AMD R9 Nano Review

[the54thvoid]

Realize you're talking about two industries. The computer industry has/is rapidly adopting DisplayPort. The TV/film industry previously adopted HDMI and is pushing to switch to the updated standard. Going back more than a decade, graphics cards have traditionally supported two or more monitors (VGA, DVI, DisplayPort) and one TV (RCA, S-Video, component video, HDMI). Fiji is staying true to that just like virtually every card before it.


As pointed out by Xzibit, HDMI 2.0 can't even handle all of the bandwidth 4K requires at 60 Hz, again, because it is a bad standard. The problem, is not AMD refusing to put HDMI 2.0 on there; there problem is the TV/film industry refusing to switch to DisplayPort. I just looked for a Home Theatre receiver that implements DisplayPort switching over HDMI and I couldn't find any. Therein lies the problem. It has nothing to do with AMD and everything to do with the TV/film industry clinging to HDMI.

All good arguments but it should be deliver for what is current, prepare for future, i.e. supply both.
I can't argue technical points but I'm taking the words of numerous established tech reviewers over forum posters here.
If HDMI 2.0 is crap, why are those sites saying it was a poor design decision for AMD not to implement it? That's rhetorical, doesn't require an answer.

 

[F-Zero]

I think it's an alright product but the 650 dollar price, coil whine (from what i have heard) and the fact that you can get the FURY X for the same amount of money is just killing the NANO.

 

[FordGT90Concept]

Because it is a feature Maxwell has that Fiji lacks.

 

[Outback Bronze]

Sweet little card

Bummer big price :-(

 

[Ferrum Master]

I also agree that that demand for HDMI 2 is rubbish. Most of 4K TV's are still plagued with serious input lag, and each specific panel problems and using them for gaming... The lack of proper 4:4:4 is hillarious. So far there are only few Panasonic and Samsung TV's/Panels utulizing display port... so far... For the sake of the industry, I hope there will be more.

Performance wise... Nano is okayish card... we all know that. The mistery remains why there are no proper two slot air colled full fiji card...

 

[BiggieShady]

Okayish is generous. That cheap VRM array reaches 93 C on open test bed ... and the card is meant for cramped ITX cases. Nevermind the coil whine.

 

[the54thvoid]

The mistery remains why there are no proper two slot air colled full fiji card...

If you join my 'All Fiji products are contrived' scenario it's easy to see.

Fury X is made with block - extremely new take on a stock top tier card. The AIO solution makes it 'unique'.
Fury lacks the water solution, thus comes in a with 'standard' albeit custom air cooling. A few cores disabled to make spec different (binning perhaps).
Nano is made smaller but only castrated via power limits and power input.

All have almost identical core (Fury X = Nano =/= Fury<core count>). The only way AMD could make a gradient to the consumer was to make different 'flavours' of Fiji. Yes, Nano could have higher TDP even with one 8 pin but the temps would be too much for the fan. A secondary BIOS on Nano for a FC water block solution would allow far higher performance but threaten Fury X's top position. Likewise, Fury is a simple Fiji castration (a la 980 to 970 without the memory nonsense). Nano can't be given the same perf as Fury X under water (TDP limit) because that would destroy Fury X sales. This can also explain Nano's Fury X pricing. If I want to fit a water solution, Fury X is already there. For Nano it's far more expensive so in a normal PC case, Fury X is the standard option. If Nano was $80-100 cheaper then it would be the go to card by adding a w/b, cannibalising Fury X. Nano's rated power draw is less than it's rated PCI-e and 8 pin potential so it's been hobbled at firmware level?

Fiji's product stack is all about fitting a requirement. Fiji is probably too difficult to section off into lower parts (or the process is too new to be so bold). I think this is AMD's answer to creating some tiers of Fiji and in hindsight, by keeping the nomenclature distinct from the 3xx series at least they are saying Fiji = prestige, R9 3xx is mainstream.

 

[Ferrum Master]

Okayish is generous. That cheap VRM array reaches 93 C on open test bed ... and the card is meant for cramped ITX cases. Nevermind the coil whine.

The VRM is same as on averge Z87 board... exactly the same... as probably the itx board will cary... it can feed around 300W with proper mosfets without problems, it is a proven old solution. The whine is an accident, you can't predict the crappiness of those inductors, lately they seem to be very bad in this aspect.

 

[Aquinus]

Is it me, or does Nano feel like it should have had Fiji Pro instead of XT in it? This feels like a whole lot of GPU for the amount of power delivery it has.

I think AMD got it backwards, Fury should have had Fiji XT and Nano have Pro instead of the other way around. Less power draw would mean less stress and heat generated by the VRMs.

 

[Ferrum Master]

Is it me, or does Nano feel like it should have had Fiji Pro instead of XT in it? This feels like a whole lot of GPU for the amount of power delivery it has.

I think AMD got it backwards, Fury should have had Fiji XT and Nano have Pro instead of the other way around. Less power draw would mean less stress and heat generated by the VRMs.

Then it would make it slower than 970 and 290X, the problem is that Fury X should have been faster, and all problems would be solved, but the thing just doesn't clock. Hello 28nm...

 

[the54thvoid]

Then it would make it slower than 970 and 290X, the problem is that Fury X should have been faster, and all problems would be solved, but the thing just doesn't clock. Hello 28nm...

Architecture makes a huge difference to clocks though. Maxwell is on the same node as Kepler yet the Maxwell top part clocks at 1400-1500 whereas Kepler struggled past 1100-1200. Same 28nm but Maxwell dropped a lot of the compute functionality which GCN has kept (for forward thinking to DX12). Clock for clock, Fiji would demolish Maxwell but what Maxwell gave up on from Kepler put it ahead enough for 2015/2016.

But yes, next node will be amazing to see from both AMD and Nvidia.

 

[Aquinus]

Hello 28nm...

That didn't stop NVidia, now did it? How much faster do you think the 390/390X would be if it could clock VRAM up to 2,000Mhz (8GT/s) and the core up to 1,400Mhz? It's not like the 980 Ti is a small die either and it's also on 28nm. My point is that AMD hasn't been doing a very good job lately, at least with current hardware with current software. Async compute might be great but, I'm still playing DX11 games for a while to come...

 

[Ferrum Master]

But yes, next node will be amazing to see from both AMD and Nvidia.

Sometimes I think we need a third player and that's it... Intel should quit screwing and do standalone chips.