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

What would they care about then? Not that I believe this rumor is true. Rumors about someone buying AMD surface at least twice per year....for the past 10 years.

AMDs integrated graphics IP is quite good, and by all accounts they're doing something with ARM :)

My point was that folks earlier in the thread seemed to be getting excited about the prospect of a company as large as Samsung wading into the fight with Intel and NVIDIA in traditional PCs, and possibly consoles. Why would Samsung want to do this? In fact why would any company want to buy AMD in order to compete with Intel on its home turf?

For the record I don't believe the rumour either, but some day or other AMD will get bought out, and personally I doubt it'll be for its x86 license.
 
Samsung already makes a ton of pc parts . AMD graphics would be a huge boon to its mobile chips , they would get a piece of the console pie (even if they don't launch their own) they will also get amd's desktop/ laptop gpu market share. Samsung already has cutting edge fabs so brining the production in house would help them save money and could give them a fab advantage over nvidia
 
Samsung is so big that it's like they are different companies, and for this they barely use their own arm chips, so why care about using their x86 in notebook?
 
AMD Fiji has two GPUs?

However, AMD raised our suspicions about the Fiji design when they announced LiquidVR SDK, where they clearly stated that you need two GPUs. The slide below clearly implies that you need Affinity multi-GPU rendering in order to reduce latency and increase content quality.

8e4d4a3e1140a63618effa6ff07b16ee_L.jpg


amd-liquid-vr1-features.jpg



http://www.fudzilla.com/news/graphics/37373-amd-fiji-has-two-gpus
 
The multi-GPU component would require multiple GPUs. Is there an AMD quote to the effect that the other parts of the tech require it?
 
Someone can tell me what he's smoking ? (Fuad i mean )

Multi gpu affinity is in the case you have multiples gpu's installed ( each gpu's are able to render " 1 eyes " panel ).. This said, it is an option IF you have multiple gpu's installed. Other techs who are part of the SDK are not concerned.. ( its like a new CFX mode available finally specific to VR ).

AMD succeed to reduce highly the latency with their SDK, by bypassing completely windows,drivers etc. In addition of the headtracking and movement detection . This absolutely dont require multiple gpu's.

But what was intend to said AMD is, with this mode enabled, 2 gpu's will offer a better experience, better quality with VR,.´than what you can got with one.

And if Fidji is made of 2 gpu's .. what it will be ? 2x Tonga ? offering 7990 performance ? lol

Following the "rumored", but coherent specs we have,

- 4096SP, so what? 2x 2048SP =( 2x TongaXT or 2x 7970 )
- 4096bit bus ? 2x 2048bit or 2x 4096bit bus ?

It dont make any sense for me.
 
Last edited:
That's the strangest rumor I've read in quite a while.

Following the "rumored", but coherent specs we have,

- 4096SP, so what? 2x 2048SP =( 2x TongaXT or 2x 7970 )
- 4096bit bus ? 2x 2048bit or 2x 4096bit bus ?

It dont make any sense for me.
Good point, since 2x 2048-bit would still give a total of 4 GB memory (normally), not 8 GB as Fudzilla claims, and 2x 4096-bit may result in a ridiculously high amount of bandwidth.
 
But what was intend to said AMD is, with this mode enabled, 2 gpu's will offer a better experience, better quality with VR,.´than what you can got with one.
Well, more specifically, this is a special Crossfire/multi-GPU mode for VR that splits the workload one GPU per eye, rather than standard AFR rendering which introduces a frame of latency (roughly speaking giving a similar latency effect as just having one GPU).
 
Good point, since 2x 2048-bit would still give a total of 4 GB memory (normally), not 8 GB as Fudzilla claims, and 2x 4096-bit may result in a ridiculously high amount of bandwidth.
Latest HBM-slides suggest even HBM1 can have 8GB on 4096bit membus (2GB per module, 4-Hi, 2x 2Gb per layer)
 
Let say this differently, if the theory of Fuad was, AMD have a dual gpu based on 2 Fiji cores and this will be the "8GB" version... why not... But if his theory is that the 4096SP /4096bit Fiji is in reality a 2x 2048SP cores gpu's.. i absolutely dont see how it will be possible.

This and the " TongaXT" with 35CU / 2240SP articles who have been made today , look more like 1st April Fool.

( This said, if AMD have prepared a dual Fiji ( that i doubt due to 28nm), and that the spec we have are correct, this will be a monster: 8192SP, in the 15-16Tflops.. )
 
Last edited:
This and the " TongaXT" with 35CU / 2240SP articles who have been made today , look more like 1st April Fool.

Having looked at it, I believe the same site (sisoftsandra) was reporting 80 shaders instead of 64 per unit, making the shader count higher than it should be (because their older stuff had 80/unit). So instead of 35CUs (64x35) it'd be groups of 80 shaders, 2240/80 = 28, 28x64 = 1792 which we know is an actual GPU.
 
Having looked at it, I believe the same site (sisoftsandra) was reporting 80 shaders instead of 64 per unit, making the shader count higher than it should be (because their older stuff had 80/unit). So instead of 35CUs (64x35) it'd be groups of 80 shaders, 2240/80 = 28, 28x64 = 1792 which we know is an actual GPU.

Yes, i know from what it come, just, im sorry but all the lately news about it posted on WCCFtech and Fudzilla, look like April Joke, posted some days too early .. I dont know.. maybe a guy have decidded to post all his april joke 2 days before, as it will be in holliday next week..

One day, i will make a backup of all thoses articles, corrected or not, and just do a full timeline on how they can report so much BS, without even try to use their brain for understand, nothing is matching... ( in the comment section, the guy apologize for the chart, and they have now updated the article) ..

( i dont even speak about all the comments done in WCCFtech about this story.. its so funny to see all the fanboys, troll, from a camp or the other starting their own story from that. .. )

Seriously, does thoses guys have at least half a brain they can use before post this type of BS in their sites ?

Sorry, it is really not thread for speak about half brain jurnalist or half brained sites, but well.
 
Last edited:
Having looked at it, I believe the same site (sisoftsandra) was reporting 80 shaders instead of 64 per unit, making the shader count higher than it should be (because their older stuff had 80/unit). So instead of 35CUs (64x35) it'd be groups of 80 shaders, 2240/80 = 28, 28x64 = 1792 which we know is an actual GPU.

Tonga in the iMac has 2048 shaders as far as I am aware.
 
From Fudzilla: "Fiji XT can support HBM2 memory."

HBM2 won't work out of the box with Fiji, but since this memory will become widely available at a later date, probably toward the end of the year, AMD has enough time to get ready.
[…]
It is not clear if Fiji needs another revision to support HBM2, or if a simple PCB revision will do the trick, but the company will make necessary adjustments in order to support the second generation of this super-fast memory for the Fiji XT core.
 
Yes, i know from what it come, just, im sorry but all the lately news about it posted on WCCFtech and Fudzilla, look like April Joke, posted some days too early .. I dont know.. maybe a guy have decidded to post all his april joke 2 days before, as it will be in holliday next week..

One day, i will make a backup of all thoses articles, corrected or not, and just do a full timeline on how they can report so much BS, without even try to use their brain for understand, nothing is matching... ( in the comment section, the guy apologize for the chart, and they have now updated the article) ..

( i dont even speak about all the comments done in WCCFtech about this story.. its so funny to see all the fanboys, troll, from a camp or the other starting their own story from that. .. )

Seriously, does thoses guys have at least half a brain they can use before post this type of BS in their sites ?

Sorry, it is really not thread for speak about half brain jurnalist or half brained sites, but well.

On OcUK there is already an effort

http://forums.overclockers.co.uk/showthread.php?t=18647127
 
Back
Top