Soldato
- Joined
- 20 Apr 2004
- Posts
- 4,410
- Location
- Rugby
http://www.anandtech.com/show/10536/nvidia-maxwell-tile-rasterization-analysis
Sure does explain a lot
Sure does explain a lot
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Maxwell uses tile based rendering ?
- Thread starter KillBoY_UK
- Start date
Soldato
- Joined
- 28 Sep 2014
- Posts
- 3,666
- Location
- Scotland
Wow really very surprised to learned that Maxwell used tile based rendering.
Maybe Maxwell was the first chip integrated 3dfx gigapixel technology that used tile based rendering properly after over a decade of many projects experiences with tile based rendering and find ways to integrated into Nvidia chips.
Maybe Maxwell was the first chip integrated 3dfx gigapixel technology that used tile based rendering properly after over a decade of many projects experiences with tile based rendering and find ways to integrated into Nvidia chips.
That does explain a lot and the efficiency was incredible really and nobody seemed to know what NVidia had done. Maxwell on 28nm Vs Kepler on 28nm was an incredible piece of work and using the same node size, efficiency was amazing.
I saw this tech program back in the late nineties and they were on about tile based rendering being. So I assumed this process was used ages ago in all modern GPU's.
Kinda surprised its only being used form Maxwell onwards by nvidia.
Kinda surprised its only being used form Maxwell onwards by nvidia.
NVIDIA must have designed Maxwell what 5 years ago? 6?
Do you think they've been sitting on their butts since then, or do you think they are researching and developing the next massive efficiency gaining technology?
I know what I think
By the time AMD manage to implement this, I'm sure NVIDIA will have rolled out the next game changing tech into their architectures.
We just need some decent competition for them - like any company they are out to make money and prices go up and R&D budgets tend to go down with no competition to incentivise making their shizzle better!
LOL, PowerVR had tile based rendering a decade ago.
Tell me how it's an nVidia innovation?
It did state GPU development. What PowerVR done with tiles was shocking, buggy and didn't real solve the issues. Nvidia have managed to get the tech to work so that is where development counts.
I don't particularly like Nvidia as a company but they offered excellent bang for buck with the 9 series. Before that I was on AMD so I can see it from both sides.
AMD was busy pushing Mantel at the time and this has allowed for other developments. They are both pushing, just in different directions at different times.
Problem with AMD is they always appear to be doing the wrong one at the wrong time and this seems to hit them hard. Should DX12 and Vulkan been pushed earlier then the cards on GCN would have seen the benefit then.
Swings and roundabouts for companies that we have to wait half a decade to see how things play out. I would suggest at the moment that due to the architecture of AMD they will have the advantage should DX12 and Vulkan pick up enough prior to Nvidia providing an architecture update that really utilises the benefits brought by DX12/Vulkan accordingly and then it will swing the other way (likely to be Volta).
Interesting article. I heard speculation of this some time ago Maxwell proved to be a massive ump in efficiency and NVidia talked about it being one of their most significant architecture advances ever, the basis of which they expected to reuse for generations to come. Maxwell is a bit like GCN version 1, AMD have made several iterations , Maxwell is gene 1, Pascal Gen 2 Volta gen 3 and likely there will be at least 1 or 2 more generations to come.
Maxwell is really where Nvidia let AMD behind at the starting blocks. It has taken AMD a whole new generation and a whole new node process down to 14nm in order for them to match the performance per watt of Maxwell. AMD simply wont close that gap until they come out with a brand new architecture but it isn't clear when that will happen. AMD hyped up Polaris as something revolutionary when it was barely any different to Fiji, AMD claims 15% architectural improvement. That seems to indicate that Polaris was just some kind of stop gap until Vega comes out. It has been speculated that Polaris was supposed to be the replacement for Hawaii on 20nm but when that process was cancelled they just did a re-spin of Hawaii.
I expect Nvidia tile based rendering will show greater benefits as games get more and more advanced
Maxwell is really where Nvidia let AMD behind at the starting blocks. It has taken AMD a whole new generation and a whole new node process down to 14nm in order for them to match the performance per watt of Maxwell. AMD simply wont close that gap until they come out with a brand new architecture but it isn't clear when that will happen. AMD hyped up Polaris as something revolutionary when it was barely any different to Fiji, AMD claims 15% architectural improvement. That seems to indicate that Polaris was just some kind of stop gap until Vega comes out. It has been speculated that Polaris was supposed to be the replacement for Hawaii on 20nm but when that process was cancelled they just did a re-spin of Hawaii.
I expect Nvidia tile based rendering will show greater benefits as games get more and more advanced
More than a decade...
Basically Kyro II in 2001 which competed pretty favourably with the GeForce 2 series. In some scenarios it matched/beat the GeForce 2 Ultra which was theoretically a far superior card, so even then showed the potential benefits of TBR. But lack of Hardware T&L, and R&D budget killed that off, unfortunately.
So not exactly innovative but still very interesting.
Kinda curious as to how much this is 'visible' to and/or can be exploited by DX12 and Vulkan, are we seeing the performance discrepancies because the low level apis favour immediate mode renderers, or that games developers are making assumptions on the Maxwell/Pascal architecture that aren't true? Or is the tile based 'bit' of the architecture hidden so far as to be not useful for exploiting even with low level API's.
Last edited:
The concept of tile-based redesign predates PowerVR. That isn't the innovation. The innovation is getting tile-based rendering within a modern GPU and getting it to achieve significant performance increases without a massive transistor budget. The innovations is going to be smart solutions to hundreds of tough engineering problems to make it a reality and to make it effective in real games.
PowerVR failed because here hardware just wasn't that fats, had missing features and just wasn't that good for real-world use at the time.
Missing T&L, yes.
Not that fast? The Kyro II a lot of us here owned absolutely smashed the previous perf/£ of the older cards. It was really cheap and performed incredibly well for its asking price.
It wasn't the fastest, no, but it was great value for money.
That said, I don't know why nobody continued making discreet PowerVR cards. Why they didn't add T&L and come out with a new gen of cards. Must have been a reason.
Maybe the margins weren't great; maybe there really is only room for one or two players in the GPU market. Heck, AMD are doing their best to make it one player only
Missing T&L, yes.
Not that fast? The Kyro II a lot of us here owned absolutely smashed the previous perf/£ of the older cards. It was really cheap and performed incredibly well for its asking price.
It wasn't the fastest, no, but it was great value for money.
That said, I don't know why nobody continued making discreet PowerVR cards. Why they didn't add T&L and come out with a new gen of cards. Must have been a reason.
Maybe the margins weren't great; maybe there really is only room for one or two players in the GPU market. Heck, AMD are doing their best to make it one player only
they made is cheap to break in to the market and suffered low margins because of it. Making value for money GPU is dad easy, just sacrifice future R&D and you can do an AMD.
they didn't have hardware TnL because they did';t have the transistor budget, spending a large chunk on the tile-based engine and they also didn't have the R&D budget to develop a competent solution. At the same time nvidia and AMD were adding new technologies to improve efficiency such as early z-culling, primitive rejection, compression. Moreover, DX was changing and a huge amount of addition complexity was going to be required for pixel sharers and the up and coming DX9
Soldato
- Joined
- 6 Feb 2010
- Posts
- 14,583
It's pretty ironic considering back then Nvidia was so far behind PowerVR in that gen, and they spent majority of the time in their events trash-talked about tile based rendering than talking about their own products
Last edited:
