Embarrassing...
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Async compute gets 30% increase in performance. Maxwell doesn't support async.
- Thread starter StormClaw
- Start date
Embarrassing...
TaintedSquirrel
[H]F Junkie
- Joined
- Aug 5, 2013
- Messages
- 12,691
Steam survey includes iGPUs, which I doubt most AAA developers are targeting. You have to look at the discrete market.
Steam's results aren't even accurate anyway, looking at their card list they have some things listed twice and others listed generally ("R9 200 Series" is listed as a video card). Just for fun I looked up the 980 Ti and it's not even listed... Maybe under the "Other" category, with god knows what else, which makes up a whopping 17%.
Not sure if this has been posted before.
https://www.extremetech.com/extreme...ading-amd-nvidia-and-dx12-what-we-know-so-far
https://www.extremetech.com/extreme...ading-amd-nvidia-and-dx12-what-we-know-so-far
TaintedSquirrel
[H]F Junkie
- Joined
- Aug 5, 2013
- Messages
- 12,691
AMD's performance is incredible in Pointless Chart Simulator 2015.
It must be a really popular game, everybody is talking about it. I'll get the season pass which includes the Manufactured Outrage DLC.
It must be a really popular game, everybody is talking about it. I'll get the season pass which includes the Manufactured Outrage DLC.
The 980 Ti isn't listed because Marketshare wise it is less than a blip, since it is a niche product.
Give it more time so the price of the 980 Ti drops enough, and then you will see it listed.
TaintedSquirrel
[H]F Junkie
- Joined
- Aug 5, 2013
- Messages
- 12,691
You would need to find at least two GPU series that are the exact same age, and had the exact same production length.
If you knew anything about me, you'd know I'd say the same thing about those benchmarks even if I had an AMD card in my rig. I mean it's not like people on this forum called me an AMD fanboy less than a month ago... And people on Reddit called me an Nvidia fanboy at the same time... If it makes you feel better I can pop my old 280X back in, and post again? Or if you don't like the 280X I can pull one of the other four AMD cards running in my house right now. Can't please everyone.
I'm doing some rebuilds later this week, I'll snap an AMD family portrait for you.
Last edited:
just ignore them people that call others biased are the biased one, pretty simple.
Well I don't agree with that, we just have to count everything from the 7xx / r2xx series all the way up till now. Even if the 7xx series came out later we will still see the market place disparity. If we only look at this gen 9xx and Fury and r300, nV will be at that 82 % or close to it. Just by a quick glance at the 980ti and titan x vs Fury, and Fury X the numbers are even more slated towards nV.
Actually no, Steam Hardware does have it's uses.
The thing with Statistical Data is that few people actually have an idea of how to use it, but many armchair warriors think they know how (or try to debunk it).
That's why the JPR numbers are the best to go by for quarterly results, its Q to Q boards being bought by AIB and OEMs. Those cards are sold not just shipped. You also have mercury numbers but there is a variance in thiers compared to JPR but still close to them.
Neither Steam nor going through a retail review list will be any good, and I was actually be satirical with my original post on both those sites.
Explain how steams numbers are no better than others? Steam targets pc gamers not just units shipped to stores. You can tease out how many people are using discreet gpus of a certain class and tally them up, same with windows 10 and dx12. How is being able to survey a large swath of pc gamers systems at the point of contact with their current system an inferior metric in any way?
Because not all people that buy graphics cards are gamers, how about all the business computers in, graphics, advertising, movies, fx studios, etc?. We are also looking at a very select group of people, as well, you can use Steam to give you a general idea like the trend of people buying which cards from different IHV's, but that's it.
If you want a better idea, take the last 2 years (or what ever you think the average person/businesses keeps a graphics card) of graphics card data per quarter from JPR and add up the total amount of graphics cards being sold to AIB's from each IHV and calculate the percentage that way.
AIB's/OEM's will not be keeping stock of cards outside of what they need for sales purposes so their turn over is going to fairly quick and shouldn't affect the end numbers much. This is also the reason why Dell, HP, any of the big OEM's give out major discounts when new generations of hardware come out, they want to get rid of the left overstock because the products based on those will rapidly depreciate because of the new hardware.
If you want to check if steam numbers on the OS, check what the windows 10 adoption rate has been and then check it on steam, they won't align.
https://www.netmarketshare.com/operating-system-market-share.aspx?qprid=10&qpcustomd=0
vs
http://store.steampowered.com/hwsurvey/directx/
at the bottom
see the difference, why is that difference there because we are looking at a select group of people with the steam survey instead of a variable group of people from different segments of the market.
Last edited:
But who cares about how many business have video cards when we are talking about gaming?
Do I care that there are 100% nvidia cards in some shitty office machines that will never play a game ever? Does that mean game developers should go for nvidia because that office has 100% nvidia coverage? No, they should go what people game with (well ideally it would be vendor neutral)
Steam numbers aren't 100% accurate either, I rarely get the hardware survey, and last time it was on my laptop vs desktop only, but at least the numbers show you what people game with over whats used for non-gaming computers.
err who cares, every developer cares if its pertinent to them, graphics engines aren't just used for games, its about applications and everything else. You have 3d real time visualization in many things, not to mention VR and 3d holographic which use similar apps that are close to game engines or modified game engines.
Also the market share numbers won't show up in steam for another quarter or two and when they will show up as a trend and for the steam survey the trend is more important because it shows buying habits, the data itself is practically useless because its a closed group of people. Just look at past JPR numbers and then compare them to 2 or 3 quarters forward on Steam.
Last edited:
I've had a boatload of Nvidia cards until the recent 290X Lightning. I think I posted my gpu history in a different thread but it's very long and mostly nvidia.
FX 5900 Ultra 256mb
8800 GTX KO ACS3
8800 Ultra
GTX 280
8800 Ultra SLI
Radeon 4870x2
HD 4890
GTX 285 SLI
GTX 470
GTX 480
GTX 480 Lightning + Giga 480 SLI
GTX 580 Lightning
GTX 580 Classified
GTX 580 Classy Ultra SLI
HD 7970 Matrix
GTX 780
GTX 780 TI
Araxie
Supreme [H]ardness
- Joined
- Feb 11, 2013
- Messages
- 6,463
Exactly... the company where I work have in a single Local Office about 150+ Machines with at least 1 Titan X per machine and also multiple Quadro cards using Unreal Engine for Medical purposes and imaging and this have multiple franchise Offices worldwide so, yes the market its just so big..
Still the other poster was right with all those machines being meaningless when talking about Async Compute and gaming
Just a perspective thing
Still the other poster was right with all those machines being meaningless when talking about Async Compute and gaming
Just a perspective thing
Async compute isn't limited to just gaming........
Tell that to CPU manufactures.
You think medical imaging wouldn't benefit from it if they were able to get what they need faster, when a patient is on an operating table?
D
Deleted member 83233
Guest
This is the main problem I have with the Steam Survey. I should be able to go in, and survey whatever machine I'm on at that moment. Or at least tell it which machine I want to survey. It seems like every time it hits I'm on my Dell laptop, Macbook, living room machine, etc. I think I've been on my actual gaming machine maybe once when it hit.
Yes, the gaming audience steam caters to has a much higher adoption rate of windows 10 so far. But if you are targeting pc gamers, don't you want to focus on numbers that correspond to that subgroup more than the general population at large?
By August the windows 10 install base seemed to be around 17% for steam users vs around 4% in that more generalized tally.
If I am a game maker, do I want to pay attention to the install base of windows 10 and dx12 on more generalized non gaming focused population data? Or population data that is more tailored to a pc gaming population of steam users?
If anything it makes you want to pay MORE attention to steams numbers, because THAT is the sample that is more relevant to the market you are trying to hit.
That's nice for gpu sales in general, but meaningless to the original question of what GAME DEVS will decide to target. A thousand titan x cards sitting in workstation machines doing rendering work is not the kind of data point a game dev ought to care about. They ought to care about the number of people using a given api like dx12 and running windows 10 for games, and the types of cards they are using and whether they can support dx12 games.
Steam data seems more focused on this subsection of the gpu market and is of prime relevance for game devs and what they will design in the future.
This is the main problem I have with the Steam Survey. I should be able to go in, and survey whatever machine I'm on at that moment. Or at least tell it which machine I want to survey. It seems like every time it hits I'm on my Dell laptop, Macbook, living room machine, etc. I think I've been on my actual gaming machine maybe once when it hit.
That's something steam could change in their analysis, perhaps giving more weight to the data from machines where people actually play the bulk of their games most of the time.
It is a subgroup of the gaming market as well. Do all steam gamer play ever other game that isn't on steam?
Also medical fields and business usually don't fluctuate hardware manufactures as much as general consumers because their software is specific to the hardware and drivers capabilities which they might need.
As an example, I looking into creating a EMR with a friend a few years back, and the nice thing about this is many of the systems that hospitals use communicate to Dicom, which is actually quite legacy stuff, but keeps it easy to plugin new machines into a large system that is already up.
Last edited:
D
Deleted member 83233
Guest
This, or the user could initiate a system scan. If that system isn't already in the survey, it can be added. (though I don't think this would handle single component upgrades very well) Not sure what the best method would be, but I'm sure there are quite a few people that this happens to.
This is why human beings were born with brains that can make extrapolations and expectations.. at least most of us.
We still have to actually test things to see if our expectations line up with reality, but seeing superior support for async compute in one game that reportedly makes relatively light use of it suggests to people with brains that can make educated guesses that future games that also make use of async compute will yield a differential performance boost on amd hardware vs nvidia hardware.
It's not about aots by itself, it's about how a given card can benefit in terms of performance with that style of rendering and compute added into the engine.
Time for an educated guess.
Do you think the steam stats for their gaming population are MORE closely related to non steam gamers or people playing games not using the steam client (like say league of legends or newer EA games) or more closely related to the general windows using population?
My guess, a perfectly reasonable guess, is that the steam stats are more typical of the pc gaming population writ large. Do you think that's untrue?
depends on what games that the people are playing, do you think the vast majority of CS players are using higher end machines on average?
We have to break down all the games out to their individual stats to get a good picture. Pretty much take out all the older games and gamers that play those games that don't push hardware to any discernible limits
We have to break down all the games out to their individual stats to get a good picture. Pretty much take out all the older games and gamers that play those games that don't push hardware to any discernible limits
So you want to take away gamers that play CS(GO) but include desktops that aren't used for gaming at all (workstation PCs)? The hell kind of results are you trying to get?
Araxie
Supreme [H]ardness
- Joined
- Feb 11, 2013
- Messages
- 6,463
and that's the biggest point to discuss vast majority of steam users are CS and dota players..
I hate to interrupt this thread but lol... are they hiring?
Araxie
Supreme [H]ardness
- Joined
- Feb 11, 2013
- Messages
- 6,463
PM..
if we are just talking games you gotta remove all the people that don't push the hardware to get a good picture. Just look at the most popular games on steam. We don't even need to look outside of steam if that is what we are talking about.
Common integrated graphics is at 20% just from intel, how many integrated graphics are there for AMD? Then exclude things like the super low end crap cards. Then the older laptops 2 gens. Then the discrete cards over 2 gens old.
I do think the number of people with nvidia gpus will cause game makers to not leave them behind, they will probably write two code paths to support both nvidia and amd.... unless they are too small to bother.
I was reading that article Joel put up on the state of the state on extremetech
http://www.extremetech.com/extreme/...ading-amd-nvidia-and-dx12-what-we-know-so-far
And there was an interesting comment from Rick La Rose talking about how nvidia got power use down by shifting away from compute workloads in favor of graphics in kepler (and it's follow on maxwell) from a previous David Kanter analysis.
His comment in full:
Rick La Rose Joe Joejoe • a day ago
While it is true that GCN is a power hungry architecture, a lot of that has to do with the extra scheduling hardware and better caching performance/larger caches (BW/FLOP and bytes/FLOP) of GCN. David Kanter wrote the following when he analyzed the Kepler architecture (Maxwell being a derivative of Kepler's architectural strategy while making some steps to improve compute efficiency over Kepler):
"The other architectural change that favors graphics is simplified scheduling. The JIT in Kepler’s graphics driver is now responsible for scheduling instructions that can execute without any register dependencies. The cores have eliminated register dependency analysis, although there is still scoreboarding for long (or unpredictable) latency instructions such as memory accesses. This approach saves power and area for graphics, which is relatively easy for a compiler to optimize. However, general purpose workloads are far less predictable and benefit from more dynamic scheduling; there is a reason that Fermi had such hardware in the first place."
"The shared data bandwidth for the Kepler core is 0.33B/FLOP with 32-bit accesses, just half of GF104. But the standard for general purpose workloads is not GF104. Fermi has 3× the shared data bandwidth (1B/FLOP) compared to Kepler. In comparison, AMD’s GCN has 1.5B/FLOP, demonstrating the advantages of a separate L1 data cache and local data share (LDS). The significant regression in communication bandwidth is one of the clearest signs that Nvidia has backed away from compute workloads in favor of graphics for Kepler."
"The catch is that the Kepler core is a poor fit for compute applications. The excellent efficiency for graphics has undoubtedly come at the cost of general purpose workloads. As our analysis showed, Nvidia’s architects made a conscious choice to quadruple the FLOPs for each core, but only double the bandwidth for shared data. The result is that the older Fermi generation is substantially better suited to general purpose workloads and will continue to be preferred for many applications." - David Kanter
Sources:
http://www.realworldtech.com/kepler-brief/
http://www.fz-juelich.de/SharedDocs.../slides/nvidia-ws-2014/02-adinets-maxwell.pdf
We're seeing the first results of this trade-off. Who knows what it will mean by the time the first few DX12 Titles hit the market. With DX12 placing a larger emphasis on Compute performance, at the expense of downsizing on Graphics performance, 2016 may shape up to be an interesting year.
The number of people using nvidia gpus will probably be the largest force retarding the adoption of more advanced dx12 enhance workloads like async compute since amd has more robust support all the way back to 79xx series cards
I think this was specific to DP performance, that's why Kepler mid to high, not Titan.
And compute performance for Maxwell 2, pure compute is very good more efficient than GCN.
What Maxwell 2 seems to be doing right now, and looks to be completely broken for most part in Dx12 where it works like its supposed to in Dx11, is that concurrent kernel execution of both graphics and compute, which is even before we get to async. Async is the ability to take take the compute instructions as the kernel is being executed and interleave them into the graphics path.
And this is why we were getting crazy results because at times, very few times it was doing concurrent execution and async was happening, but most of the time, it was screwing up.
This is what raised the reason of the preemption and the context switch, which isn't accurate either, well an assessment can't be made on that because it seems to be being used when it shouldn't be.
Last edited:
Unfortunately, future games workloads are going to have Mixed graphics/compute workloads, not pure compute.. below a certain compute workload threshold where the cpu can keep up nicely.
Edited by post after that
sorry yes I agree, but the problem isn't that it can or can't do it, it does concurrent in Dx11 just fine just not in Dx12, don't confuse concurrent or parallel with async, two different things async needs concurrent kernels to be executed to function. Even if kernels are done concurrently it still might not be doing async.This is the main problem I have with the Steam Survey. I should be able to go in, and survey whatever machine I'm on at that moment. Or at least tell it which machine I want to survey. It seems like every time it hits I'm on my Dell laptop, Macbook, living room machine, etc. I think I've been on my actual gaming machine maybe once when it hit.
Problem is, this isn't how surveys work, at all.
Survey's are random because random is what actually certifies the result, otherwise it would be easy to tamper them.
EDIT TO ADD:
Surveys are the basics of statistics analysis, and are actually very important. They are done the way they are done because it has been proven to work, wether the information gathered from it works towards the aspiration of any one group is meaningless.