Discussion: Polaris, AMD's 4th Gen GCN Architecture

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
@17 seconds, that bit is key though. AMD need to compete against the 1070 as opposed to the 1080. I don't see that as out of the question at all. Eventing we've heard about full fat Polaris 10 puts it in a similar performance bracket. Then the question becomes, which is more cost effective, selling a cut down large gpu, or a smaller full gpu? I guess yields determine which strategy makes more sense- high yield would favour AMD, poor yields might mean AMD don't get enough full die parts to realise a cost saving... Interesting times ahead me thinks...
 
The thing is it is not easy for amd to pull another HD4k series moment. Plus nvidia brand image is even more stronger than it was 5-6 years ago. Right now Nvidia actually giving pressure to amd slowly. With maxwell nvidia give the pressure with 970. Right now they repeat the similar strategy with 1070. Looking at the price there is large gap between 980 vs 970 and 1080 vs 1070. The repeat of this strategy shows that nvidia have no problem with this strategy. They use the x70 card to initiate indirect price war with amd but at the same time they also have no problem selling x80 at much significantly expensive price.
 


Of course Pascal is cheaper to make than Maxwell. That's one of the main advantages of new process nodes. The interesting question is how Pascal/GP104 compares with Polaris 10. Both of those should easily crush all the old 28nm hardware.
 


From what they're saying, Polaris 10 is a *much* smaller GPU that isn't going to outpace the top end parts of last gen... although most things point to it being close to the top parts. If you think back to the HD7XXX release, from what I'm hearing Polaris 10 = the HD 7870 in terms of positioning, rather than the HD 7970.

Whilst some people have taken that as 'the sky is falling' look like AMD are releasing the top end card in October under the Vega code name (with an even bigger Vega card to follow in 2017)...
 


Right. I'm talking cost vs. performance here, not raw performance per se. The GP104 in the GTX 1080 should really be compared to the GM204 in the GTX 980, not the GM200 in the GTX 980 Ti. And GP104 is in fact smaller than even GM204.

And yeah leaks say Polaris 10 is just a little bigger than Pitcairn, but it will obviously perform MUCH better than a 370/270/7870 etc. Just like a GTX 1080 destroys a GTX 980.
 
I'd say it's a good bet AMD will either release Vega, or expand on the Polaris lineup by/for xmas. Time flies, and it won't be long before the holidays. Waiting for Vega or higher end Polaris models, as well as Zen, before making a choice, would be wisest for those consumers looking to upgrade. Being patient til Polaris/Zen/Vega are available should result in better prices in general for the consumer. Just my .02$.
 


Oh yeah fully agree with you there, I think what is interesting is the fact that it looks likely that Polaris will be sitting in a 'gap' in NV's line up- it's probably going to be slap bang in between GP104 and GP106. That gives them a shot of being very competitive on price vs GP 104, assuming the performance is there to challenge the cut down 1070 part.
 
The issue with AMD's release sitting in a gap of nVidia's lineup and them not competing at / for the high end, is the "halo" effect. A lot of buyers don't necessarily know all the ins and outs of each parties lineup. However they are likely to know that the "king of the hill" is nVidia and thereby they will transfer that perceived awesomeness on to the lower end cards not knowing that there might be something from AMD that competes better at their price point. For instance when the 980Ti was tops, nvidia sold a lot of 960's and 950's because of the "halo" effect when AMD had very compelling alternatives at that price point.

Overall, I think nVidia plays the game better than AMD. They are very strategic when releasing new products. They release a card that isn't going to be their top performer, however at it's release it's the best. So it cuts AMD out of the top end and forces AMD to lower prices. Then they release one or two mainstream / mid range cards. Then before AMD has an opportunity to release something that can compete with nVidia's top end release, nVidia releases their actual top end card al a the Ti / Titan. The biggest issue is nVidia is releasing more actual new cards rather than rebranding their old ones like AMD has been doing. A re-spin of the silicon to get higher clockspeeds, maybe slap more VRAM on the card and bump the number of the hundreds place in the model number doesn't make it a new card. The analogy of chrome plating a turd comes to mind, except you don't have to chip away at the plating to much before you can smell what's underneath.

Don't get me wrong, I like AMD / Radeon, I really liked them (better) when they were ATI. That said, they just haven't been executing too well lately. They needed the Fury / Fury X out sooner so they held the performance crown longer. As it was the Fury / Fury X wasn't too much faster than the 980Ti and there were a few games that the 980Ti was still faster. I love AMD's open source initiatives, and they were the one that got the ball rolling on DX12 / Vulkan with their Mantle initiative. This is something that nVidia really needs to do better. I wish nVidia would abandon all their proprietary tech releases (GSync, PhysX, etc) and just release tech to help developers to make better games. They need to stop trying to tie it to just their GPU's. Especially now with VR trying to get a foot hold. Nothing will kill off VR faster than two different platforms competing for what is likely going to be, a limited market until prices come down. Laterly, nVidia's drivers suck too. Reminds me of AMD's drivers from a year ago and back from there.

 
Nvidia go proprietary ways because they don't want something they work on go unpaid. Ultimately they don't want to be the one that do the heavy lifting while their competitor get the free rides. If competitor want something similar then they have to work they own effort for it. At the very least in nvidia case they were always very clear with their intention. For example they were ask if they going to license gsync to others the answer was very clear: No (it's not like amd going to license gsync anyway). Unlike AMD with Mantle. They say mantle is open source but they refuse to give other IHV to look at mantle spec (intel was the most persistent one asking for it) while at the same time have no problem to give mantle access to game developer and being sole vendor capable of running mantle.

Nvidia actually slowly opening their tech. Still not in the same manner as open source but some of stuff like source code for gameworks effect and cpu based PhysX are more accessible to public without the need to pay licensing deal with nvidia just to look at the source code.

As for VR....so far gpu maker did nothing to fragment nor make any deal with HMD maker to support specific feature that only work on certain vendor. If anything the one that will kill VR itself will be non other than company like Oculus themselves.
 


AMD said they would release Mantle eventually (once it was complete), which they then did in a roundabout way by giving the code to Khronos to form the basis for Vulkan.
 
Some people say AMD intend to give mantle to khronos group since the very beginning and will kill mantle later on in favor of new API by khronos. They said mantle very peupose was only to kick start low level API on PC gaming nothing more nothing less. Except it was nothing like that at all. Richard Huddy himself already mention about the existence of Mantle 2.0. How mantle will continue to exist along side other low level API. how they will handle mantle development. And how they want mantle to be the forefront on what feature should be incorporate by other 3D API.
 
After they made the original Mantle publicly available as promised (just part of Vulkan instead of a Mantle SDK), they changed focus and future developments of Mantle will not be for general use. It'll be an internal prototype for developing new API features that can then go into Vulkan or even DirectX. Nvidia surely has some internal API development projects too, they have regularly provided input to DirectX and OpenGL.
 
No, you're wrong. Polaris will almost certainly NOT deliver 'Fury perf'. According to the leaked RX 480 info, that part will deliver GTX 970 performance. So FYI, thats TWO whole classes lower than Fury X / 980Ti. So its nice that it will cost 199 dollars, but that also means that its defintely NOT a competitor for the GTX1070 (which does deliver 980Ti performance).

A week ago, I as well still thought Polaris was going to hit Pascals cheaper card, but that doesnt seem to be the case, unfortunately. Therefore, I'm very interested in which cards AMD is going to use to fill the 480X and 490 placeholders. TDP wise, they could even go for Polaris 10 Crossfire as the 490X. Now if that costs 400 bucks it would be a nice hit against Pascal since it would sit between the GTX1070 and GTX1080.
 


The Videocardz leak showed the top single card (480? 480X?) at 18060 3Dmarks, where an R9 Fury was doing 18287 and a GTX 970 sat at 14929.

But it's all unreliable info at this point. We should wait for benchmarks.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom