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

[Lightman]

For what it's worth, if AMD's methodology for typical board power is consistent with other boards its given figures for, the stock consumption for FuryX appears to be the same neighborhood as Tahiti and Hawaii instead of being higher.

edit: found a more clear figure for the 290X's typical power, so Fury is slightly higher.

Which can be a result of improved PowerTune managing to dance even closer to the TDP limit.
I'm quite intrigued by AMD's open claim of building this card for overclocking. Hawaii was very medicore at overclocking even with very good coolers, but if Fiji can consistently reach 1.3GHz+ (10TF) it can be quite a popular card for overclockers and gamers regardless of power consumption in that state.

BTW do we already know where is Fiji manufactured? Is it still TSMC or did AMD move to GloFo?

 

[silent_guy]

Because you have a lot more clue, please list the complete BOM and I will tell you.

There is no need to list the full BOM. What matters is the difference due to using HBM.

- PCB size and less layers: a couple of dollars
- Power regulators: depends on overall power consumption. It doesn't matter much how it's split. Let's say that it's the same for Fury X and GTX 980 Ti. (We know the former is worse than the latter.)
- die size: roughly similar? Nvidia probably has better volume deals.
- Interposer vs none at all
- HBM memory vs GDDR5
- Watercooling vs air cooling

If you think that those last 3 will be less than $10 in extra cost, then there's really not much else to discuss. The Interposer alone is going to be more than that. But HBM is the elephant in the room.

 

[Jawed]

 

[fellix]

<pic>

Damn, it looks so much simple and easy without all the bulky memory chips.

 

[Deleted member 13524]

- Watercooling vs air cooling

The 980 Ti uses a vapor-chamber copper heatspreader combined with a huge dual-slot aluminum heatsink. Let's say it goes for $25 for volume orders?
The Fury X uses an Asetek AIO cooler, very similar to the 550LC with a 120mm fan. For retail, the 550LC can be had for 50€ and a silent 120mm fan goes for ~8€.
Wouldn't be surprised if volume orders for this AiO + fan would cost much higher than $50, assuming it also has an interface for the VRMs aside from the GPU+HBM.

I don't understand all the fud surrounding the watercooler, honestly.

 

[SimBy]

MOOR? What is that?

 

[Infinisearch]

Review NDA is June 24th supposedly.

Same as the release date, was hoping for something sooner.

 

[fellix]

http://www.moorinsightsstrategy.com...e-to-Radeon-by-Moor-Insights-and-Strategy.pdf

 

[Dave Baumann]

http://www.moorinsightsstrategy.com...e-to-Radeon-by-Moor-Insights-and-Strategy.pdf

"R9 390 & R9 390X each with 8GB of GDDR5 and 275W for 4K gaming "

 

[SimBy]

http://www.moorinsightsstrategy.com...e-to-Radeon-by-Moor-Insights-and-Strategy.pdf

Tnx, that's interesting. So that's some kind of a tech analyst company. I'm guessing it's just some sort of marketing deal?

 

[Gipsel]

Apparently just 64 ROPs.

 

[Scott_Arm]

64 ROPs is the max for the earlier 300 series, right? I guess with the shift to compute, maybe they feel an increase is not necessary?

 

[Alexko]

Any word on the FP64-rate?

 

[3dilettante]

64 ROPs is the max for the 200 series, right? I guess with the shift to compute, maybe they feel an increase is not necessary?

ROPs haven't ceased to be relevant yet, and with its glut of other resources I'd imagine Fiji would be more comfortably above Maxwell if something like peak fill rate for high resolutions were bumped up to match everything else. The 2x Tonga moniker seems more appropriate with that data point.

 

[fellix]

Well, at least AMD don't have to scale the number of setup pipes to match the ROP throughput.
And the 64 ROPs in Hawaii are already BW starved for FP16 writes and blends, so the extra juice from HBM could level off the utilization.

 

[UniversalTruth]

It just won't be future proof with 4K.

 

[silent_guy]

I don't understand all the fud surrounding the watercooler, honestly.

But you just came up with a cost difference of $25 between the two? I think the difference is going to be much lower than that, say, $15. And that's still huge.

 

[UniversalTruth]

But you just came up with a cost difference of $25 between the two? I think the difference is going to be much lower than that, say, $15. And that's still huge.

I have an idea. If the CEO cuts her compensations which are actually even greater than JHH's, and directs this money into engineering power, I think the company would feel much better. Lower margins or higher margins aside.

Ok, let's assume that you are correct and FIJI costs more, and so what given that idea?...

 

[lanek]

I will say that differently, lets assume Fiji cost a bit more .. why make you think that they will not do any profit at 650 $ or said differently, why do you assume that the 650$ cost of the Nvidia gpu's seems leaded by the manufacturing cost .. they could surely sell it at 500$ and their margin will still be good .

AMD need to go back to competitions, the 980TI is at 650$ , even if they are a bit faster, why will you they put their gpu at 750$ ?

 

[silent_guy]

I will say that differently, lets assume Fiji cost a bit more .. why make you think that they will not do any profit at 650 $ or said differently, why do you assume that the 650$ cost of the Nvidia gpu's seems leaded by the manufacturing cost .. they could surely sell it at 500$ and their margin will still be good .

Nobody said they won't make any profit. But at the same price, their margins will be lower, and probably quite a bit less than what they hoped to get.