Radeon R9 Nano thread

[Nagorak]

I know a number of us are interested in seeing how the Radeon Nano performs, and it occurred to me that with the help of someone with a Fury X, we might be able to predict how the Nano will perform.

Current rumors suggest that the Nano will be a fully unlocked Fiji GPU, but possibly be clocked around 800 MHz or 850 MHz. Well, we already have fully unlocked Fiji chips in the Fury X. So, if it's possible to reduce the core clock of the Fury to 800 MHz, that should approximate the performance of the Nano. I know that Catalyst Control Center lets me down clock my HD 7870 down to as low as 300 MHz, so you'd think it would be possible on the Fury X as well. We can assume that the memory will remain 4 GB and set to 500 MHz.

Is anyone with a Fury X willing to try to help out with this?

 

[Hapatingjaky]

AMD already stated that the Nano would perform at roughly the 290x performance at a lower power usage and much less heat. Which is of course sweet.

 

[SSXeon]

AMD already stated that the Nano would perform at roughly the 290x performance at a lower power usage and much less heat. Which is of course sweet.

Where is that statement?

 

[@md_Guy]

AMD already stated that the Nano would perform at roughly the 290x performance at a lower power usage and much less heat. Which is of course sweet.

IIRC, the only "official" statement was that Nano was nearly 2x perf /watt of the 290X. This could mean either 2x the perf for the same wattage OR same performance for half the wattage. I myself am leaning towards the middle, about 20% more perf for like 60% of the 290X power costs... all while being about 7.5" long and perfect for mATX setups.

 

[razor1]

Well if the card pulls 225 watts max and we look at what Fury X does, just drop its performance by 20%, thats a max power draw for the card,

Frequency is supposed to be at 800 mhz (leaked rumors) that too is 20% less (we know frequency has a 1:1 relationship to wattage). So 20% less than the Fury X seems likely.

 

[SSXeon]

Well if the card pulls 225 watts max and we look at what Fury X does, just drop its performance by 20%, thats a max power draw for the card,

Frequency is supposed to be at 800 mhz (leaked rumors) that too is 20% less (we know frequency has a 1:1 relationship to wattage). So 20% less than the Fury X seems likely.

From the slides i saw its 175w, sure it will be clocked lower but also have lower voltage. And i dunno about a strait up 20% slower performance. The Fury has ~14% less shading power but only shows 7-8% slower than Fury X. If we go by that, and if in fact the nano retains the 500Mhz HBM, worst case it will be 12% slower.

 

[razor1]

From the slides i saw its 175w, sure it will be clocked lower but also have lower voltage. And i dunno about a strait up 20% slower performance. The Fury has ~14% less shading power but only shows 7-8% slower than Fury X. If we go by that, and if in fact the nano retains the 500Mhz HBM, worst case it will be 12% slower.

*snip*

I stated max power usage, and this chip is a full fiji chip not just cut down some parts, bottlenecks won't shift as much when we are looking at the same chip, actually games right now aren't fully utilizing the Fury X shader arrays at lower resolutions due to a geometry bottleneck so the shader bottleneck is actually not the real concern here, there might be more then 20% drop due to that on lower resolutions (lower then 4k)

If the TBP comes out to 175, I expect its performance to be somewhere between the gtx 970 and gtx 980. Its pretty obvious it won't match the gtx 980 with performance per watt, and the 980 pulls around 175+ I think.

Actually the gtx 980 is more like 150 watts, so at 175 it will remain somewhere between the gtx 970 and gtx 980.


http://www.techpowerup.com/reviews/ASUS/R9_Fury_Strix/31.html

there is a 24% difference in performance (I know its only one review but just easier to do this way cause its already tabulated) from Fury X to gtx 980, expect more then a 20% drop in performance at 4k, and even more in lower resolutions.

 

[BababooeyHTJ]

That looks really slick.

 

[IamHere]

Think about how small a form factor it could have as a dual chip card. Should still be able to air cool but have amazing cost/performance numbers.

 

[noko]

Well if the card pulls 225 watts max and we look at what Fury X does, just drop its performance by 20%, thats a max power draw for the card,

Frequency is supposed to be at 800 mhz (leaked rumors) that too is 20% less (we know frequency has a 1:1 relationship to wattage). So 20% less than the Fury X seems likely.

Razor1, there is another factor which affects power and that is voltage. With the lower frequency voltage no doubt can be lower. Combine the two it will be more than 20%. Voltage is exponentially and not linear, a 20% drop in voltage will give more than 20% drop in power, up to 44%. Memory speed could also be reduced even if HBM is very efficient there maybe no need for 500mhz for it, saving an additional few watts.

Plus Fiji may just be way more efficient at lower speeds so power and performance we will need to see once released. It could well match the 980 in general in performance winning some and loosing some (gameworks crap stuff).

 

[razor1]

Razor1, there is another factor which affects power and that is voltage. With the lower frequency voltage no doubt can be lower. Combine the two it will be more than 20%. Voltage is exponentially and not linear, a 20% drop in voltage will give more than 20% drop in power, up to 44%. Memory speed could also be reduced even if HBM is very efficient there maybe no need for 500mhz for it, saving an additional few watts.

Plus Fiji may just be way more efficient at lower speeds so power and performance we will need to see once released. It could well match the 980 in general in performance winning some and loosing some (gameworks crap stuff).

Yeah that's true, but I don't think its going to beat the gtx 980 in performance per watt from what we have already seen from Fury X, and Fury.

its either going to perform more then the gtx 980 while using more power or use less power and perform less.

If they hit 175 watts, its pretty much guaranteed its won't match the gtx 980 in performance. I would be very surprised if they were able to achieve the gtx 980 performance with that low of power usage.

 

[pax]

How much would the one 8 pin connector affect ocing. I mean if cooling isnt a problem if you slap on an EK wc...

 

[noko]

Yeah that's true, but I don't think its going to beat the gtx 980 in performance per watt from what we have already seen from Fury X, and Fury.

its either going to perform more then the gtx 980 while using more power or use less power and perform less.

If they hit 175 watts, its pretty much guaranteed its won't match the gtx 980 in performance. I would be very surprised if they were able to achieve the gtx 980 performance with that low of power usage.

It will be interesting how this plays out, from current data it would be hard to beat the 980 perf/watt. Let see what happens.

Now how will the Nano compete with the R390x? Price? or the R390? Why would someone pick up a 390x if the performance is about the same? or the price. I am wondering how in the world AMD is going to price the Nano? Any guesses?

My guess if it beats the 390x it will be $499. Now if it competes against the 390 and 390x but is slower - $449 due to form factor and coolness but also it has 4gb less memory. So the more performance orientated would go with the 390x while the smaller case size - Nano.

As for OCing - cooling would be the #1 issue due to size. Water block alleviates that but then power from one less pcie connection to the actual power circuitry onboard combined with what the bios will allow for voltage changes (if that ever comes about).

 

[noko]

Then again maybe AMD will price this like an Apple product at $599 and call it a day.

 

[Nagorak]

Well, it just seems to me that if someone gets their hands on a Fury X they would be able to test this directly. For example, how much less power does the X use when running a 800 MHz (how much can you lower core voltage and still have it stable). How much does performance drop off, when keeping memory at stock 500 MHz?

Obviously we can't account for every single variable, such as the cooler, but we could at least see how fast a Fury clocked ~25% lower still is.

That being said there are so few people who actually have Fury X's and most of them likely have no interest in the Nano, so finding someone willing to test this stuff is probably hard.

 

[Gandalfthewhite]

without voltage control it would be hard to verify.

 

[Nagorak]

without voltage control it would be hard to verify.

That's right, I forgot there is no voltage control yet.

 

[LordHawkwind]

As I've stated a few times I think the Nano is supposed to go head to head with 970 and beat it in most cases and even come close to a 980. It's clearly intended for competing with the SFF 970 so to do that they have to price it right IMHO.

TBH AMD's pricing strategy is a mess with the Fury X, fury and 390x marketed for 4k gaming with significant price differentials that don't translate into their performance. I also think when the nano arrives the 390's will be EOL.

 

[noko]

From the slides i saw its 175w, sure it will be clocked lower but also have lower voltage. And i dunno about a strait up 20% slower performance. The Fury has ~14% less shading power but only shows 7-8% slower than Fury X. If we go by that, and if in fact the nano retains the 500Mhz HBM, worst case it will be 12% slower.

Using AMD numbers:

The Nano will be up to 2 x 175wNano/250wR290x = 1.4 times the performance of a R290x Information does not say what is the basis for the performance/watt comparison (I was guessing R290) and what performance it is being tested and compared to. But it does give a reference to the Rage Fury X

So new formula would be:
2/1.5 x 175w/275w = .85 meaning the performance of the Nano is 85% of the R9 Rage fury X

 

[SSXeon]

Using AMD numbers:

The Nano will be up to 2 x 175wNano/250wR290x = 1.4 times the performance of a R290x Information does not say what is the basis for the performance/watt comparison (I was guessing R290) and what performance it is being tested and compared to. But it does give a reference to the Rage Fury X

So new formula would be:
2/1.5 x 175w/275w = .85 meaning the performance of the Nano is 85% of the R9 Rage fury X

http://wccftech.com/amd-confirms-radeon-r9-nano-launching-august/

Also noko i want to add, as with the Fury(1040Mhz) with ~15% less shader power, but shows ~7-8% less performance. Nano @ 800Mhz it will have about 31% less shading performance, so 14-16% lower? If it is at a $450 price point i will be jumping on that! I need a ~$400-450 videocard for fallout 4.