Mobile Polaris Discussion

Best quote of the articles:

"This is the last part of our RX 480 rumor series. My next post will not include rumors, but something ‘more official’ "

 

First reviews are out, it seems to only barely surpass a 970 in gaming benchmarks for DX11 games, reaching 980 levels in DX12 .

http://www.guru3d.com/articles-pages/amd-radeon-r9-rx-480-8gb-review,1.html

 

Disappointing to say the least, I really expected 980 performance. Probably will come in 2 months after driver updates

 

http://www.notebookcheck.net/AMD-Ra...-Polaris-desktop-card-at-launch.168250.0.html

Best part:

So we need now to wait for mxm version.

 

It's going nutz out there, 1 hour past NDA embargo lift, and already dozens of reviews.

Spoiler

Published on Jun 29, 2016
DX11 and DX12 1080p benchmarks, with all frame data using FCAT capture. Jump direct to the game of your choice here:

00:02 - The Division
01:39 - The Witcher 3
02:34 - Rise of the Tomb Raider
03:52 - Hitman
05:30 - Ashes of the Singularity
07:08 - Far Cry Primal
07:56 - Assassin's Creed Unity
09:06 - Crysis 3

Full Review:

More benchmarks:
RX 480 vs R9 390/380/GTX 970 1440p: https://www.youtube.com/watch?v=Z5YrK...
RX 480 vs R9 380X/280X/270X 1080p: https://www.youtube.com/watch?v=ksN3V...

https://www.youtube.com/results?sp=CAI%3D&q=rx480+review

 

Reference RX 480 is 7% slower than stock 980 according to notebookcheck (34 games tested):
http://www.notebookcheck.net/AMD-Ra...-Polaris-desktop-card-at-launch.168250.0.html

 

If the mobile Version will be about 980m Performance for 200 bucks as well, it might attract some Attention....

 

Except from those of us who already have GTX 980M.

 

It won't be 200$...there is no MXM card that debuts for $200...more like 600-800$

 

I meant more upgradewise or for those who wants to buy a new rig in getting 980m performance for considerably lower Price.

 

That would be a bummer....

 

So, in Europe the real prices of RX 480 today seem to be:

RX 480 (4 GB): 220 EUR
RX 480 (8 GB): 270 EUR

http://www.heise.de/preisvergleich/eu/?fs=rx 480&cat=gra16_512
http://geizhals.de/?fs=rx+480

For comparison the cheapest GTX 970 and GTX 980 now sell for:

GTX 970 (4 GB): 240 EUR
GTX 980 (4 GB): 370 EUR

http://www.heise.de/preisvergleich/?fs=gtx 970&cat=gra16_512
http://www.heise.de/preisvergleich/?fs=gtx 980&cat=gra16_512

http://geizhals.de/?fs=gtx+970
http://geizhals.de/?fs=gtx+980

 

That 50 euro premium for 8gb version is just not worth it. The card is already struggling a lot even at 2K, so not many reasons to play anything at 4K and thus no reason for such high VRAM.

 

8GB version is 8000MHz effective while 4GB is 7000.
However it's possible that AIB partners still overclock 4GB memory to the same 8000 like this Sphire possibly did according to Newegg. 204$ for 4GB RX 480! It's REAL!
http://www.newegg.com/Product/Product.aspx?Item=N82E16814202222

 

Don't forget that the new polaris rx 480 core clocks is only 1266 mhz
Compared to the new gtx 1070 and gtx 1080, per MHz performance should be around the same. Plus, rx 480 has lower core count and lower memory bandwidth
If you look at performance/MHz :

RX 480 2304 core counts
3DMark11 = 18000/1266MHz = 14.22/MHz


GTX 1080 2560 core counts
3DMark11 = 30000/1797MHz = 16.69/MHz

Running in the same core clocks
The RX 480 would've scored 25553 in 3DMark11

Now the question is TDP
According to the reviews, the RX 480 is only running 110-120 Watt in maximum load, while gtx 1080 is running in full 180-200 watt.
So yeah, RX 480 is definitely not a disappointment.
It can easily go movile version without compromising anything from RX 480

 

Anandtech:

http://www.anandtech.com/show/10446/the-amd-radeon-rx-480-preview

 

But it doesn't.
It struggles to compete with 970 in some games so it's a disappointment to me.

 

http://www.purepc.pl/karty_graficzne/premiera_amd_radeon_rx_480_polaris_test_karty_graficznej
http://pclab.pl/art70318.html
After such a hype it's sure a dissapointment, still decent card (in USA, not in Europe).
GTX 1060 should wipe it down.

 

http://videocardz.com/61667/what-re...80-exceeding-pci-express-power-specifications
Nvidia stopped reference designs for MXMs with Pascal and if ever AMD makes it to the market with reference designs I think this won't be efficient to beat the 980M with temp/power limits to the reference boards, 980N is a long shot... Massive disappointment from AMD.

 

Of course not, but it should be close enough. This problem with pcie is strange, I can't find now, but sometime ago, I see that PCIE 2.0 have higher limit of power to 150w, and 3.0 to 300w. But I can't find it now, maybe it was just false rumour.

 

Still kinda curious to see the AIB versions of RX 480, seems like they might clock up to 1,6ghz, here's benchmarks for a 1,4ghz+ OC, looks a bit better, that puts it in Fury territory, but still quite disappointing .

http://oc.jagatreview.com/2016/06/t...on-rx480-ke-1-4ghz-dengan-cooler-3rd-party/2/

 

That 8pin can theoretically do 300w fine.

 

http://videocardz.com/61587/amd-radeon-rx-490-listed-on-amd-website
In other news...

 

Another hype ? Just loose steam out of it...

 

It sounds like a good card for 199/239. If you just want 1080 it's a no brainer at that price. I would imagine the drivers will improve.

Sent from a 128th Legion Stormtrooper 6P

 

Finally a neutral guy.
There are too much amd haters here.
Yep, definitely worth the price 199$ for a near gtx 980 (370$) performance
Not to forget less power consumption

 

Not a hater, but a lover of accurate info

The AMD RX480 is showing just about 970 performance in games and a little better in some benchmarks, even OC'd it's only just in the range of the 970, it never breaks out and hits performance levels of a 980, nowhere near.

For $200-240 the RX480 is competitive against the 970, but not by that much $ difference - the best differentiator is the 8GB model, since you can't get a 970 with 8GB

 

Plus, that DX12 performance is strong, so it sets you up for the future.

Sent from a 128th Legion Stormtrooper 6P

 

I haven't seen a benchmark where the RX480 pulls very far away from the 970 even on DX12... if it's faster, but not so much as you would notice a difference in the use experience.

 

X-factor will be the price and performance of the GTX 1060.

Nvidia can shut down the RX 480 hype if their card is 980 speeds and less than $300.

 

Oh wow, they teased the RX490...

But, what you really want is the RX500 => 2x RX490

 

To be honest I am also quite disappointed, but if we look at bigger number of games we'll see that reference RX 480 performs really close to stock 980 and even closer at 4K. It shows relative performance of 91.2%, while 980 shows 97.1% for 27 games (notebookcheck).

 

It's a matter of finding the sites with the stats that support ones preferences

I don't know what section of notebookcheck you are looking at to get those stats, but here is what I found as summary for synthetic benchmarks and gaming comparisons.

AMD Radeon RX 480 Review - The fastest Polaris desktop card at launch
http://www.notebookcheck.net/AMD-Ra...-Polaris-desktop-card-at-launch.168250.0.html

Synthetic Benchmarks

Spoiler

"On average, the new Radeon RX 480 is ahead of the GeForce GTX 970 (-6 %) and Radeon R9 290X (-5 %) in our benchmarks, but is usually behind the GTX 980 ( Notebook) (6-11 % faster) and R9 390X (4 % faster). The RX 480 can even just beat the GTX 980 in the old 3DMark 11 Graphics subscore, but the Unigine Heaven 4.0 performance is pretty weak in return.

The new Geforce GTX 1080 (Pascal) is – as expected – out of reach and manages an impressive advantage of 86 % on average."

Gaming Benchmarks

Spoiler

"The general gaming performance is obviously more important. During this test, we checked the performance of the RX 480 with 34 games and numerous settings. The average results in 1920 x 1080 and maximum settings as well as 4K with high settings are on par with the R9 290X, and slightly ahead of the GTX 970 (-2 %, but inside the MSI Gaming Dock and less benchmarks). The R9 390X is 5 % faster, while the GTX 980 about 5-15 % faster depending on the device (desktop or notebook). The Radeon R9 Fury (Pro) manages 17 % more frames per second on average (it is particularly faster in 4K).

The RX 480 performs excellent in the targeted Full-HD resolution and manages smooth frame rates in all tested games. Only Anno 2205 and XCOM 2 are on the limit, but high frame rates are not that important here.

The GTX 980 is actually beaten in some titles in Full-HD ( Dragon Age Inquisition, GTA V, Rainbow Six Siege, Black Ops 3, Hitman 2016, and Overwatch), but it also manages a significant lead in other games ( Rise of the Tomb Raider +25 %, Fallout 4 +19 %, Mad Max+16 %). It is 7 % faster on average."

That the 980 being outperformed by the RX480 in 6 out of 34 of these game tests is surprising, and I would like to see those particular game tests redone using the same motherboard / CPU / 980 / RX480 test bed to see a more direct comparison between the GPU's.

The main problem with the notebookcheck review is that the results are taken from many different tests and "test beds" collected over time, so it's quite possible that many of the results aren't good examples for comparison.

Spoiler: Test systems, GPU's

All the other same "test bed" reviews I have seen show the RX480 consistantly below the 970 desktop GPU FPS on every real time comparison, which puts it way behind the 980 desktop GPU, viewed real time watching actual game play.

For me that's the best comparison, from a real gaming pov, and using the same "test bed" to host the same 980 and the same RX490 for all the testing.

 

I'm no hater. Just all the rumours and amd propaganda was way off the truth, it's something that makes a little sad. If nvidia said gtx 1070 was faster than gtx titan x and 980ti - it was even a 2-6%, but it was. Rx480 = gtx980? Just no. Rops cutted half is also very bad, the card itself has worse specs than R9 290 and it can be seen in games.

 

I am also not a hater, I really want AMD to do well and keep Nvidia honest, because if that were the case I doubt we'd see things like gimpworks, and shoddy drivers that lower your performance when the next generation of cards get released .
For there to be a healthy market, there needs to be competition, offering 970 level of performance 2 years after 970 release is just a let down, specially after all the hype, even at 239$ I really cannot say I am impressed .
On top of that, stock vs stock cards, I am sure the 970 is a waaaaay better overclocker, the power consumption is also a bit of a let down a 150w TDP card pulling around 160w in games, when people were predicting around 100w of usage under load .
Mobile cards will end up being trash, under 980m performance, with over 980m power draw, with the 1080m around the corner you can't help but wonder why AMD chose to compete with a 2 year old generation of cards rather than it's real competitor Pascal .
And yes I am aware AMD are aiming for market share, and this was never meant to be more than a mainstream card, If that is really what they are aiming for, maybe that 8gb card should have been the one priced at 199$ just to help them secure market share away from Nvidia .
I am still impatient to see OC'ed cards from AIB's, with custom cooling, hopefully they manage to get a good perf/watt, though from what we have seen so far these cards will probably end up being guzzlers when OC'ed past 1,4ghz .

 

Well guys, I have also looked through many sites and I really cannot say RX 480 is consistently slower than 970. In fact, I took my time to make a summary gaming benchmarks list (by copy-pasting to paint, I am THAT bad at design) from the review of overclockers.ru
There are 8 games with ultra presents at 3 resolutions (1080, 1440, 2160).
Testing bed:
ASUS Maximus VIII Hero (Intel Z170, LGA 1151)
Intel Core i7-6700K 4500 МГц (100МГц х45)
Arctic Cooling МХ-2
DDR4 G.Skill 3333 МГц, 2 x 4 GB

 

Ok.. so this raises certain questions... just what/how did AMD demonstrate performance in Hitman pro resulting in consistent 60 FPS across the board with locked fps?
Correct me if I'm wrong, but locked FPS indicate that the GPU will basically achieve that targeted FPS if capable, and avoid fluctuations.

These reviews for instance are out of sync with that demo. So how did AMD manage to achieve that performance level on 2k, and all highest settings as they claimed?
Also, who the heck took the power draw tests to release the info that 480 was consuming only 110W under load if it came from AMD?

It is actually disappointing that this GPU is offering only close to 980 level performance on a 14nm node with slightly better power efficiency.
Essentially, AMD seems to have managed to achieve Maxwell power efficiency through a die shrink... meaning that the architectural changes brought very little, or they are not evident at all right now with early drivers (which probably also negates the idea that a primitive discard accelerator will increase performance - unless its already doing its job for the most part by compensating for 'castrated' hardware or is not being utilized due to drivers).

I guess we will have to wait for better drivers to see whether performance increases or not.
As for mobile versions... this does kinda put a kink into the idea that we'd be getting at least 980 or Fury Nano level of performance in mobile if the power consumption is that high already (which begs the question, how is the GPU drawing more power than what the PCI-e interface can provide)?

Also, is it possible that power draw is high because AMD might have let the voltage run amok too much for potential overclocking while not reigning it in for it's base clocks?

 

It's not close to 980. It's in the territory of 970, it's for first. Second, as we know amd likely is manipulating materials and their presentation doesnt show all of the truth, just like nvidia. They maybe have some drivers optimalization going on, but even if - it's not going to be more than 10%, with one simple reason - it's still the same architecture - gcn, just improved. Kepler and maxwell were different architectures, and you can see it. Pascal is similar to maxwell, but still I think it has more changes than polaris vs. Gcn 2.0

 

This is somewhat disappointing. The price is good, but in terms of performance per watt, they've basically caught up to 28nm Maxwell. The RX480 has the same TDP as the GTX1070 and is nowhere near the performance of the latter. I suppose there is some hope that because Polaris 11 is a separate chip specifically designed for laptops, it may do better than one expects based on the RX480, but given that the architecture is mostly the same, this is not looking good for AMD.

 

On another note, the 1070 and 1080 for example use GDDRX5... that lowers power consumption in turn by 30%, which might explain why 1070 is in the same TDP range like 480 while having larger performance.
Efficiency-wise, Polaris could be almost in line with Pascal (since it's basically just improved Maxwell on 16nm)

If AMD opted to use GDDR5X too, it's possible that power draw on 480 would be lower.
Then again, we might not have a complete picture of the gpu just yet and what's happening.
It might be worthwhile to wait and see how AMD responds to these claims.

As for performance increase through drivers... it's more than possible. We've seen many AMD cards gaining performance all round on various GPU's as time went on. How well this translates to Polaris though remains to be seen.
I was hoping they will at least be able to give us Fury Nano or Fury (nonX) level performance in DX11... but then again, they might be looking more forward than backwards.
So AMD could be content at providing acceptable performance under DX11 that fluctuates between 970 and 980 at times, while in DX12 it's an obviously different story and things look different.

 

And amd has basically shrink the old gcn into 16nm, so this isnt any real based metrics. Pascal is more energy efficent than maxwell, and so is new gcn. But the improve on the amd side isnt enough. For the card for 199$ is ok, but it isnt revolutionary as they wanted to present it

 

AMD used 14nm, not 16nm.
As I said, if AMD used GDDR5X for 480, energy consumption would likely end up lower.
That and we don't know specifically if they increased the voltages too much to allow for overclocking headroom.

 

I keep reading (not here but out on the interwebs) how with this card AMD has the mainstream all to themselves because of the price. I don't get it, the card is only 20-50 bucks cheaper than the 970 depending on the model. Now if this thing ends up being on par with the 980 with new drivers soon, that's a different story. They should have aimed for that out of the gate.

 

@Deks - The 1070 is GDDR5, only the 1080 is GDDR5X.

@mufferer18 - A 1080 dialed back to 980 clocks shows zero per-clock (IPC) improvement, but there is efficiency one, and it is ~2x, a forum member tested it. That's a die-shrink related improvement, not architectural, just like with Polaris. I mean you can get efficiency improvement from the architecture alone, but that's not the case with Pascal and sadly not the case with Polaris, although it was expected to be - die-shrink + new architecture = huge increase in both performance and efficiency.

Everything I saw lead me to believe that it should be AT LEAST matching 980 @~100W. What we see is worse than that. And I don't mean performance. Well, it is performance as well, but it will be fixed, it's not the first time they've done it (as of now it's ~10% behind 980). The efficiency is pretty off though. The supposed performance improvement would bump the efficiency up, but not by much. All in all AMD is architecture/die-shrink behind nGREEDIA. The RX-480 is still a great (midrange) desktop GPU, but it wont translate to anything groundbreaking in mobile, like I was hoping. And this is the reason for my disappointment, I can't get why desktop people are disappointed. Well, maybe these in Europe where the 4GB is 220EUR (that's NOT 200USD) and don't even want to mention where the 8GB is, or the prices in my country...

Another thing worth mention, the PCIe power draw. Suddenly everyone became engineer. I'm not one myself, but usually things are "a bit" overbuilt than the rated specs. That's why you can chip your turbo engine, or to make it more laptop-friendly - the MXM 3.0, rated according to specs to 100W. There were 120W and even 125W GPUs, there's also a supposed 150W M5500M, which slots in a standard MXM connector, no external power. Why? Because by the same specs that say the MXM is "limited" to 100W, say that the power line should be able to handle 10A (* 19V = 190W). That's of course peak power, but still, that's just over-engineering, which some GPUs took advantage of. How I know? Because same GPUs run fine in older platforms, given that one is able to cool them (so it's not like these GPUs were designed for certain beefed-up machines). So, do you think that the desktop motherboards are designed for exactly 75W with zero peak-power tolerance? Then again it only indicates that it consumes more than 150W and that's the main problem for me. Was expecting kick-ass mobile part

 

More custom overclocking info

http://videocardz.com/61697/how-fast-radeon-rx-480-can-be

https://www.reddit.com/r/Amd/comments/4qk6ug/4chan_uses_a_mounting_bracket_to_mount_a/

(computerbase.de) reported that by manually reducing the voltage on the card, they dropped consumption by 30W and temperatures by 15C while at the same time increasing performance by 8%

Can anyone who understands german confirm this?

 

I'm don't know where you're getting your numbers, but the lowest price on a new GTX 970 I can find on Amazon is about $270 which is about $70 more than RX 480's list price. Most are in the $280-$300 range.

 

Only GTX 1080 uses GDDR5X, GTX 1070 uses regular GDDR5.

 

There's also this review - Fury ~ 390X performance, ~110W... These reviews are all over the place.

 

Leave it to AMD to screw up a launch. I was hoping for fury nano performance.

390x is 10% faster than it at 1440p overall. And 390x is a rebranded 290x.

 

Two interesting threads on Reddit, I didn't see them posted here yet, check'em out.

If you are thinking of buying an RX480, read up on the PCI-E Power issue in thread #2 below before moving forward.

RX 480 - Disparity in benchmarks explained(?)
https://www.reddit.com/r/amd/comments/4qiffg/_/

Briefly, there are disparities between review sites, large performance results differences that the author thought was explained by using 2 different driver versions, 16.6.1 and 16.6.2.

He has a large list of reviews compiled, but it's still not clear what is the disparity in results root cause.

" shinryu744 54 points 10 hours ago

Wow the disparity is HUGE, hardware canucks review makes the RX 480 look like a complete badass. Man I am so psyched for the after market models"

RX480 fails PCI-E specification - original(?)
https://www.reddit.com/r/Amd/comments/4qfwd4/rx480_fails_pcie_specification/

RX 480 powergate problem - solution(?)
https://www.reddit.com/r/Amd/comments/4qmlep/rx_480_powergate_problem_has_a_solution/

The RX480 seems to be drawing too much power from the PCI-E source, higher than maximum tolerance for PCI-E spec, high enough to potentially damage a low spec MB over time(?)

" IDoNotAgreeWithYou 16 points 11 hours ago
Getting a lot of value until your mobo is fried."

There are a number of other reddit AMD threads on this PCI-E Power drawn subject.