Fury Nano, biggest bang in smallest package.. FPS/inch baby!

[William Gaatjes]

I will also wait for now until the nano arrives. Then i will make my decision what gfx card to buy. I do not know if the interposer tech is a new technique or a proven one. I know it is just a die without any logic circuits in it. Only metal layer routing and via's. But i am curious if it has been used before and if they are not looking for a cheaper technique. Seems to me that the interposer is quite expensive because of the sheer size. Or could they be using a really old photolithography technique to lower the cost for the interposer as much as possible ? There is a 4096 bits wide databus, probably single ended i/o, address lines single ended i/o, control lines. The pcie x 16 bus. And a lot of power connections.. Some more control lines and hdmi/ display port signals.

EDIT:
I do wonder how much those hbm dram chips cost a piece.

 

[Stuka87]

I am really hoping the Nano is close to 390X performance, but in a much smaller package. My case can only fit a 10.5" long card, so the full length 12" Fury cards wont fit. But the Nano would be absolutely perfect for it!

Although I hope that Sapphire or the like makes one that is around 10" long with the heatsink overhanging the end like on the normal fury but with two fans instead of three. That way the fans could run at very slow speeds making it very quiet.

 

[NTMBK]

Seems to me that the interposer is quite expensive because of the sheer size. Or could they be using a really old photolithography technique to lower the cost for the interposer as much as possible ?

According to S|A, the interposer is made on a 65nm process: http://semiaccurate.com/2015/06/22/amd-talks-fiji-fiji-x-odd-bits-tech/ So yeah, pretty old!

 

[tential]

Fury is in 980 class, so, having Nano in the same class makes no sense at all.
It is either going to compete against the 970 (though, the 390x does that as well), or, it is going to be a premium product with a premium price tag.

Good to see some more sensible posters come in (including others after yourself). This whole discussion was getting utterly ridiculous.

A card for $450, that is faster than a 290x significantly (Which makes it faster than a GTX 980. Remember please the 290x and the GTX 980 aren't miles apart or anything....), so therefore faster than a GTX 980 but at $450? What happens to Fury? It's FASTER than a GTX 980. So is Fury. And not by much. So how far is the gap between Fury Nano and Fury? Not much if the Fury Nano is faster than a GTX 980 or even JUST as fast. So now you're telling me a $450 card (some who speculated even $400 lol), is lets even say 5% slower than a GTX 980 would be priced at $450? This card will DESTROY it's own cards sales then.

Ugh, I don't even feel like this needs an explanation. It does NOT make sense at $400-450. Or the card is significantly slower. Please lets move on.

 

[Silverforce11]

Rumored to be using full-dies, low clocks, low vcore, so I don't think it would be cheap, it's not meant to be the "cheap option", that's Fury.

I hope I'm wrong and they slot it in at $499, a tiny bit slower than Fury, still faster than 980.

But who knows how popular mITX builds are? Maybe AMD has some market research we don't and they see a potential for a premium mITX GPU.

 

[monstercameron]

Good to see some more sensible posters come in (including others after yourself). This whole discussion was getting utterly ridiculous.

A card for $450, that is faster than a 290x significantly (Which makes it faster than a GTX 980. Remember please the 290x and the GTX 980 aren't miles apart or anything....), so therefore faster than a GTX 980 but at $450? What happens to Fury? It's FASTER than a GTX 980. So is Fury. And not by much. So how far is the gap between Fury Nano and Fury? Not much if the Fury Nano is faster than a GTX 980 or even JUST as fast. So now you're telling me a $450 card (some who speculated even $400 lol), is lets even say 5% slower than a GTX 980 would be priced at $450? This card will DESTROY it's own cards sales then.

Ugh, I don't even feel like this needs an explanation. It does NOT make sense at $400-450. Or the card is significantly slower. Please lets move on.

even if the fury and r9 nano are the same price they target different market segments. The r9 is limited to 175W max tdp and the fury is pretty much unlocked. The fury might be for overclocking while the r9 for sff, low power, htpc rigs.

 

[Elixer]

You know, looking at the fan on the nano, it looks to be the same OEM that sapphire is using on their cards.
I just wish AMD would let the OEMs have at at.
By that, I mean let them do everything custom if they want, like they did with the Fury.
AMD can't afford to blunder on the cooler for the Nano, and if it isn't whisper quiet, then, that just means more lost sales, something they don't need.

With only 1 8pin connection, we pretty much know the max power it could use, and thus calculate the max performance capable by looking at the Fury...
So, the only unknown in this whole thing is the price, and if AMD will allow the OEMs to customize or not. (Like Asus did with the DVI-D connector)

 

[Kenmitch]

Depending on price vs performance the Nano might be a hard card to get. Could see Dell, HP, and others buying them by truckloads.

 

[tential]

You know, looking at the fan on the nano, it looks to be the same OEM that sapphire is using on their cards.
I just wish AMD would let the OEMs have at at.
By that, I mean let them do everything custom if they want, like they did with the Fury.
AMD can't afford to blunder on the cooler for the Nano, and if it isn't whisper quiet, then, that just means more lost sales, something they don't need.

With only 1 8pin connection, we pretty much know the max power it could use, and thus calculate the max performance capable by looking at the Fury...
So, the only unknown in this whole thing is the price, and if AMD will allow the OEMs to customize or not. (Like Asus did with the DVI-D connector)

Huh? Then what is the max performance possible if we know?

 

[shady28]

Huh? Then what is the max performance possible if we know?

Tom's did an analysis by power limiting the Fury X.

I wouldn't call it max performance possible but it's probably a very good estimate.

The avg isn't hit too bad but the minimum tanked.

Even though the power consumption decreases from 267W to 170W when the power limit is set to -50 percent, the resulting frame rates just aren’t in the playable range any more. Higher power limit settings do not show imported results, either. The minimum and average frames per second, as well as the power consumption, stay the same. Things actually look about the same at Full HD, but the power consumption is low enough in that scenario that there’s no point to changing the settings anyway.

 

[Stuka87]

Tom's did an analysis by power limiting the Fury X.

I wouldn't call it max performance possible but it's probably a very good estimate.

The avg isn't hit too bad but the minimum tanked.

Not sure we can take that test as being the same as a card that is designed to run at that TDP. Performance will be lower, AMD stated 290X like performance when they first talked about it. A fully unlocked, highly binned chip that is running at say 800MHz isnt the same as a fully unlocked, chip that wants to run at 1050MHz, but cant because its being artificially limited.

 

[IEC]

Huh? Then what is the max performance possible if we know?

Speculation is that it is using chips binned for power efficiency/lower voltage and that it will come clocked in the ~890 MHz range (vs 1050 MHz for Fury X). If true, you could expect based on clockspeed changes alone to get ~85% of Fury X performance provided it is a fully-enabled Fiji chip. Perhaps more if HBM speed isn't changed.

It would track well with what those of us who undervolt/underclock Radeon 290/X in the summertime have observed: you can get up to 85%+ of the stock performance for a 30%+ reduction in power consumption. My personal sweet spot seems to be to run the core at 880 MHz, which doesn't seem that far off from the speculated core clock speed of the Nano.

That said, unless you need the extra performance of the Nano it doesn't seem likely it'll offer a better value than a 290/X...

 

[Elixer]

Huh? Then what is the max performance possible if we know?

We know this:
http://forums.anandtech.com/showpost.php?p=37489068&postcount=63

BTW, I think that slide is telling us something else, since, a 8pin PEG offers 150W, and they have 75W from the interface, so that means, 225W.
They are showing 175W, which is 50W lower than the most they could use.
Does that mean that the Nano is capped at 175W?

Peak: Metro: Last Light at 1920x1080, representing a typical gaming power draw. Highest single reading during the test.

shows us this:

and to put it in perspective, this is what the Fury is getting at that resolution

So, this Fury uses a 1000 MHz clock, using 226W getting 97.7FPS to get you a rough idea where the Nano will sit.

The performance will be less than the Fury pretty much across the board, except for some specific tests, where it could match it.
That leaves 290x/390x, and how it will do against those, and in most cases, the Nano will match or be slightly faster than those, using a much lower power envelope, and also brings GCN 1.2 as well.

At resolutions higher than 1920x1080, the nano will carve out a small little space for itself.

The big gamble here is, how much is AMD willing to gamble that people will pay a premium because of the size/power envelope it has compared to other products at this price point.

Since TMSC is now giving discounts for 28nm, to get orders up, one has to wonder if this will play a role in AMD's pricing.

 

[Silverforce11]

@Elixer
Good points but TDP is greatly affected by vcore.

You can gain massive perf/w by undervolting R290X for example and you lose ~10% performance but gain 30-40% efficiency. No joke. I've done it myself.

It's quite easy with the CC power limit, set it to -50% and watch as GCN downclocks/downvolt, an R290/X will run at around 850-880mhz instead of 947/1000mhz and power usage drops from ~240W to ~160W. For me, mining power load drop from 275-300W to ~200W.

GCN can be very efficient, the lower OC headroom suggests AMD clocks it pretty high out of the gate.

I suspect Fury GCN will behave similarly, so if they take Fury X, drop clocks to ~850mhz, downvolt it, set a lower TDP limit in the bios (like Asus's Fury), it will definitely drop down to <175W while losing about 10-15% performance tops.

 

[Elixer]

While that is true about VCore, the problem with that is, unless AMD is really cranking out stellar yields, to be able to bin enough chips to achieve a low VCore, while still keeping the clocks at acceptable levels, and have enough supply to sell, it don't seem viable at this point.

They have major supply issues, and nobody has been able to nail down where the issue is. Is it they are selling all available units? Is it issues at UMC, or another plant or what.

 

[Silverforce11]

Yields or binning issues aside, technically its very possible given how GCN behaves with downclock/undervolt.

You can take an R290X, -50% power limit, it drops from ~240W to ~160W in games and lose about 10% performance. What kind of efficiency level is that? Well, ~Maxwell 2 efficiency.

The reverse is also true, if you OC it, power usage goes nuts really quickly with +vcore, so I didn't like pushing OC, but run it at the highest stock vcore allows.

 

[Shehriazad]

So uhm...when AMD/Lisa Su talked about the Nano at first (which was at the Fury X presentation day I think, right?).


Didn't they say the Nano was gonna deliver performance beyond that of the 290X while having this tiny size and low-ish TDP?
I mean correct me if I'm wrong and this statement doesn't exist...but shouldn't we expect this card to be above 290X/390X by a bit?

For some reason everyone is now like "It's gonna be just as fast as 290X"...and that cherry picked benchmark we have (the only one so far) also indicates that it's above the 290X.

So where exactly is the talk of same performance coming from?

25% faster in best case scenarios and probably like 10-15% faster on average seems like a good base to work with.

 

[Silverforce11]

So uhm...when AMD/Lisa Su talked about the Nano at first (which was at the Fury X presentation day I think, right?).

Didn't they say the Nano was gonna deliver performance beyond that of the 290X while having this tiny size and low-ish TDP?

2x perf/w and significantly higher performance than R290X was her words.

But then one of them also said "Fury X will be an overclocker's dream" and that was a damn lie. Without vcore mod released by AMD (seriously fail) on launch, its an overclocker's nightmare.

 

[PPB]

@Elixer
Good points but TDP is greatly affected by vcore.

You can gain massive perf/w by undervolting R290X for example and you lose ~10% performance but gain 30-40% efficiency. No joke. I've done it myself.

It's quite easy with the CC power limit, set it to -50% and watch as GCN downclocks/downvolt, an R290/X will run at around 850-880mhz instead of 947/1000mhz and power usage drops from ~240W to ~160W. For me, mining power load drop from 275-300W to ~200W.

GCN can be very efficient, the lower OC headroom suggests AMD clocks it pretty high out of the gate.

I suspect Fury GCN will behave similarly, so if they take Fury X, drop clocks to ~850mhz, downvolt it, set a lower TDP limit in the bios (like Asus's Fury), it will definitely drop down to <175W while losing about 10-15% performance tops.

How dare you talk about undervolting and underclocking GCN! You cant do that acording to the super well informed posters in here!

Sarcasm aside, anyone with a minimum knowledge that bought a GCN product with a dual BIOS, the first thing they did was to mod the BIOS with some famous tool and change the clock/voltage/pstates to their best interests. At least in my case the very first thing I did was disable that awful boost on my 7950

 

[Headfoot]

Anyone who mined on 290s knows how far down you could push power consumption without losing much performance. Consequently, the only time gpu power consumption actually mattered was when calculating mining profitability

Of course the power consumption "debate" had nothing to do with people actually caring about power consumption and everything to do with the fact that nVidia happened to be winning in that metric, thus it was important.

 

[Kallogan]

there big chances the gpu will be undervolted quite agressively to fit the TDP, not much of an overclocker. But i couldn't care less, stock perf will be nice and one fan isnt fit for OCing anyway.

 

[Elixer]

2x perf/w and significantly higher performance than R290X was her words.

Yeah, but we don't know what metric she was talking about, she could have just meant bandwidth.

But then one of them also said "Fury X will be an overclocker's dream" and that was a damn lie. Without vcore mod released by AMD (seriously fail) on launch, its an overclocker's nightmare.

What is worse, the person who said that was the chip designer himself...
Maybe it will be enabled in a future driver, but, it leaves a very bad impression when he says that and nothing on launch.

The whole product stack seems like is was rushed for all the wrong reasons.
First, we got Fury X, while a good product, suffered from mechanical issues (pump), very low supply(?), and wasn't really the killer card people hoped it would be.
Then we got the Fury, which again, has very low supply(?), but, released the OEMs to have perhaps one of the best cooling systems out there, and we can see that the 290x/390x is nipping at the Fury's heels.

Now, we are waiting for the Nano, which is supposed to use binned chips, which either means they have been binning them since Fury X, or they got fantastic yields, or, we will yet again see very low supply.

Why does this matter?
Well, August is the start of back to school, and that is when the big OEMs want the new stuff, so they can sell them, and you can bet that they want thousands of cards (which might explain the low supply issues), however, if this is not the case, then, AMD has yet again missed the boat.

Look on launch day of Windows 10, and see what the big OEMs are offering, and if you see lots of machines with Fury & Nano products, that will answer if they caused the supply issue.
Nano would seem to be the perfect card for the back to school crowd.

 

[shady28]

Not sure we can take that test as being the same as a card that is designed to run at that TDP. Performance will be lower, AMD stated 290X like performance when they first talked about it. A fully unlocked, highly binned chip that is running at say 800MHz isnt the same as a fully unlocked, chip that wants to run at 1050MHz, but cant because its being artificially limited.

But it's better than people guessing.

Those benchmarks put the downclocked Fury X with a substantially lower minimum fps than a 290X in Thief at 4k, and about the same avg fps as a 980 Ti. Looks like it will be a stutterfest at 4k.


My guess is this will be a good high-performance 1080p/1440p card though, since they also say it didn't make much difference there.

And If this card gives 980+ performance at 1080p/1440p though, I think that would make it a winner for most people given the slow adoption rate of 4k on PCs.

vs

 

[thilanliyan]

@Elixer
Good points but TDP is greatly affected by vcore.

You can gain massive perf/w by undervolting R290X for example and you lose ~10% performance but gain 30-40% efficiency. No joke. I've done it myself.

Actually, sometimes you won't even lose performance. I was able to undervolt my 7950s by 150-200mV all while keeping stock clocks.

 

[Headfoot]

Actually, sometimes you won't even lose performance. I was able to undervolt my 7950s by 150-200mV all while keeping stock clocks.

You won't get as much power savings that way, but definitely true. I can undervolt my 290 -75mV @ 1000mhz