FP64 will still have a poor ratio vs FP32, and naturally people will complain about applications written for the GCN architecture performing badly on Pascal as well. Or that NVIDIA's OpenCL drivers for macOS are terrible. Or something like that.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
AMD Confirms RX 480 At $199 USD, Other APU & Polaris Announcements
- Thread starter lowendlinux
- Start date
- Sort by reaction score
FP64 will still have a poor ratio vs FP32, and naturally people will complain about applications written for the GCN architecture performing badly on Pascal as well. Or that NVIDIA's OpenCL drivers for macOS are terrible. Or something like that.
All. Even 1080 appears to be weak in some tests. Without that high clock speed it would be clearer that Nvidia has been weakening compute capabilities for each generation.
Ok I got some hint that the MacBook Pro is going to contain an Nvidia chip but I'm not believing it until I see it. Look for clues in Sierra DP4-5
RX 490. Few possibilities.
1) Polaris 12 GPU with 3072 GCN4 cores, and 4 GB of HBM1, with performance between GTX 1070 and GTX 1080.
2) Dual Polaris 10 GPU.
3) Vega 10 with GDDR5(X) memory.(unlikely because that would increase power draw of the GPU, and RX X9X GPUs are supposed to have wider than 256 Bit bus).
4) Vega 10 with HBM2 memory.
The most tempting possibilities are 1 and 4. Everything else will not be that interesting.
1) Polaris 12 GPU with 3072 GCN4 cores, and 4 GB of HBM1, with performance between GTX 1070 and GTX 1080.
2) Dual Polaris 10 GPU.
3) Vega 10 with GDDR5(X) memory.(unlikely because that would increase power draw of the GPU, and RX X9X GPUs are supposed to have wider than 256 Bit bus).
4) Vega 10 with HBM2 memory.
The most tempting possibilities are 1 and 4. Everything else will not be that interesting.
Just because the current lineup stops at the RX 480 doesn't mean there will be a RX 490. Once Vega comes out AMD could just iterate to its 500 series. Basically, these names are marketing and don't indicate future products.
I don't think any form of Polaris is getting any bigger. The only other Polaris chip we know about is 11, which is smaller. There could be a ~3000 GCN core chip but it would probably be some iteration of vega.
The only thing I could imagine is if AMD brought out an overclocked version of Polaris 10 with GDDR5X. Its not a very exciting product as it would probably be something like 390X performance at 200 W.
There is no reason to make a dual GPU card for a midrange chip. This wouldn't even beat out the pro duo.
As for Vega, I think whatever AMD is bringing out at the high end will have HBM2 but that doesn't mean they won't adopt GDDR5X on some product.
AMD already by mistake leaked the information about RX 490 that is going out before the end of the year.
Code you can redeem up to the end of February 2017. Computer with eligible GPU or CPU must be bought before the end of the year.
What information? Do you have a link? If it is coming out this year then I doubt it will based on silicon we haven't seen yet. There have been several stories over the last few days saying that vega is coming next year, which AMD has said itself.
Ah, well either its a mistake by marketing or maybe we will see an overclocked Polaris 10 that could beat out nvidia's upcoming gtx 1060.
That will happen with OC'ed aftermarket RX 480. If it will really be faster than RX 480, in the first place.
I could see it being something like the 290X -> 390X or they could add GDDR5X to it.
Well, that would make RX 480 into RX 485. Even naming scheme would indicate it as possibility.
Well, that would make RX 480 into RX 485. Even naming scheme would indicate it as possibility.
I am sure you could put some marketing spin on it to say that GDDR5X makes it effectively greater than a 256 bit bus.
Nvidia will announce mobile Pascal on August 1st. Check the beta builds soon after.
I see you constantly quoting theoretical FLOPs and GBs, and trying to convince us to buy (or hold off buying) based on mathematical calculations of peak performance from suspect or leaked specs of unannounced (and possibly not even taped-out GPUs).
I'll be waiting for honest benchmarks from real applications before I pull anything out of my "next Nvidia card" savings account. I want to see CUDA and OpenCL numbers for real applications. And I don't want weird cherry-picking of FP32 vs FP64 benchmarks. FP64 is irrelevant for me - I need compute performance on FP32 and FP16.
ALL the first batch of 4GB 480s are actually 8GB models that can be flashed to the full spec. If you're lucky you could even get an overclocked gold sample. There will probably a list of product SKUs to look out for.
wccftech.com/amd-rx-480-4gb-retail-cards-8gb/
wccftech.com/amd-rx-480-4gb-retail-cards-8gb/
I don't know what are you referring to, however, it was just posted as informative post, my friend.
You could use it whatever way you would like.
That is because... you can't buy 4Gb RAM chips. Only 8 Gb models are produced currently. 4Gb is EoL both on SK Hynix suite and Samsung.
Last edited:
Let's see what the driver update today brings to the table.
No performance penalties they promised.
The card is already sucking juice from both sources so no miracles there, diverting from the PCIe will only overload the other source.
No performance penalties they promised.
The card is already sucking juice from both sources so no miracles there, diverting from the PCIe will only overload the other source.
