AMD Polaris Thread: Radeon RX 480, RX 470 & RX 460 launching June 29th

[cytg111]

Sure if you ignore history of company that made it.

For months their only product was Mantle benchmark, then they announced they have 3 Mantle games in development which were conveniently used to prove how Mantle is getting great traction among developers.

And they barely released first one recently.




Now it gets more hilarious - after skipping on content just to push game out they are spending resources on making engine work on 20 core cpus

- Working towards optimizing DX12 constructs
- Working towards >4 core utilization in games

So, we call them heroes? Or what?

 

[selni]

Sure if you ignore history of company that made it.

For months their only product was Mantle benchmark, then they announced they have 3 Mantle games in development which were conveniently used to prove how Mantle is getting great traction among developers.

And they barely released first one recently.




Now it gets more hilarious - after skipping on content just to push game out they are spending resources on making engine work on 20 core cpus

Wasn't there that postmortem their founder published that essentially said that yeah, mistakes were made on the actual game/budget due to how risky it was for them? It's not news and they've been pretty upfront about all this - Ashes really was an engine development experiment with Oxide with a low budget game as a proof of concept.

 

[Bacon1]

Sure if you ignore history of company that made it.

For months their only product was Mantle benchmark, then they announced they have 3 Mantle games in development which were conveniently used to prove how Mantle is getting great traction among developers.

And they barely released first one recently.




Now it gets more hilarious - after skipping on content just to push game out they are spending resources on making engine work on 20 core cpus

So a new company developed a great engine for use for multiple games with a tiny budget and thats a bad thing how exactly?

http://www.gamasutra.com/view/news/...de_Games_Ashes_of_the_Singularity.php#tophead

Try giving that a read.

They give regular updates to the game just releasing the 1.2 patch. Sounds like solid support to me. More than Rise of the Tomb Raider has gotten yet its AAA.

http://forums.ashesofthesingularity.com/forum/1123

 

[JDG1980]

i wonder if single slot 480 variants are possible?

FirePro W7000 (Pitcairn) and even FirePro W7100 (Tonga) are single-slot cards. Both have 150W TDPs with 1x6-pin. Presumably the same will be true of the FirePro Polaris 10 card.

 

[trane]

Up until HD 5000 / GT 200 series, pretty much every reference ~100W card was single slot.

 

[JDG1980]

Not impressed with that cooler. Should be pretty easy to improve on that from the AIB's

That cooler would suck if it was on a 200-250W card, but for a card with a TDP of 150W that will probably not exceed 110-120W in most games, it should be fine. The leaks we've gotten indicate it does fine on thermal performance, it's reasonably aesthetically pleasing, and Gibbo from OCUK says it has low/moderate noise levels.

It's important to remember that this is a mainstream part with mainstream pricing. It looks like it was very well designed for good performance at a low bill of materials. Of course, there will be plenty of aftermarket versions available for people who want better heatsinks.

 

[thedavexp]

Most of the leaks are running on 16.5.2 driver version. If you take Tonga, double the ROPS, add 256 cores, clock it about 25% higher, and add a small boost due to a faster cache, the leaks basically reflect that.
What it does not reflect is literally any of the new logic hardware in Polaris. That discard accelerator doesn't run itself. It is operated by driver code which does not exist at all in current AMD drivers.
The launch drivers in reviewers hands right now, 16.6.2, are just the first draft of support for the new features of the architecture, and I don't think it would be that surprising if performance in some areas improved in excess of 10%.
A 10% higher FSU score based on arch improvements. Plus a 20% OC from 1266 -> ~1500 on the GPU and the known 12.5% OC on the VRAM. This alone puts the card at Fury X.
Why would production drivers out in public have beta driver code for an unreleased product? It just doesn't go in the build. There is no optimization or hardware support for Polaris right now in the leaks, since the reviews just that missing code in the new press drivers less than 24 hours ago.
I just don't see an AMD developer leaking it. And since all the leaks are right around the same level, differing in 1080 base vs 1266 boost clocks it seems, it is even more unlikely that one guy leaked it to multiple people with different builds. Occam's Razor, 480 drivers didn't leak.

Thoughts? Perhaps drivers are not released for polaris yet so all leaked benchmarks are unoptimized.

 

[3DVagabond]

AMD stock coolers are usually pretty poor...

Unlike the 1080 which throttles like a dog?

Threadcrapping and trolling are not allowed
Markfw900

 

[BlitzWulf]

RX 480 Overwatch at 4k max settings?

https://youtu.be/933LBq34y8I

 

[3DVagabond]

Glo. you picked 2 of the few games AMD cards are known to perform much better at than in the majority of games... Biased.

I think you are going to see many more DX12 titles like that.

 

[3DVagabond]

What about the new tomb raider game clearly being on the side of nvidia?

The problem is Async utilization and aots has that issue clearly and games with it will favor amd for awhile it seems.

Tomb Raider is broken in DX12 on both brands. Let's hope that's not the norm.

 

[Silverforce11]

RX 480 Overwatch at 4k max settings?

https://youtu.be/933LBq34y8I

Impressive. Very smooth performance. Though Overwatch maxed 4K isn't terribly taxing!

https://twitter.com/GFXChipTweeter/status/745887920809218049

^ Also definitely Vega celebration event where Raja is at, so they must have good news on the qualifying stage of the chips.

 

[zlejedi]

- Working towards optimizing DX12 constructs
- Working towards >4 core utilization in games

So, we call them heroes? Or what?

No we call them con artists when they are selling full priced game with barebones content and ask us to spend 1000$ on cpu to play what is over glorified indie game.

 

[flopper]

I think you are going to see many more DX12 titles like that.

going to see a load of excuses from Nvidia owners the coming months and years ahead.

Looks good for the 480 as far.

 

[3DVagabond]

Why do mods allow this kind of trash in this thread?

One can only speculate to themselves. Anything else will get you an infraction. No warnings. Straight to infraction.


Mod callouts are not allowed
Markfw900

 

[BlitzWulf]

Impressive. Very smooth performance. Though Overwatch maxed 4K isn't terribly taxing!

https://twitter.com/GFXChipTweeter/status/745887920809218049

^ Also definitely Vega celebration event where Raja is at, so they must have good news on the qualifying stage of the chips.

Assuming that it's holding 60 FPS mins at those settings that puts it roughly in OC 390X vanilla Fury range.
I'm sold, my 390X made my PC room too warm for my liking.
What I'm really excited to see is how it performs in Fallout 4 with God rays and shadow distance at Ultra.

 

[DarkKnightDude]

RX 480 Overwatch at 4k max settings?

https://youtu.be/933LBq34y8I

Need more info. Looks smooth, but don't see any telltale FPS counter damn it!


Overwatch 4K benchmarks:

http://www.techspot.com/review/1180-overwatch-benchmarks/page4.html

 

[William Gaatjes]

My guess is Microsoft didn't want to use an open source API, after all DX12 is one of their last remaining vendor lock ins keeping large number of customers on Windows.

It is still impressive however that AMD managed to influence the entire industry and contribute to the birth of DX12, Vulkan and Metal..

Microsoft promotes a smartphone, console, IoT, pc and laptop ecosystem that all runs windows and that is compatible with each other. With exception for IoT strong graphics capability is needed and the consoles already showed the advances of bare metal programming. Apple had this for years, because Apple controls the hardware, Apple can exploit more low level hardware features. On windows because everything hardware and software will work, there was more abstraction needed, preventing the hardware to be fully utilized. Now Because AMD is in the consoles and in laptops and pc's with dx12, more can be done in windows.

We should thank who ever came up with the rapport that explained that directx11 and opengl is holding modern graphics cards back.

 

[Silverforce11]

Need more info. Looks smooth, but don't see any telltale FPS counter damn it!


Overwatch 4K benchmarks:

http://www.techspot.com/review/1180-overwatch-benchmarks/page4.html

That's Ultra. There's Epic which is max.

Edit: He recently covered Overwatch, lots of GPUs tested and yes, max = Epic. 390X only averages 40 FPS at 4K Epic.

https://youtu.be/pmLG7YRgpgs?t=4m28s

That would imply the RX 480 is a lot faster than 390X in Overwatch... I wan't to believe, but that's way too fast to be believable!

 

[krumme]

Hey so question to you VR owners? Can you do non VR content just fine? Like can you just surf the web and post on forums with it, without the monitor?

I am thinking about using it to code with. Like how much "screen real estate" do you really get.. and how long can you use it before it gets too uncomfortable?

I was originally into it for professional issues also. You need more res for programming as it is now. Like 4x plus at least meaning 2x 4k at least. But absolutely its getting there.
The real estate is well like the software allows. Its absolutely the future for much professional use.
What we need besides resolution is something like pinch and zoom. The vive controllers can certainly do it but we need the tools. But imo its going to be eg a programmers dream in 3 years.

 

[krumme]

It is just the by AMD chosen vrm controller. It is the case that more phases allow for better response to dynamically changing loads. Meaning that the current drawn by the gpu is rapidly changing all the time and the vrm has to keep the voltage at exactly the same level independent of current draw. That is quite the task to perform.
The current the gpu draws depends on the calculations it has to do. The more it has to calculate, the more current it draws. But this is not constant. It varies all the time.


There is something i am wondering about.
Fury and the other GCN 290x and 390x cards had CU to spare for compute. Because not all CU could not be fully utilized for graphics.
The RX 480 has less memory bandwidth but culls more pixels that are not visible and therefore does not need to calculate them. Saving bandwidth. And it has better memory compression. But that purely benefits graphics.
Also, the command processor is new. They would do that to utilize for more graphics.
I am wondering if RX 480 will be a compute beast as the other gcn cards are.
I think RX 480 is improved for graphics + a set of compute but mainly graphics. Not for compute only.
Because that is the target market for polaris. Graphics and added (asynch) compute.
Vega will be the compute beast.

Ok. So to stay more efficient?

 

[Thala]

That would imply the RX 480 is a lot faster than 390X in Overwatch... I wan't to believe, but that's way too fast to be believable!

Excuse my ignorance. How can you possibly judge framerate using video which is recorded at 30fps? Anything above 30fps and you would not see a difference.

 

[William Gaatjes]

Ok. So to stay more efficient?

That is my assumption too :
And that is a real impressive piece of technical engineering. Because the CU are less idle (meaning more power consumption) but also do more work for each cycle (means less power consumption) and still using way less power because of being optimized for 14nm process. Also, there are less CU. I would not be surprised if a polaris would pop up in the near future with more CU enabled. I am betting that 36 is not the finite number. But i also believe that there is a point of diminishing returns. If you have to much CU and not enough memory bandwidth, you can never saturate the chips calculation power(unless you have lots of on chip cache to keep the cu busy but that is chip real estate and power consumption goes up and larger caches are slower for the same amount of power draw). It keeps power consumption down, but why make a chip larger and then not be able to saturate the available CU with computations. It does not make sense. It is bad economics.

That is why i also find the GTX 1080 an amazing piece of engineering Nvidia pulled off. Both companies are limited by the same physics and have to find ingenious solutions to work around power draw and utilization. In the end, they have to walk the same route.

 

[William Gaatjes]

This is from AMD async compute white paper :

However, submitting all of that work to the GPU and scheduling it efficiently on the available
processing resources is still a complex problem. GPUs rely heavily on pipelining to achieve their
performance, and bottlenecks at any stage of the pipeline can leave other stages under-utilized.
These stages include command processing, geometry processing, rasterization, shading, and raster
operations. Each stage can make extensive use of parallel processing, but the throughput of each
stage is also dependent on other stages that come before and/or after it.

Rumor :
I read somewhere that somebody mentioned that GCN4.0 has out of order execution. I do not know if the previous generations of GCN can do that too.
But that would make much more efficient use of the CU if the CU previously did not have out of order execution.

https://en.wikipedia.org/wiki/Out-of-order_execution

In computer engineering, out-of-order execution (or more formally dynamic execution) is a paradigm used in most high-performance microprocessors to make use of instruction cycles that would otherwise be wasted by a certain type of costly delay. In this paradigm, a processor executes instructions in an order governed by the availability of input data, rather than by their original order in a program.[1] In doing so, the processor can avoid being idle while waiting for the preceding instruction to complete to retrieve data for the next instruction in a program, processing instead the next instructions that are able to run immediately and independently.[2] It can be viewed as a hardware-based dynamic recompilation or just-in-time compilation (JIT) to improve instruction scheduling.

 

[sze5003]

If Vega comes out in October then I'll wait to change my monitor at that point. Seems like the 480 will do me good until then.