AMD R9 Nano official specs (and later, reviews)

The performance/price ratio of this graphics card puts it at an extremely niche market, to which AMD is terrible at catering to because of their lack of popularity (unlike nVidia with the Titan series).
Moreover, the crowd with the really tiny mITX cases will probably want to have them in the living room. That means HTPC, which in turn means they want HDMI 2.0. Which the Nano doesn't have, and I doubt the expected DP->HDMI 2.0 adapters that are coming will handle HDCP so there's a big problem right there.

What's the point of developing an interesting product if you're not interested in selling it?

 

$650!?!?

Have you lost your ever loving f-ing minds AMD?!?!

/me is frightened, confused, and displeased!

 

The idea seems to be that they're offering something unique, in that the Nano performs at a level that is unmatched by any card of this size, which has value for people who want compact systems.

That may be true, but in practice I'm skeptical that it will convince a significant number of buyers.

 

It's just telling me how badly they're banking on Quantum, they NEEDED the Nano for Quantum regardless of what the MSRP would be of the product.

Heck, mebbe they're holding back a bunch for a fall Quantum release and that could also be driving up the price. (<-Purest speculation, nothing to back that up!)

 

A more reasonable assumption - they only have a few of these to go around, and are trying to skin the handful that'll actually be interested. By pricing it this high they ensure that they don't end up with demand they cannot satisfy (under viable economic conditions, this appears pretty boutique). I do expect numerous "we're selling all that we can make" claims, without any substantiation as to how many units make up an "all".

I'll give them the benefit of doubt here. If any product planner actually thought "OMG this is a totally unique and enticing piece of kit that will sell itself on being small and cute, and will not look ludicrous if priced ludicrously" it's just...sad.

 

Mebbe if you can put one in a laptop, that might be cool...

 

But the Quantum uses either a full fledged Fury X or a Fury MAXX dual-Fiji graphics card. The Quantum doesn't need the Nano at all...

 

All the disclosures about Quantum so far indicate it's more niche than Fury Nano, and it's capable of handling Fury X, and possibly the dual-GPU card as well.

http://www.pcworld.com/article/2973...y-and-powerful-project-quantum-dissected.html

If the idea was that Nano's volumes would match a non-product like a very unorthodox form factor with custom machining and motherboard, AMD probably could have doubled the price for all the volume it'd get.
Perhaps the FirePro variant might be useful, or some Fury Nano successor might once memory capacity increases.
The virtualization features and form factor might provide benefits in a remote rendering server or some workstation/compute capacity. The older monitor output standards would matter less for the remote rendering case.

 

This is only tangentially related to the Nano, but the reduction in size enabled by HBM has me wondering about integrated voltage regulation. AMD claims to be working on it but they haven't specified (to my knowledge) whether that's for CPUs, APUs, GPUs, some or all of the above. I don't know how much power it might save on a GPU, or how much more efficient power management could be, but it should allow IHVs to make seriously tiny graphics cards, perhaps so tiny that the PCIe slot would become the limiting factor.

 

It seems like IVR can go either way, at least for now, given the tradeoffs Intel has made in using and not using it, and their materials science is beyond what AMD's manufacturing options have demonstrated so far.
AMD's projections for using interposers to de-integrate specialized silicon provide a way to put a regulator module on-interposer, which should provide a good amount of the benefits, but then it would have to be cooled right next to the die and HBM.
The fact that Fury Nano has a separate heat pipe for its adjacent power phase circuitry may mean power density is not a solved problem.

I did note this:
http://www.anandtech.com/show/9564/amd-announces-radeon-r9-nano-shipping-september-10th

This is the first time I've specifically seen 85C discussed as a thermal ceiling. I've been wondering if that would come up now that DRAM is under the same cooler, as that temp does show up in research for processing in memory and DRAM thermal behavior.

 

Hawaii's thermal ceiling on the 290X is 94ºC, and AFAIK the GDDR5's heatsink is connected to the GPU through heatpipes.

 

Which model are you referencing? The 290x standard cooler has the GDDR5 chips first touching a base plate, not the plate touching the GPU. This is significantly more indirect, and the required large amount of space that is a supposed disadvantage versus Fury means that heat has far more opportunity to dissipate when the DRAM isn't less than a millimeter away.

edit: Does that plate touch the heatpipe assembly in a significant way?

 

My model has 6 GDDR5 chips (left row + single chip below the GPU) in direct contact with the "big" heatsink, which also covers the VRMs:





I thought the reference cooler did the same.

 

The pictures of an unmounted standard cooler show the GDDR5 thermal pads touch a black metal base plate, not the copper plate of the main cooler. The copper GPU cooler base bulges downward to couch the GPU. I haven't seen a shot of how tightly mated the rest of the copper heat chamber/heat pipe assembly is to the metal the GDDR5 touches.
The pictures you show for that cooler also shows that there is a different piece of metal that the GDDR5 touches, and even then I think it's only touching some of the modules. How tightly bound that metal plate is to the actual internals of the heat pipe assembly is unclear.

The DRAM shouldn't need that much cooling, particularly since there's a mounting plate over a lot of it. It does not look like the heatpipes in that portion are trying all that hard to be uniform in length or shape, so it might have poor contact there.

 

The DRAM chips I mentioned (left row + below the GPU) are definitely in direct contact with the GPU's heatsink, since there are even thermal pads for those:



My only point is that at least those chips are reaching 94ºC and operating at those temperatures (and they did so in my cards quite a few times until I changed my old case.. this summer is being a bitch for my gaming rig..). Maybe I misunderstood, but from this post I thought you suggested that the DRAM's thermal limit would be lower than the typical GPU 90-100ºC limits?

 

If FuryX is already very aggressively priced compares to production cost, then the price of the Nano kind of makes sense since the only difference is the water cooler.

Whether it makes sense for a consumer is something else...

 

DRAM can function at those temps, but refresh rates typically enter a high-temperature mode at 85C that hurts bandwidth.
Depending on the model, the memory may be unable to report anything but a value indicating it is hotter than 95C, which makes operating above that limit risky.

http://www.hynix.com/datasheet/pdf/graphics/H5GQ1H24AFR(Rev1.0).pdf (pg 119)

This is of course for non-stacked GDDR5.
If that cooler was for some reason heating a subset of the GDDR5 chips to 94C, then I would consider that a flaw. It might be a few degrees short of some kind of emergency throttle.

 

But the throttle kicks in - pretty aggressively - at 94ºC. I've never seen the GPU temps reach 95º.

What? Where did you see that?

 

That would be the GPU's thermal management kicking in.
If the cooler itself has reached a steady state where something as far away as the GDDR5 chips touching that baseplate are at 94C, the GPU throttling to keep at 94C would not stop the GDDR5 from hitting a stretch of higher than normal consumption.

When you say the GDDR5 chips hit 94C, which monitoring tool was giving that data?