Thursday, June 23rd 2016
AMD Radeon RX 470 Could Surprise with Pricing
AMD could deal yet another shock to NVIDIA after the Radeon RX 480, with its smaller sibling, the Radeon RX 470. This card is expected to be priced at $149 for the 4 GB variant, and $179 for the 8 GB variant. The card is rumored to feature 2,048 stream processors, spread acrosss 32 compute units, down from the 2,304 stream processors of the RX 480. Its memory ticks slower, at 7 Gbps, with a memory bandwidth of 224 GB/s. The most spectacular specification, however, is its typical board power, which is rated at 110W. The card should be faster than at least the R9 380X, and at its given specs, offer a very interesting option for 1080p gamers, at $149.
Sources:
WCCFTech, VideoCardz
67 Comments on AMD Radeon RX 470 Could Surprise with Pricing
Available 29th aswell?
There are a lot of things that we could not do before that we can now or that has gotten a lot easier with time and progress.
DX12 is new and different right? the idea that was sold to us was that if there are components that can render graphics etc, those components would help out.
So yeah unless indeed its all in the hands of the drivers which constantly change and update...
I thought DX12 would be a bit like that Lucid Hydra chip...although that has not taken off really...maybe for a reason, a well, idk I guess, but it sure would be nice to look back at this in 2 - 3 years and see proper DX12 multi gpu of whatever variety support :)
Very curious...
That makes things much harder since they can't off-load the work to someone else and there isn't (as far as I know) a tool to do the work for them. And even if there was, it's not certain they'd use it.
Intel (and a consortium of other companies) has created an open source tool (OpenMP) to easily make multi-threaded software including games. Developers refuse to use it even though it's a great tool that allows great scaling.
But to remain on-topic:
Of course the features in DX12 are advertised even if they aren't used because the capability is there for anyone to use but it's not a requirement nor an automatic process.
I'm thinking that DX12 might make integrating some multi GPU features (specifically Implicit Multi Adapter) easier than SLI or Crossfire ever was because it's built-in and perhaps the documentation is better but it probably ends there. Implementing the ability to use any mix of GPUs will be hard to implement and harder to implement well.
Making things more complicated never makes things easier, so just because it's possible now doesn't mean it's because it's easier. It just means the tools are available if you have the resources to use it.
If you think about it: things have become easier for the end-user but harder for the developer. And that even goes beyond IT-related products.
Heck the RX 460 should offer great 1080p (aka 960/380) at a $100 price point and no 6-pin, however I understood that would be a Polaris 11 (smaller) part compared to the Polaris 10 of the RX470/480.
But you can Crossfire two of theses for ~300. :peace:
I think that AMD is doing the smart thing with only having 4 chips in development. In the past they would have used 1 GPU per product stack (Tahiti for 7970/7950, Pitcairn for 7870/7850) now we're seeing them using 1 chip for 2 product stacks (Polaris 10 for 480 and 470).
It's a smart way for using their limited resources and once they get better yields and higher binned chips they can offer X versions with increased clockspeed (think 480X = 480@ >1.5Ghz)
Still, this may well be something that will be unannounced until Vega's release - sort of a mid-cycle refresh.
Layman's terms: Polaris is just too different from what came before to work together with older video cards in crossfire.
DX12 solution: permit multiple game-managed graphics queues and compute queues which allows completely different GPUs to work together more or less seamlessly.
"One of the most exciting parts of Microsoft's DirectX 12 API is the ability to pair graphics cards of varying generations, performance, or even manufacturers together in a single PC, to pool their resources and thus make games and applications run better."
arstechnica.com/gaming/2016/02/directx-12-amd-and-nvidia-gpus-finally-work-together-but-amd-still-has-the-lead/
If they are trying to gain the market share of the 750ti, this won't do it. It will be faster than a 750ti by probably at least 50% though.
1. 7950 was closer to 7970 in terms of price (82%) than what 470 at $149 would be (75%), so there ought to be more separation between 470 and 480.
2. As 14nm is a new process, the probability of a more cut down version is always higher in order to maximize yields.
So I would agree with you that there is a possibility for 2048 cores, since 2048@1-1.05GHz would give roughly same performance as 1792@1.15-1.2Ghz, it's just that I think the latter is more probable.
Looking at it another way, 950 was 5-10% faster than 370. Now 460 will have the same core count as 370 (1024 cores) and will feature a higher clock... So it should be around 10-12% faster. Thus effectively it should be 2-5% faster than 950.
So the newer cards hierarchy is looking like:
RX 460: Good 1080p performance (Firestrike score ~6k)
RX 470: Excellent 1080p performance (Firestrike score ~9k)
RX 480: Outstanding 1080p performance (Firestrike score ~12k)
or Good 1440p performance (Firestrike Extreme Score ~6k)
GTX 1070: Excellent 1440p performance (Firestrike Extreme Score ~8K)
RX 460 is very interesting too, I'm pretty positive that it'll touch 960's performance, for $100, I think it would be my next card for my second PC.
Same thing happened with gtx980 at around 400mm2 die size vs hawaii(r9 390x) at 440mm2, amd in this case is somewhat more competitive than tonga was because hawaii is a newer revision of gcn so with only 10% extra die space they competed, however the power consumption was far greater on amds side.
With all this being said I would also like to make a suggestion for reviews to focus a bit on this aspect of not only performance/watt and performance/dollar, but also performance/mm2 of die size. Because that in my opinion is a clear indication on who has the superior architecture or the better business model, and basically in other words who can achieve higher performance with better cost efficiency