though, real devs seem to agree with AMD that black boxes are a bad thing for performance and optimisations. As an example of what driver optimisations can do, look at 7970 performance before and after one of the early updates; it went from being generally behind a 680 in BF3 to solidly ahead.

[edzieba]

I expect it is more down to these monitors being a low sales volume product. FPGAs can be fairly cheap, you can get a breakout hobby board with the FPGA, config flash, voltage regulator and crystal all tested and shipped to you for about £15, so the chip itself can't be that much.

There exist a whole range of FPGAs, yes, from cheap to expensive. However, there are G-sync monitors handling 2560x1440 at up to 144 Hz. At 8 bits per subpixel, that's up to 12,740,198,400 bits/second (12.7 gigabits) ignoring overhead, blanking intervals, etc. And that's just input; the FPGA also needs to buffer that (and the reason the G-Sync board has such a massive amount of RAM is to get decent bandwidth by using multiple ships in parallel, due to the limited speed of each of the FPGA's memory lanes) and output pixel driving data to the panel (after calculating overdrive, comparing to the lookup table, etc), all simultaneously. That's a massive amount of data to handle, so you need a high-end FPGA to do anything useful with it.

I'd also argue that a SPU is an ASIC. Its specific application (in the case of a desktop PC) is handling x86 instructions. You can't feed in, say, a raw DVI datastream and expect it to do anything with it. This is unlike an FPGA, where feeding it Verilog or VHDL does not result in any sort of inherent output, but you can arrange the internal logic in order to handle a wide variety of tasks.

[watercooled]

I'd also argue that a SPU is an ASIC. Its specific application (in the case of a desktop PC) is handling x86 instructions.

Isn't that a bit like saying the specific purpose of a transistor is to switch things on and off?

This is unlike an FPGA, where feeding it Verilog or VHDL does not result in any sort of inherent output, but you can arrange the internal logic in order to handle a wide variety of tasks.

The 'application-specific' part distinguishes it from simply 'integrated circuit', otherwise anything could fall under that description. After all, all ICs are designed for a purpose which could be called specific on some scale; e.g. an FPGA can be used for any number of logic applications but you're not going to convince it to clean the toilet for you. There's no dividing line between what's a general purpose IC and an application-specific one, but a programmable CPU is pretty much the opposite end of the scale to what ASIC generally refers to in my experience.

CPUs can be programmed to carry out any number of tasks whereas ASICs are typically designed for a very specific process e.g. an SSD controller, modem IC, network transceiver, and something with a lot of attention recently, cryptocurrency miners. A CPU can be programmed to perform many of these tasks with appropriate ADCs, but it generally wouldn't be very efficient or cost-effective in such a role.

GPUs started as something closer to ASIC but have arguably blurred the lines more recently with their more general-purpose use and hugely increased programmability.

[george1979]

I'd forgotten about this thread and just came across it again. Wasn't sure whether to reply but...

Actually, instead of going for a cheap shot you might want to actually READ the Q&A - what RT was doing was giving real world examples of where AMD tech is being used. And while Hexus readers are going to be interested in PC graphics cards, your "man on the Clapham omnibus" is going to be more impressed with PS4 and XBone. Oh, and is "TressFX" any more ridiculous than NVidia's bigging up of their "grass simulation"? Judging on the hype around that, some PR flacks are smoking grass rather than seeing it simulated.

Fair enough, I suppose it was a bit of a cheap shot but don't make the mistake of assuming my comment was fanboy-ism on my part. I would have said something very similar had it been an Nvidia PR guy going on about a grass simulation
Also not my intention to demean the work of the people who work on the tech, I mean I'm glad their working on increased realism I just don't see using grass or hair as a selling point swaying many people.

So you laughed at PhysX then?? Funny how the first physics implementation from AMD actually worked on my GTX660 fine,and yet only after AMD did that did Nvidia start to make some of its own physics effects more cross vendor!

So are you laughing at all the "turf" simulation they have been bigging up:

http://techreport.com/news/27272/the...a-turf-effects

Again - if it had been an interview with an Nvidia exec going on about turf then, yes, I would have taken the mick out of him too

Wasn't trying to be mean so I hope it didn't come across as such - just taking the pee out of an article that I genuinely feel deserved to have the pee taken out of it!

I actually have a great soft spot for AMD, My first 'gaming' PC was all AMD/ATI and I sincerely hope they pull something out of the bag cos the last thing anyone needs is Intel and Nvidia not having any viable competition. As many have said, it is cyclical so I'd expect them to come good soon.

Tell me that a 980 which performs worse than my R9 290X with Mantle on and matches it in DX11 is worth a Additional £140 premium after they finally made a 4GB card? No didn't think so.We're done here your dismissed

I have literally no idea what you're on about here but it was an amusing collection of words lol. Keep up the good work