Tuesday, August 25th 2015

AMD Radeon R9 Nano Core Configuration Detailed

AMD's upcoming mini-ITX friendly graphics card, the Radeon R9 Nano, which boasts of a typical board power of just 175W, is not a heavily stripped-down R9 Fury X, as was expected. The card will feature the full complement of GCN compute units physically present on the "Fiji" silicon, and in terms of specifications, is better loaded than even the R9 Fury. Specifications sheet of the R9 Nano leaked to the web, revealing that the card will feature all 4,096 stream processors physically present on the chip, along with 256 TMUs, and 64 ROPs. It will feature 4 GB of memory across the chip's 4096-bit HBM interface.

In terms of clock speeds, the R9 Nano isn't too far behind the R9 Fury X on paper - its core is clocked up to 1000 MHz, with its memory ticking at 500 MHz (512 GB/s). So how does it get down to 175W typical board power, from the 275W of the R9 Fury X? It's theorized that AMD could be using an aggressive power/temperature based clock-speed throttle. The resulting performance is 5-10% higher than the Radeon R9 290X, while never breaching a power target. Korean tech blog DGLee posted pictures of an R9 Nano taken apart. Its PCB is smaller than even that of the R9 Fury X, and makes do with a slimmer 4+2 phase VRM, than the 6+2 phase VRM found on the R9 Fury X.
Sources: VideoCardz, IYD.kr
Add your own comment

101 Comments on AMD Radeon R9 Nano Core Configuration Detailed

#2
mr2009
Reviews gonna bring out the true color of this card potential. I really hope they do deliver...
Posted on Reply
#3
TheGuruStud
If power throttling can be disabled (or limit significantly increased), then this will be a great card.

Otherwise, it's a dog turd like Fury X.
Posted on Reply
#4
btarunr
Editor & Senior Moderator
TheGuruStud
If power throttling can be disabled (or limit significantly increased), then this will be a great card.

Otherwise, it's a dog turd like Fury X.
I don't think with that VRM, its throttling can be relaxed enough to match Fury X performance.
Posted on Reply
#5
the54thvoid
TheGuruStud
If power throttling can be disabled (or limit significantly increased), then this will be a great card.

Otherwise, it's a dog turd like Fury X.
I think the reduction in core VRM from 6 to 4 (33% down) will mean any power limit bypassing will have to use little extra voltage. FWIW, Fiji isn't a turd, it was just a marketing mess.
Posted on Reply
#6
Chaitanya
interested in reading the review.
Posted on Reply
#7
geon2k2
Well, since it has the full core, this is gonna be expensive, so it is mostly a small form factor enthusiast card, not a mid-range mass market card :(
Posted on Reply
#8
john_
Up to 1000MHz? AND 4096 SPs? WTF? Is this at 20nm? I was expecting the GPU to be close to 800MHz.
Posted on Reply
#9
MxPhenom 216
ASIC Engineer
john_
Up to 1000MHz? AND 4096 SPs? WTF? Is this at 20nm? I was expecting the GPU to be close to 800MHz.
28nm. Why is it so hard to believe? 28nm is a very mature process. We have been running high SP and highly clocked cards for a while now.
Posted on Reply
#10
HumanSmoke
john_
Up to 1000MHz? AND 4096 SPs? WTF? Is this at 20nm? I was expecting the GPU to be close to 800MHz.
I suspect the " up to" might be a significant factor in most situations.

AMD are clearly marketing the card for SFF applications, but I doubt too many reviews - if any, are going to use a SFF test environment (maybe an open air version at best).
Posted on Reply
#11
RejZoR
It's the clock increase that can require voltage increase beyond reasonable numbers.

For example, my HD7950, I could overclock it from stock 900MHz to 1000MHz without any voltage change (stock 1075mV I think). For 1100MHz I needed around 1.2V, for 1200MHz it was at 1.38V even though I've made the same 100MHz increase as before.

Anyway, the point I'm making is that voltage isn't receiving linear increase with same clock increases. Once you get to a certain point, voltages have to skyrocket in order to maintain linear clocks increase.

Maybe 1050MHz is already that point for Fiji (used on Fury X), that's why it needs water cooling and we know how power usage spiked on overclocks. But Nano is clocked 50MHz lower and that gives it all the headroom to lower the voltages. It just won't be a mad overclocker due to power circuitry limitations.
Posted on Reply
#12
15th Warlock
I wonder what this baby would do if put under water, I mean, it packs quite a punch, too bad they cut the number of VRMs, but if somehow people would find a way to overvolt this little card, in theory it could be pushed beyond the performance of a vanilla Fury, right?
Posted on Reply
#13
nem
Nano... *o*
Posted on Reply
#14
Xajel
Well done AMD, the first HBM is very promising for those who want to upgrade this time...

but for those who have other plans ( 2016 plan ), then waiting for HBM2 will be interesting, not because AMD will be able to have 8GB and more of VRAM... but we will be able to see both AMD and NV offerings also...

I'll be very interested in Nano like products, Nano 2 from AMD or what ever NV will come with will be interesting for my next build which should be a miniITX build... hell we might get AMD Zen also in the mix to see how it will do against Skylake or Kaby Lake as we're looking at 2H16 timeframe
Posted on Reply
#15
john_
MxPhenom 216
28nm. Why is it so hard to believe? 28nm is a very mature process. We have been running high SP and highly clocked cards for a while now.
Yes I know. I just wasn't expecting reading 1000MHz.
HumanSmoke
I suspect the " up to" might be a significant factor in most situations.

AMD are clearly marketing the card for SFF applications, but I doubt too many reviews - if any, are going to use a SFF test environment (maybe an open air version at best).
Yes, that performance slide do say in every way possible that, that 1000MHz, is just to make the average john_ to post "wow!". Well, AMD is shooting it's foot in marketing, every time they come out with a new product and probably that "Up to 1000Mhz" could be a nice excuse for many tech sites to post something negative in their reviews, like "We only managed to see those 1000MHz for 5 seconds in that game's intro, while testing the system with an open case, in north pole". If this does happen, then I would say that someone should fire the person who doesn't fire the marketing department. Then fire the marketing department too.
RejZoR
Maybe 1050MHz is already that point for Fiji (used on Fury X), that's why it needs water cooling and we know how power usage spiked on overclocks.
The first time I looked at a slide pointing at Nano's efficiency compared to Fury X's, I assumed that Fury X cards are in fact all factory overclocked. And why not. AMD is doing it with 200 and 300 series cards. Why not with Fiji?
Posted on Reply
#16
tabascosauz
the54thvoid
I think the reduction in core VRM from 6 to 4 (33% down) will mean any power limit bypassing will have to use little extra voltage. FWIW, Fiji isn't a turd, it was just a marketing mess.
AMD usually isn't too bad in this VRM department (after all, the GTX 480s were blowing up, as well as certain 780 Tis); plus, I feel like bulk of the aggressive power targets is going to affect the core.
geon2k2
Well, since it has the full core, this is gonna be expensive, so it is mostly a small factor enthusiast card, not a mid-range mass market card :(
Never marketed as, never meant to be, and never will be sold as a mass market card. Need a GTX 970 competitor without killing your wallet? Go buy a 390.
HumanSmoke
I suspect the " up to" might be a significant factor in most situations.

AMD are clearly marketing the card for SFF applications, but I doubt too many reviews - if any, are going to use a SFF test environment (maybe an open air version at best).
"Up to" is AMD's favourite term. We witnessed this with the HD 7xxx, R9 2xx, R9 3xx and now the Fiji family. Always "up to xxx MHz". Only when the card is released will we be able to see for ourselves what the stock base clock is like; however, I have a feeling that the R9 Nano is going to bring something to the table that we haven't seen before in this regard, due to its TDP requirements.
RejZoR
It's the clock increase that can require voltage increase beyond reasonable numbers.

For example, my HD7950, I could overclock it from stock 900MHz to 1000MHz without any voltage change (stock 1075mV I think). For 1100MHz I needed around 1.2V, for 1200MHz it was at 1.38V even though I've made the same 100MHz increase as before.

Anyway, the point I'm making is that voltage isn't receiving linear increase with same clock increases. Once you get to a certain point, voltages have to skyrocket in order to maintain linear clocks increase.

Maybe 1050MHz is already that point for Fiji (used on Fury X), that's why it needs water cooling and we know how power usage spiked on overclocks. But Nano is clocked 50MHz lower and that gives it all the headroom to lower the voltages. It just won't be a mad overclocker due to power circuitry limitations.
This. It's why I hope that the Nano's VRM is in line with AMD's tradition of good VRMs, yet is engineered so that users cannot take advantage of the binned Fiji XT's voltage characteristics to push it past a Fury X when equipped with a FC Block. It's a niche card, and isn't meant to be selling like hot cakes, despite the fact that it probably will due to the sheer hype and anticipation from mostly uninformed users. The Fury X and Fury are what AMD is trying to steer us towards; overloading AMD's supply chain for the Nano isn't going to do them any good. I can see it now: Nano out of stock @ NCIX for the next 8 months, while Fury X is out of stock because of lack of demand, while R9 Fury is just sitting on the shelves, waiting on something that'll never come.

@john_ Firing the marketing team isn't going to change the mentality at AMD, even though the bad marketing is at the centre of their problems. They've made countless mistakes over the past 4 years alone, and now it's as if every part of AMD is adopting an unmotivated, slacking, waiting-for-the-end attitude.
Posted on Reply
#17
Ebo
If these specs holds out, and its true then I wish for W1zzard to test it, and come up with a way to work arround the voltage, since the card only have 175 watt, that would be awsome:clap:, just like allways when he does his magic:laugh::peace::respect:.

I have a few concirns about HBM1 since it cant go above 4GB, so Im waiting for HBM2 instead, and I dont feel like any of the Fury series will bring a lot to the table over my R9 290 from Sapphire, since my card gives even the new series a run for its money especially if I turn up the juice on it.

But the Nano might do it fore me;), so I will upgrade, especially under water and turn up the juice, I have the itch for new hardware:eek:. I really dont need any new hardware, but the feeling in the right fingertip is becomming close to hit the buy button:toast::nutkick:.
Posted on Reply
#18
Folterknecht
tabascosauz
Never marketed as, never meant to be, and never will be sold as a mass market card. Need a GTX 970 competitor without killing your wallet? Go buy a 390.
And that is exactly AMDs problem. They need a card that can really compete with the GTX970, because 390(X) sucks balls at that price point. Same performance and price (in case of 390X totally overpriced) as the 970, which you can OC easy to 390X levels. I wont pay the same money for a card that doesnt ofter more performance but draws 100-150W more with all the related issues.

And please dont start with the 8GB nonsense, its a marketing gag (GT620 4GB much). For single GPUs setups it's useless in most cases - Hawai doesnt have the horsepower to make use of it. Hawai (290/390) and 970/980 are 1080/1440p cards.


I was really hoping for a 970 alternative to replace my GTX570.
Posted on Reply
#20
Assimilator
"Up to 1000MHz" = you'll see 1000MHz for a couple of milliseconds at best. Most of the time it will run at a far lower core clock.

Overclocking on this card is going to be nonexistent because it will probably catch fire, but that's okay because Fury/Fiji overclocks like s**t anyway, so why would you even bother.
Posted on Reply
#21
Sony Xperia S
Assimilator
"Up to 1000MHz" = you'll see 1000MHz for a couple of milliseconds at best. Most of the time it will run at a far lower core clock.

Overclocking on this card is going to be nonexistent because it will probably catch fire, but that's okay because Fury/Fiji overclocks like s**t anyway, so why would you even bother.
Just enjoy it the way it comes in default. It is not supposed to overclock and noone gives you such options. :D
Posted on Reply
#22
buildzoid
So 4 phase VRM... Nothing an E-power can't fix.
Posted on Reply
#23
uuuaaaaaa
I hope this has cherry picked chips!
Posted on Reply
#24
commando55555
uuuaaaaaa
I hope this has cherry picked chips!
what would be the point doesn't oc for shit any way.
Posted on Reply
#25
Sony Xperia S
commando55555
what would be the point doesn't oc for shit any way.
When will you understand that the card itself will be damn fast even without overclocking and with overlock you will achieve nothing but nonsense?
Posted on Reply
Add your own comment