Leyvin (StudioRaven)2 days ago
The Consumer (Gaming) Drivers that will be available as part of the 17.7.3 Drivers on the 15th August, will provide up to +40% Performance over GCN 3.0 (Fiji-Based) Architecture at Identical numbers of Compute Units and Frequencies.
This was something that AMD has revealed during additional events, that are not RX VEGA / Threadripper Related but were instead Investor/OEM Events where they were discussing the AM4 7th Generation APU Release (A12 9800-Series) but were also discussing the future Ryzen-Based APU successor, which contained GCN 5.0 (Vega-Based) Architecture Graphics Component.
In the slides presented, they showed the direct comparison between the A12 9800 (4C/4T + 8CU R7 "Fiji" Graphics) Vs. AR5 2400 (4C/8T + 8CU RX "Vega" Graphics)... the former being released this week for $120, while the latter will be available in the 1st Half 2018 for $120. There will also be 4C/8T, 6C/12T and 8C/16T Ryzen-Based APU.
A general overview being that the AR5 2400 (you might've seen benchmarks of it on Futuremark and AotS) will have +50% CPU Performance, +40% GPU Performance all using 50% the Power (i.e. it's a 35w as opposed to 65w APU)
Both were clocked at 1100MHz (although the RX Graphics can boost to 1300MHz, which I've included in Brackets)
1080p
Rocket League (High, DX9) • R7G / 87 FPS • RXG / 122 FPS (135 FPS)
World of Warships (High, DX9) • R7G / 67 FPS • RXG / 94 FPS (112 FPS)
Overwatch (Medium, DX11) • R7G / 68 FPS • RXG / 96 FPS (113 FPS)
DOTA 2 (Best Looking, DX9) • R7G / 84 FPS • RXG / 118 FPS (138 FPS)
CS:GO (Ultra Settings, DX9) • R7G / 71 FPS • RXG / 100 FPS (115 FPS)
DOOM (Medium, Vulkan) • R7G / 26 FPS • RXG / 37 FPS (44 FPS)
Battlefield 1 (Medium, DX12) • R7G / 31 FPS • RXG / 44 FPS (50 FPS)
Now while these might not seem like impressive scores, remember we're talking about an 8 Core Integrated GPU at 1100/1300MHz
If you dropped down to 720p, then Doom and Battlefield; should Avg. 30+ and 60+ (on Average) what's more is the RX "Vega" Graphics supports FreeSync and is in (most) Games sitting within the FreeSync Range.
Still this post obviously isn't here to blow smoke up the Ryzen-Based APU rear-end.
Rather it's a Like-for-Like between the GCN 3.0 and 5.0 Architecture.
It provides a very good example of the sort of performance uplift that you can expect to see on the RX Vega 56/64/64-LCE Vs. as similarly clocked Fiji (i.e. FURY / FURY X) from the Day One Drivers in Popular Games.
For Reference Purposes:
RX VEGA 56 Vs. R9 FURY X = x1.765
RX VEGA 64 Vs. R9 FURY X = x2.164
RX VEGA 64L Vs. R9 FURY X = x2.348
[GTX 1080 Ti Founder's Edition] {R9 FURY X}
Battlefield 1 (2160p, Ultra, DX12) • [43 FPS] {10 FPS} 45 FPS / 56 FPS / 61 FPS
Deus Ex MD (2160p, Ultra, DX12) • [38 FPS] {21 FPS} 37 FPS / 45 FPS / 49 FPS
DOOM (2160p, Ultra, Vulkan) • [85 FPS] {55 FPS} 97 FPS / 119 FPS / 129 FPS
Fallout 4 (2160p, Ultra, DX11) • [46 FPS] {28 FPS} 49 FPS / 60 FPS / 65 FPS
Witcher 3 (2160p, Ultra, DX11) • [59 FPS] {35 FPS} 61 FPS / 75 FPS / 82 FPS
Now there is the obviously elephant there being Battlefield 1, where the FURY X numbers take a nosedive... which I think is a Memory Limitation; as both the 1080 and 1080 Ti scale almost perfectly between 1440p to 2160p at 0.52x Performance (which is about right); as such with 8GB the R9 FURY should be getting 26 FPS at 4K.
As such the Vega projected figures are taking that into account.
Now it should be noted that in some cases such-as Doom, the 1080 Ti will gain ~20% Performance Increase from 15% OC (2000MHz) ... but on average 10-17% is the average range of performance uplift that is common to see.
Now while we don't know much about the Overclocking Potential of the RX Vega, AMD themselves have gone on record to state that a minimum 1700MHz Stable Clock can be achieved on ALL Vega.
Obviously for the 64 Liquid Cooled Edition (Stock 1677MHz) ... this is a guaranteed increase of 23MHz. Yet, with this said the Polaris Architecture was guaranteed to clock to 1340MHz Stable... beyond Silicon Lottery (I won quite well, as mine will OC to 1520MHz; although I typically keep it at 1300MHz Stock/OC)
Another thing to note, is obviously I can't account for Thermal Throttling as I don't have an RX VEGA Frontier Edition to compare to and most Reviews on it are "Eh" when it comes to actually recording Min - Avg - Peak Clocks during Benchmarking.
This is important to note, because while the Liquid Cooled is almost certainly going to be able to maintain it's Clock, thus yeah THOSE are the performance numbers we can expect (£700, not looking like such a bad price now... eh)
For the Air Cooled Vega 56 and Vega 64, I almost Guarantee with Stock AMD Settings, Fan Curves and Voltage... it's going to be more power hungry than it needs to be, and it'll throttle a damn sight sooner and harder than it needs to as well.
Now I heard that the Air Cooled Frontier Edition, typically sits around it's State 4/5 while in Operation ... which makes sense as my RX 480 and WX7100 does as well; which means while peak it'll certainly hit it's Theoretical Performance Figures for it's Clock. More often than not I'm actually getting ~88% Performance due to Throttling and the Reference Coolers typically are a little more inefficient than the AIB like I have.
We can essentially say roughly x0.80 is likely to be what we can expect from actual Real-Time Performance.
i.e.
Battlefield 1 (2160p, Ultra, DX12) • [43 FPS] 36 FPS / 45 FPS / 61 FPS
Deus Ex MD (2160p, Ultra, DX12) • [38 FPS] 30 FPS / 36 FPS / 49 FPS
DOOM (2160p, Ultra, Vulkan) • [85 FPS] 77 FPS / 95 FPS / 129 FPS
Fallout 4 (2160p, Ultra, DX11) • [46 FPS] 39 FPS / 48 FPS / 65 FPS
Witcher 3 (2160p, Ultra, DX11) • [59 FPS] 49 FPS / 60 FPS / 82 FPS
In essence all games tested with the Vega 64 at 4K are within FreeSync Range... so this I'd wager heavily is why they're marketing it along side FreeSync 2.0. Now this performance is almost certainly due to Primitive Discard, which greatly reduces workloads (by about 50-90% Geometry but Workload by 20-40% on average) ... HBCC will have very little effect on performance due to nothing using > 8GB VRAM and there being more than enough Bandwidth Available to keep the GPU fed; however it will almost certainly eliminate Memory Call Latency, thus increasing Minimums (Frame Rate Jitter) to be substantially closer to the Average.
< • >
Now keep in mind this isn't using the DDR Pipelines, I mean it could be of course and just really poorly optimised ... which alright this is AMD so plausible in Release Drivers. Still given most of these gains can easily be explained by things like Primitive Discard, Tile-Based Rendering, etc. and I'm sure we might see some more performance gain from FP16 (I'm looking at you Wolfenstein New Order II)
As such there is still 2X Performance that the RX VEGA could still have available to be unleashed.
Yet there are certainly some caveats to be made concerning this that I didn't go over before.
API Support... as in Legacy APIs would almost certainly need a Driver / Developer workaround, very similar to how the R9 295X2 was (or rather often wasn't) supported. For the moment Legacy APIs are still quite prevalent; so while sure they could essentially out-perform SLI 1080 Ti with a Vega 64 (Air) this would only be in Select (primarily the most modern) Games, and very dependant upon the API.
So if for example in DirectX 11 you're seeing the 1080 Ti and Vega 64 trading blows, but then in DirectX 12; you need a second 1080 Ti to keep comparative Frame Rates; well what conclusions would you as a Gamer take away from that?
Chances are you'll be sat there thinking "Wait, if Vega can get 2X Performance from DirectX 12... why is my RX 480/580 only seeing 20% performance improvement?"
You see the problem there? As I said this is as much about Brand and Product Image as it is Performance.
AMD need to be seen as Competitive, but they can't be seen as "Top Dog" without the expectations that come with it from Consumers.
On top of this, they can't alienate their own established Consumer Base.
Now, if each Generation (GCN 5.1 in 2018... GCN 5.2 in 2019) they increase the Thread Throughput by +50%, Reduce the Clock by 15% (depending on what NVIDIA release). This would result in a natural reduction in Power Consumption, along side actual improvements in the 14nm process; and costs would continue to fall, which allows for bigger profit margins or again another drop in price that NVIDIA are forced to follow to remain competitively priced.
This would keep the RX 690 and 699 close enough to what NVIDIA will release that via Overclocking they can appear Competitive, while AMD looks like it's Edged ahead. Yet what's more important better Power Consumption, Lower Temperatures ... things that even with the RX Vega Frontier Edition, which is essentially on-par with the 1080 Ti in this respect (while being quieter); it's AMD still getting blasted for being "Power Hungry" (they list the peak, not the Common Draw) and "Hot".
Like with Ryzen Vs. Bulldozer, AMD need to dispel this Stereotype of Radeon Products.
We can see this plain as day with AMD essentially being in the Technology News 24-7 for the past Year... they're starting to drown out their competitors and that is where they're going to start winning, and winning big.
They've always had the Technology and Innovation, what they need is the Market Share for that to actually mean something and be supported by Developers.
Show less
Biggest issue of course will be availability and it may just come down to which card you can actually buy.
