AMD Vega MegaThread! FAQ and Resources

Page 5 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.


those card also exist but AMD probably will not going to release them until may or june time frame. this RX500 is just 'refined" existing RX400 series.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985


To be sure :)
 


Lets just hope people with reference cards don't try that....could be ugly.
 


Interesting, I've seen no other mention of the RX 550, only the RX 560 that is expected. I have to admit, I'm surprised AMD hasn't released a card based on the full Polaris 10 chip. IIRC, the 460 was cut down. Maybe that will be the RX 560, but never can tell.

AMD took too long to get their GPU's into laptops, I ended up getting a ASUS laptop with a GTX 1060 3gb (which oddly enough isn't cut down like the desktop part, it's just half the memory of the 6gb version.). I was waiting for the RX 470 or 480 but Black Friday came along and I saved $400 on the laptop and AMD's had nothing to offer at the time...
 


with new process node AMD aim to increase their GCN efficiency with Polaris. when AMD demo polaris 11 against GTX950 back in CES 2016 i think AMD intend the full polaris 11 to be sub 75w part. back then they clock the GPU at 800mhz and it shows very good efficiency. AMD even mention they still not done tweaking the chip. but it seems beyond that polaris start losing it's efficiency very fast. remember the talk about polaris cannot clock pass 800mhz? well i think that is conclusion they get mixed with fanboy reaction but the initial rumor probably talk about polaris efficiency issues at high clock speed, not that polaris cannot be clocked very high.

on the lack of AMD GPU on laptop that one maybe you can blame it on past management. the situation AMD in right now stem from the decision Rory Read made when he helm the company before. we know nvidia usually dominated discrete gpu market share on desktop but on laptop the situation is used to be direct opposite. during 5k and 6k era AMD own 60 to 65 percent market share on laptop discrete gpu!! the table turned to nvidia favor in one generation only because of Rory Read decision. now AMD need to gain back the trust with laptop OEM after the damage that being done by Rory.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
Hey guy's... I came across something last night in the design specs of the new xbox scorpio... I posted it in AMD Future Chips...

It has a Directx12 Chip between the CPU an the GPU.. This is supposedly Dx12 in a chip.. An it is supposed to take halve the rendering load off the CPU. Have you guy's heard of this before.. What's the story with this. Is Dx12 running in hardware as opposed to software ?

Anyway what is this Dx12 chip on the scorpio an can it be implemented on the PC ?

I have tried googling this to no avail.. so far anyway.
 


If you're talking about the dedicated draw call processor, it's because the ability to process a huge amount of draw calls was DX12's selling feature; stuff like shaders compilation is probably still in-CPU, but HSLS isn't DX12-specific and is currently cached after first use anyway.

What AMD and Microsoft did (AFAIKT) was improve the CPU's instruction decoder so that when a game makes use of DX12, the decoder catches DX12 calls and provides dedicated instructions and/or pipes draw calls directly to the GPU. This, I guess, sidesteps the need to decode the instruction in-CPU then route it to the GPU, instead the instruction decoder sends it to the GPU directly. This can easily save a dozen cycles or more per draw call.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
I found some more info on this "GPU Command Processor" as MS is calling it..

An I have posted the details an links in the AMD Future Chips thread.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
What's going on is Volta Early or what ??

"NVIDIA reportedly set to unveil Volta-based GeForce cards in Q3 2017"

Reportedly being the operative word... Wccft is behind this rumour quoting some chinese site as their source.. can't help wondering if it's just wccft spicing things up a bit... Or is it a GPU shootout... they just dropped the 1080ti are they worried now it's not enough ?

This could be just waffle, Volta isn't due till the end of the year...
Or could it be that Intel and now Nvidia are moving their product launches forward to put pressure on AMD ?

Either way I wonder if they have sorted out Dx12 in Volta...?
That's the million dollor question I guess.

http://wccftech.com/nvidia-geforce-20-volta-graphics-card-q3-2017/
 
From other sites I've seen, Volta should be out at the end of the year....for supercomputers. Consumer products are not expected, except in rumors, to be out until 2018. This always happens, trying to steal AMD's Vega thunder.

As for DX12, that would take some modifications, as I understand it, nVidia's current DX12 support is all drivers. They don't have native hardware support like AMD. AMD tends to be very forward thinking.
 


people do that with 290s as well. but the main point of this refresh is it is based on more refined and mature process. so the new card should be (some of them) a whole lot more stable when running very high clock. so even if they flashed their RX480 into RX580 it doesn't mean now their card can easily run 1500mhz clock stable on 24/7 basis.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
Looks like Direct3d is baked into AMD's GPU in Xbox Scorpio.. check it out I posted in AMD Future Chips with links. I taught it was on a seperate chip... but not according to latest article.. If this is baked into AMD's future GPU's could be a game changer.
 


the rumor doesn't really make sense. they said the reason nvidia pushing volta early is because nvidia are not satisfied with the profit generated by pascal generation. but if you look at nvidia earnings for this past few quarters nvidia revenue keep increasing and each quarter they even beat their own estimation by quite a bit. and yet they said nvidia somehow still did not meet their revenue target? i haven't look too deeply into it but sometimes things like this originated from financial analyst random prediction. for example some of analyst out there believe that Sony will be coming out with PS5 sometime next year to counter MS Scorpio.
 


DX12 is more complicated than that. right now AMD async compute implementation seems better than nvidia but at the same time when you look at nvidia and AMD architecture, AMD architecture tend to have more "bubbles" in their execution. that's why despite being 8.6Tflops monster Fury X did not significantly out pace 5.6Tflops 980ti. and speaking of native hardware right now AMD still did not have FL12_1 support in their hardware.

AMD tend to be forward thinking but sometimes that forward thinking did not benefit them at all. remember tessellation hardware that exist in pre 5k series? in the end to use tessellation the hardware must be fully compliant with SM 5 making those pre 5k series card cannot use the feature at all despite having the hardware for it. this time AMD have some advantage in DX12/Vulkan because of Mantle influence on those two API.
 


could be game changer for future GPU but not current GPU. since this is part of DX12 standard that's mean nvidia might also going to have similar access to the said feature.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
looks like AMD developed it with MS though if so.. well if that's the case can't see them licensing it to Nvidia.

Maybe it's just wishful thinking but maybe they may have made a deal with MS on it for the right's already who know's.. They are working very close together an this is baked into one AMD GPU already !

Maybe AMD will partner with MS on this in the future.. Or licence the IP or maybe they developed the tech with them... certainly looks like they did...we shall see I guess.
 


since it is part of DX12 itself i don't think AMD can keep it to themselves only. if AMD have such intention then MS will not going to make it into DX12 component. same case what happen to tessellation earlier. AMD have the hardware for it even in their DX10 class of hardware. with DX11 tessellation is part of DirectX itself. but everyone have the access to it not just AMD. nvidia for their part did not have tessellation engine until Fermi generation. and they said it is also one of the reason why fermi being late to the market (other that the yield issue on TSMC 40nm) because it is one of the last thing they include in fermi to be fully compliant with DX11 SM5 spec.
 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965
One thing to consider. What are the implications of having this in next gen GPU's? FPS will be a joke for supported API's. Will they be able to sell another Video card? Their are $billions$ of reasons why we don't have this already.
 


I guess the main interest is that it allows the lucky ones to use a signed VBIOS that provides lower idle clock speeds/voltages and slightly higher boost frequencies without having to keep an eye on Wattman profiles.

 


Unlikely - considering DX12 is essentially Microsoft's latest bout of Not Invented Here syndrome when AMD introduced Mantle (which it intended to open source but instead gave to Khronos to spawn Vulkan), and the way they proceed with all their APIs and technologies (if not open sourced then at least publicly documented), I don't see them making such a deal with MS.

However, DX12 includes several profiles and extensions, and this could just be one of them; since AMD makes both CPU and GPU, tweaking the instruction decoder to fast reroute a few select instructions is rather easy for them (one could consider this a new step in HSA, actually), but Nvidia would need to synchronize with AMD, Intel and now Via (again) to do so on their own cards; considering the company's history, that would be akin to making the whole upper management staff swallow a murder of crow - each.
 
Status
Not open for further replies.