AMD Radeon R9 300 Series MegaThread: FAQ and Resources

[de5_Roy]

AMD Radeon R9 Nano Reportedly Launching on 27th August - Fiji Based Compact-ITX Card Arriving Next Week
http://wccftech.com/amd-radeon-r9-nano-delayed-september-wip/
AMD's next-gen GPUs will feature between 8-16GB of super-fast HBM2
http://www.tweaktown.com/news/47165/amds-next-gen-gpus-feature-between-8-16gb-super-fast-hbm2/index.html
AMD readies three new GPUs: Greenland, Baffin and Ellesmere
http://www.kitguru.net/components/graphic-cards/anton-shilov/amd-readies-three-new-gpus-for-2016-greenland-baffin-and-ellesmere/

 

[cdrkf]

A little article that (supposedly) gives the code names for the Artic Islands range due next year:
http://www.kitguru.net/components/graphic-cards/anton-shilov/amd-readies-three-new-gpus-for-2016-greenland-baffin-and-ellesmere/

This is all from 'sources close to AMD' so take it as you will.

 

[nikoli707]

A little article that (supposedly) gives the code names for the Artic Islands range due next year:
http://www.kitguru.net/components/graphic-cards/anton-shilov/amd-readies-three-new-gpus-for-2016-greenland-baffin-and-ellesmere/

This is all from 'sources close to AMD' so take it as you will.

this does make sense since pascal is also supposedly on the same timeline.

 

[fudgecakes99]

I doubt it, marketing wise it's already a niche market ontop of a niche market if you want high end parts. Like iphones, add a few features call it new.

 

[Eggz]

Why they would alienate their customer when they bring big improvement faster?

Because advancement usually comes at the top-end of product lines, which sit at the $1,000 price point. If you make those products obsolete too fast, people will lose (what little) incentive there is to buy at that price point. No one wants the only difference between their $1,000 product and a worthless brick to be a short time span. Sure, everything eventually becomes obsolete, but obsolescence shouldn't come too quickly for premium customers.

 

[renz496]

They will not be too quick like once a year. In general nvidia will come up with new architecture for every two years. Those that buy 1k gpu actually can be quite crazy. There were few of them will change to new gpu in a blink of an eye when new faster gpu comes out.

 

[nikoli707]

maxwell released in feb 14, we got about 6 months until two years are up. and their current cycle is about 2.5 years, so next year this time we will see pascal.... if everything keeps going to nvidias plan. otherwise will it be a 1080ti or a flagship model, probably not. but every single chip manufacture is now switching to the 14/16nm process and ditching everything else, so theoretically if all of them are on board then yield improvements should come quickly for all. the smartphone/smart anything demand is now so damn high that they want to put an iphone 7/samsung s7 cpu/gpu into every car dash, tv, grocery store cart, etc. the future is coming much sooner than many think. otherwise i think amds rearrange caused them to ditch this gen and just release what they got unrefined and it was very very smart. put all your efforts into zen and the next gpu generation, they knew they would never catch intel or nvidia in a single generation. also its a great time to do that. either you need 1080p gaming, which any cheap skate r9-290 can do already, or you want 4k which no card can do at all anyways. the hdtv market forced them a great hand and early released hbm will benefit them well in the longer run, even if it makes almost no virtual difference right now, unless you already have quad furyx to handle 120fps@4k.

 

[renz496]

To win the future they need to win the present first.

 

[Cazalan]

HotChips spilling R9 Nano info

http://wccftech.com/amd-reveals-radeon-r9-nano-performance-efficiency-numbers-hot-chips-2015-prototyping-hbm-july-14/

Confirming the fully activated Fiji. Only slightly faster than R9 290X but 1.9x the efficiency.

The slide published by AMD in their Hot Chips presentation shows that the Radeon R9 Nano will deliver 33 FPS compared to 30.5 FPS (Radeon R9 290X), while the Radeon R9 Fury X blisters past 40 FPS at an exact FPS rate of 42.

 

[cdrkf]

Interesting look at Project Quantum:
http://www.pcworld.com/article/2973970/software-games/exclusive-amds-amazingly-tiny-and-powerful-project-quantum-dissected.html

 

[renz496]

so....AMD afraid for tech sites to review nano?

http://www.techpowerup.com/215776/amd-radeon-r9-nano-review-by-tpu-not.html

 

[fudgecakes99]

Yeah but if we look at it from a micro atx or super micro atx builder, it's pretty amazing when comparison to the form factor of an r9 290x. Thats pretty much on par with a 980. At a little over or under half the size. Running multi gpu setups won't be impossible. Though i don't think most micro atx boards allow that.

 

[renz496]

for mixro AXT i've seen lots of them have second x16 PCI-E slot though most often they only operate at x4 only. but for mini ITX i think it is impossible to add second slot x16 on it.

 

[fudgecakes99]

for mixro AXT i've seen lots of them have second x16 PCI-E slot though most often they only operate at x4 only. but for mini ITX i think it is impossible to add second slot x16 on it.

maybe if you find some special low power x4 gpu you could probably run it but at that point it's not worth the hassle.

 

[renz496]

at x4 the slot should be able to provide 75w. At least that's what i see from pci-e spec. But for cf setup that probably going to bottleneck the second card.

 

[Eggz]

for mixro AXT i've seen lots of them have second x16 PCI-E slot though most often they only operate at x4 only. but for mini ITX i think it is impossible to add second slot x16 on it.

I have a mATX with four PCI-e slots. As for physical size, three are PCI-e 3.0 x 16 and one is PCI-e 2.0 x 4. One of the x 16 slots can switch between x 4, x 8, or x 16. Everything else is fixed.

Something like that probably gets limited, though, to the extreme edition boards because they have enough PCI-e lanes.

 

[renz496]

so more drama than i initially thought.....

http://techreport.com/news/29011/updated-amd-vp-explains-nano-exclusion-apologizes

http://hardocp.com/article/2015/09/09/amd_roy_taylor_nano_press#.VfFTtPQu2Uk

 

[-Fran-]

I wouldn't classify it as "drama", in all honesty. I take it as a "was about time!". They've been neglecting the GPU division by cutting costs and forcing it to abide to the CPU division's Fusion dogma and other stuff. From what I've been reading on the subject, it's a streamline of the GPU Business Unit.

If you want another way of saying it, they want the GPU division to function a tad more like an independent unit and not tied to other units goals and objectives.

I have high hopes for this, since they might be able to scrap some of the fat from GCN and even create a mid arch much like nVidia does to fight back in the mid segment. They really really need it.

Cheers!

 

[renz496]

how did you get into that conclusion? the talk is about how sites like [H, TPU and TR are not getting their samples becaus AMD taught those sites cannot give fair review to their product. and recently amd also cancel the plan to give kitguru their supposed Fury X sample because they commented on how most of the R 300 series as rebrand of R 200 and 7k series.

 

[-Fran-]

how did you get into that conclusion? the talk is about how sites like [H, TPU and TR are not getting their samples becaus AMD taught those sites cannot give fair review to their product. and recently amd also cancel the plan to give kitguru their supposed Fury X sample because they commented on how most of the R 300 series as rebrand of R 200 and 7k series.

Ah, my mistake, hahaha.

I misread the links. There were 2 news going around:

1.- The GPU business unit getting a streamline.
2.- The Tech Report (and others) not getting a Nano for review.

I commented on the first one, but you put info for the second one. My bad.

In any case, yes, drama. That's why PR exists. To create drama and more looking eyes 😛

Cheers!

 

[renz496]

in regards to then i already heard stuff that ATi might exist once more lol as separate company from AMD.

 

[cdrkf]

The Nano looking good

http://www.tomshardware.co.uk/amd-radeon-r9-nano,review-33301-12.html

Not a perf / $ winner, but still damned fast for it's size, and matches Maxwell in power efficiency.

I've said this before and I'll say it again- those who say GCN is inherently a power hog are missing the point, AMD have essentially clocked all their cards up well past the perf / w sweet spot, which makes the power consumption look bad. GCN just doesn't like going above 1ghz- I mean dropping the power from 275W in the Fury X to 175W in nano whilst keeping 90% of your performance shows that nicely.

Now if AMD can keep this philosophy for Artic Islands maybe all isn't lost just yet

 

[-Fran-]

The Nano looking good

http://www.tomshardware.co.uk/amd-radeon-r9-nano,review-33301-12.html

Not a perf / $ winner, but still damned fast for it's size, and matches Maxwell in power efficiency.

I've said this before and I'll say it again- those who say GCN is inherently a power hog are missing the point, AMD have essentially clocked all their cards up well past the perf / w sweet spot, which makes the power consumption look bad. GCN just doesn't like going above 1ghz- I mean dropping the power from 275W in the Fury X to 175W in nano whilst keeping 90% of your performance shows that nicely.

Now if AMD can keep this philosophy for Artic Islands maybe all isn't lost just yet

Keep in mind the GTX970 is using GDDR5. That is like a 25% difference in overall power usage, isn't it? So, you can't say "Maxwell vs Fiji" in this case. It is impressive, yes, but don't forget that "little" detail to put things in perspective when analyzing Fiji's GCN revision.

Cheers!

 

[cdrkf]

The Nano looking good

http://www.tomshardware.co.uk/amd-radeon-r9-nano,review-33301-12.html

Not a perf / $ winner, but still damned fast for it's size, and matches Maxwell in power efficiency.

I've said this before and I'll say it again- those who say GCN is inherently a power hog are missing the point, AMD have essentially clocked all their cards up well past the perf / w sweet spot, which makes the power consumption look bad. GCN just doesn't like going above 1ghz- I mean dropping the power from 275W in the Fury X to 175W in nano whilst keeping 90% of your performance shows that nicely.

Now if AMD can keep this philosophy for Artic Islands maybe all isn't lost just yet

Keep in mind the GTX970 is using GDDR5. That is like a 25% difference in overall power usage, isn't it? So, you can't say "Maxwell vs Fiji" in this case. It is impressive, yes, but don't forget that "little" detail to put things in perspective when analyzing Fiji's GCN revision.

Cheers!

I don't think it's as simple as that- it depends very much *how many* gddr5 chips are used on the 970- given it's fairly conservative memory bus I think the answer is *not many*. I think the 25% reduction was an estimate going from Hawaii to Fiji (which is looking at the biggest Gddr5 interface ever made at 512 bits). The narrower memory interface on GM204 is probably one of the reasons it's so power efficient compared to AMD cards all with wider memory interfaces...

I guess it isn't an apples to apples comparison- but then, whilst the memory *should* save power- fiji does have over 4k stream processors compared to less than half that number on the 970- it's astonishing it's anywhere near the same power use imo.

 

[juanrga]

The Nano looking good

http://www.tomshardware.co.uk/amd-radeon-r9-nano,review-33301-12.html

Not a perf / $ winner, but still damned fast for it's size, and matches Maxwell in power efficiency.

I've said this before and I'll say it again- those who say GCN is inherently a power hog are missing the point, AMD have essentially clocked all their cards up well past the perf / w sweet spot, which makes the power consumption look bad. GCN just doesn't like going above 1ghz- I mean dropping the power from 275W in the Fury X to 175W in nano whilst keeping 90% of your performance shows that nicely.

Now if AMD can keep this philosophy for Artic Islands maybe all isn't lost just yet

First, you are comparing (GCN+HBM) vs (Maxwell+GDDR5). Therefore you are not comparing microarchitectures. Once Nvidia moves to HBM2 the next year the gap will increase again.

Second, AMD increased efficiency of Nano by reducing clocks, which is a pretty easy trick and again has nothing to do with GCN efficiency.

Third, GCN is an inefficient gaming architecture, because it is an architecture that targets both gaming and compute. RISC is less efficient that VLIW, because requires more hardware to do the same work.

Fourth, nowhere AMD Nano maintains the 90% of performance of Fury X. In fact, according to AMD, the gap is somewhere 30% above 1080p.