AMD: Pirate Islands (R* 3** series) Speculation/Rumor Thread

So the rumour for the next chip's performance is, wait for it, about the same as Titan X. Oh what a surprise.

Yeah, but will be interesting to see an apples-to-apples comparison when all three GPU's are "under water".
 
"Up to 8GB of ultra-high bandwidth High Bandwidth Memory video memory"

HBM2, is 4 or 8GB.. someone can enlight me if you need a 8192bit bus for HBM2.0. or it is just a question of the density of the stacked die ? (2Gb instead 1 ? )
 
HBM2, is 4 or 8GB.. someone can enlight me if you need a 8192bit bus for HBM2.0. or it is just a question of the density of the stacked die ? (2Gb instead 1 ? )
Hbm2 isn't ready yet, options are 4gb and fake slide, 8k bit membus or hynix doing double density even though specs don't mention it being possible
 
Hbm2 isn't ready yet, options are 4gb and fake slide, 8k bit membus or hynix doing double density even though specs don't mention it being possible

i dont speak about the slides or rumors we see, but about the technology of HBM2 .. because if you need a 8K bit memorybus, thoses rumors are fake effectively.
 
It depends, I while the specs don't mention it being possible, I don't think there's any technical issue preventing Hynix from using 4Gb dram chips in 1st gen HBM
 
I wonder if there's a latency cost associated with double-density "version 1" HBM, if that is what this turns out to be for 8GB.
 
I also have a hard time believing that if it has similar performance to GTX Titan X that they'd price it at almost half of that and undercut GTX 980ti as well. Then again with the massive gap in marketshare, perhaps they'll undercut massively just to try to get marketshare back.

Regards,
SB
 
I also have a hard time believing that if it has similar performance to GTX Titan X that they'd price it at almost half of that and undercut GTX 980ti as well. Then again with the massive gap in marketshare, perhaps they'll undercut massively just to try to get marketshare back.

Regards,
SB

The question is maybe more, where Nvidia go search this price ? .. 3 years ago, we had dual gpu's in the 1000-1300$, peoples was cry about the price of an Asus Mars, custom, limited edition dual gpu's ( 1200$ ) ..now if they sucess to sold a single gpu at this price.... Good job.

Nvidia is just trying to push their lineup prices to higher level since the 700 series...
 
Last edited:
In addition, this paper, is a research work on what will be needed for end to TOP-PIM ( Just read this paper entierely and you can be sure we are not ready to see it on a product before a long time).. I will not base myself on the "image" like this who are just used as "example" in this paper.
 
Last edited:
So the rumour for the next chip's performance is, wait for it, about the same as Titan X. Oh what a surprise.

http://wccftech.com/amd-r9-390x-nvidia-gtx-980ti-titanx-benchmarks/
http://wccftech.com/amd-r9-390x-nvidia-gtx-980ti-titanx-benchmarks/
And it uses more power. Though nothing worth getting excited about, but I'm sure the trolls will be all over the tiny difference.

Looks like a repeat of R600 if this really is an HBM chip.
What's the point of having unlimited bandwidth when you don't have the computing resources to use it? HBM and 28nm don't mix well.
 
Whats the point of having anything then?
That's a deep philosophical question for which I'm ill equipped to give it the answer it deserves. But I'd say that the point of having is primarily to use it for something.

R9 290X has 352GB/s in BW. A GTX 980, which outperforms the 290X quite easily most of the time, has 224 GB/s. Now I know that the 980 has compression etc, but I don't think that will compensate for a 57% difference in BW. It's thus very reasonable to say that Hawaii is not significantly held back by BW. And an imaginary Hawaii with the Tonga compression improvements even less so.

If Fiji has 64 CUs instead of the 44 of Hawaii, that's only an increase of 45%. Yet the bandwidth that was already generous goes up to 640GB/s or 81%, even without compression improvements. Adding this much BW would have been fantastic if the number of CUs could have been increased accordingly, but the practical limits of 28nm make that very hard. IOW: HBM and 28nm don't mix well. We'll only see the real power of HBM when process technology catches up as well.

Edit: I see that WCCFTech speculates that Fiji will have a die size of 550mm2. That's probably completely made up, but if it's true, then that'd be a crucial mistake on AMD's part: it'd be leaving a ton of HBM performance on the table.
 
Last edited:
That's a deep philosophical question for which I'm ill equipped to give it the answer it deserves. But I'd say that the point of having is primarily to use it for something.

R9 290X has 352GB/s in BW. A GTX 980, which outperforms the 290X quite easily most of the time, has 224 GB/s. Now I know that the 980 has compression etc, but I don't think that will compensate for a 57% difference in BW. It's thus very reasonable to say that Hawaii is not significantly held back by BW. And an imaginary Hawaii with the Tonga compression improvements even less so.

If Fiji has 64 CUs instead of the 44 of Hawaii, that's only an increase of 45%. Yet the bandwidth that was already generous goes up to 640GB/s or 81%, even without compression improvements. Adding this much BW would have been fantastic if the number of CUs could have been increased accordingly, but the practical limits of 28nm make that very hard. IOW: HBM and 28nm don't mix well. We'll only see the real power of HBM when process technology catches up as well.

Edit: I see that WCCFTech speculates that Fiji will have a die size of 550mm2. That's probably completely made up, but if it's true, then that'd be a crucial mistake on AMD's part: it'd be leaving a ton of HBM performance on the table.
I think its a nice addition even if it is not being used to its full potential. The bottleneck is going to have to be somewhere and if bandwidth bottleneck can be ignored, the whole system will perform better in some cases where bandwidth does matter. There is likely 128 ROPs on Fiji and that will need the bandwidth if you are pushing higher resolutions I'd imagine. We can assume the latency improvements might also be good as well. This is all on top of a power reduction too. I think HBM is a great technology to be implementing even if its not being pushed to max potential.

AMD likes higher density dies to larger dies. Its the same with hawaii. They are already at the power limit they are comfortable at, adding more CU will likely not improve efficiency or performance by much at that point. They save some money by making the die smaller and clocking it slightly higher to get the same performance out of the same power budget as a larger chip.
 
Fiji could simply be filler. I doubt anyone really expected to still be on 28nm for flagship gpu's in 2015. So perhaps this is AMD's answer to NVidia in the first half of the year.

Put out a power hungry gpu on 28nm and reduce costs and power draw with HBM . Then in the fall when they might actually be able to get a 16nm gpu out there go full out with the die drop and HBM .
 
if it's true, then that'd be a crucial mistake on AMD's part: it'd be leaving a ton of HBM performance on the table.
ROPs could eat that bandwidth tho, even a game like world of warcraft stutters on a 290X at 1440P if you throw in a little supersampling, especially whenever there's alpha blending involved.
 
Back
Top