AMD R9 Nano official specs (and later, reviews)

I'm just wondering how long it'll take for an AIB to make a mITX GTX 980 and give the Nano a run for its money and power efficiency.

The Nano is faster of it, so untill they OC the 980 and sold it at 450-500$ .. For a niche market like that.. what will be the point ? Make a 980 with less performance and a small cooler ( ala Nano ), so end basically with the 970 mITX ?

well this is suggested retail price, so its not the market that is driving up the price.

Yes, but it is not as the price cant be adapted quickly then.

For me the price is too high for this market.. this said, i see allready peoples who dont care at all of the price, they build micro PC, and they can have a nice performance with it.

The thing is, it is a full Fiji XT core but with an aggressive thermal / power management.
For what i have see, its absolutely not a problem to get it running at 1070mhz ( simply adjust the slider in the Catalyst driver ), and you have basically the full FuryX performance.
 
Last edited:
err the oc 970 miTX gets pretty close to a stock nano ;)

I don't think this card will drop in price, only way is if demand isn't there and stock piles up.
 
err the oc 970 miTX gets pretty close to a stock nano ;)



I don't think this card will drop in price, only way is if demand isn't there and stock piles up.

And the Nano OC put again a distance with it. ( do you want we compare my 1500mhz 7970's with a 680 at stock ? )

Seriously, if wanted buy a FuryX, me who put all my gpus under h2o, and i dont plan to OC it much, ( for my Blender rendering setup ), and the price was at 450-500$, i will take the Nano, put a waterblock on it, and set it at 1050-1100mhz.
 
well if you want to that, I think the 970 with a water cooler can run circles around the nano with a water cooler. Have you seen the the 970 with water cooling it reaches 1500+ mhz.

http://www.modders-inc.com/bitspower-msi-gtx-970-full-cover-waterblock-review/5/

I dont think you need a watercooler, 1070mhz was with the fan set at default on the Nano.. thing is i just said, that ME, if the price of the Nano was lower, under the FuryX, i will take it instead of the FuryX, OC it to 1100mhz, i will get the full 4096SP, a similar gpu of the FuryX for way less.

hence why the price is fixed at 650$..

The watercooling example was just there, because i have allways use H2o thoses last 15-18 years on every system and GPU's i have use. I have never run a gpu with his stock cooling, this way the little fan on the Nano will not suffer and whatever i allways put a waterblock on them ( so pricing will not be different what i choose a 980TI, a FuryX or whatever, i will allways got in any case the 100-110$ cost in addition for my EK waterblocks )
 
Last edited:
Damien compared GM200 to Fiji at 185W: http://www.hardware.fr/articles/942-23/fiji-vs-gm200-185w.html

GM200 comes out ahead, but not by as much as you'd think. I wonder whether it's because Fiji was pushed far beyond its optimal operating frequency in the Fury X, or because AMD's DVFS is better than NVIDIA's at minimizing power and maximizing performance in tightly constrained power budgets.
That article forgot about binning. There's a wide variation in perf/watt between individual chips because of semiconductor manufacturing. AMD told us they binned the Nano for best perf/w. TitanX, on the other hand, is not just golden samples.

A better comparison would be Fury (standard) at 185 W and 980Ti at 185 W.
 
For what i have see, its absolutely not a problem to get it running at 1070mhz ( simply adjust the slider in the Catalyst driver ), and you have basically the full FuryX performance.
What does that do to fan noise though? That's a tiny cooler for a huge GPU...
 
That article forgot about binning. There's a wide variation in perf/watt between individual chips because of semiconductor manufacturing. AMD told us they binned the Nano for best perf/w. TitanX, on the other hand, is not just golden samples.

A better comparison would be Fury (standard) at 185 W and 980Ti at 185 W.

They got even better power-efficiency from a Fury X with PowerTune set to -50%: http://www.hardware.fr/articles/942-4/consommation-efficacite-energetique.html
 
What does that do to fan noise though? That's a tiny cooler for a huge GPU...

As with every air cooling, it will obviously be more noisy, this said, i was just trying to explain what can be the reason behind this price.. not much a comment about the card itself.

Bear in mind that Fury X is running 65C by virtue of the water cooling, while Nano is running ~75C. The temperature difference directly translates to power.

And that is the reason advanced too by hardware.fr...
 
What does that do to fan noise though? That's a tiny cooler for a huge GPU...
I actually tried that in my usual testing (1h heavy gaming-like load inside a case). I just pushed the power limit to the max (+50%) without overclocking but that wouldn't make any difference as I hit the power limit (~280W instead of 185W and average clock of 907 MHz instead of 849 MHz). GPU temp went up from 74°C to 79°C, fan speed up from 2400 to 2765 RPM and noise up from 33.2 dBA to 40.5 dBA which is still much lower than for the reference R9 290X even in "Quiet" mode. So it gets noisy but not so terrible as one might have feared. My concern was more about the VRM temperature but no big issue for such components according to thermal imaging (91 to 101°C). I wouldn't push the power limit if the board is inside a mini-PC lacking proper cooling though.
 
What does that do to fan noise though? That's a tiny cooler for a huge GPU...
The thermal mass on Nano is actually not too dissimilar from a blower solution for a ~300W solution by virtue of the fact that a lot of the length on those boards is taken by the blower itself. If I look at a W9100 that I happen to have on my desk, the finned portion of the thermal mass is 5.7", where the Nano is ~5.2"; the fins on the Nano have a cut out for the Axial fan, but the loss isn't that great there as the fins on the blower don't reach the full dual height anyway due to the shroud and the gap for running multiple cards. The vapour chamber element is smaller on the Nano, but that is augmented by the heatpipes.
 
The thermal mass on Nano is actually not too dissimilar from a blower solution for a ~300W solution by virtue of the fact that a lot of the length on those boards is taken by the blower itself. If I look at a W9100 that I happen to have on my desk, the finned portion of the thermal mass is 5.7", where the Nano is ~5.2"; the fins on the Nano have a cut out for the Axial fan, but the loss isn't that great there as the fins on the blower don't reach the full dual height anyway due to the shroud and the gap for running multiple cards. The vapour chamber element is smaller on the Nano, but that is augmented by the heatpipes.

For be honest, i was ask me why you have not use this design on the 290x before.. it look for me way more efficient finnaly that the standard blower used, ofc, due to the ram, the cooling point was different.. For me, looking at the noise and cooling performance of the nano cooler, we are really far far of the blower model...

Some peoples seems have think that the cooler will just be enough ( hency why all the rumor with a cut off Fiji core with less SP activated ), but on the end, this new cooler seems have got a lot of attention and seem reasonablly well thinked.

The blower model, had an interessant configuration in the past due to his capacity to really push the air through the fins at high pressure....
 
Last edited:
Guys, we could maybe made a separate thread about why some reviewers ( i know only 2 sites ) have not got a sample for review, because it completely remove all discussion about the Nano so far...

+1

Let's try to focus on Nano. If you really need to discuss how AMD picks reviewers, I'd suggest starting a new thread.
 
It seems that GCN's power consumption problems are all due to texturing.

Look at the performance of Nano in the 3DMark Vantage Texel fillrate test:

http://anandtech.com/show/9621/the-amd-radeon-r9-nano-review/14

At nominal 1GHz, Nano scores 177GT/s. At nominal 1050MHz, Fury X scores 229.9GT/s. But Nano should be scoring 219GT/s.

177GT/s has to be the result of power throttling. It implies 808MHz.

So, what we need now is the same test but with the power limit of the card increased.

Two further data-points for non-power-limited cards:

Fury has 224 texture units at 1GHz, which is 83.3% of Fury X. 83.3% of 229.9 = 191.6GT/s. Tested result is 193.1.

390X has 68.75% of the theoretical texture rate = 158.1 GT/s. Tested result is 161.5GT/s.

Obviously all of these test results are lower than the pure theoretical numbers (e.g. 268.8GT/s on Fury X), which is why some of the slower cards show slightly faster results. But Nano's result is clearly much further from theoretical than all the other cards.

By contrast, pixel fillrate doesn't show any power-throttling effect on Nano.
 
It seems that GCN's power consumption problems are all due to texturing.

Look at the performance of Nano in the 3DMark Vantage Texel fillrate test:

http://anandtech.com/show/9621/the-amd-radeon-r9-nano-review/14

At nominal 1GHz, Nano scores 177GT/s. At nominal 1050MHz, Fury X scores 229.9GT/s. But Nano should be scoring 219GT/s.

177GT/s has to be the result of power throttling. It implies 808MHz.

So, what we need now is the same test but with the power limit of the card increased.

Two further data-points for non-power-limited cards:

Fury has 224 texture units at 1GHz, which is 83.3% of Fury X. 83.3% of 229.9 = 191.6GT/s. Tested result is 193.1.

390X has 68.75% of the theoretical texture rate = 158.1 GT/s. Tested result is 161.5GT/s.

Obviously all of these test results are lower than the pure theoretical numbers (e.g. 268.8GT/s on Fury X), which is why some of the slower cards show slightly faster results. But Nano's result is clearly much further from theoretical than all the other cards.

By contrast, pixel fillrate doesn't show any power-throttling effect on Nano.

Im not surprised, this benchmark push small batch, my gpu's are whining like crazy when i run this ( even more that when i run 3Dmark2003 first scene where the fps are up to 8000fps ) it will offcourse make kick the power limiter really hard ( and so reduce clock speed ).

Its in some aspect simillar of the OCCT test or Furmark.
 
Back
Top