How come AMD's hardware specs are always superior (bandwidth, FLOPS, texel fill rate, etc) but the cards either end up being on par or slower than Nvidia?
>>54812900
nvidia pays devs to sabotage AMD
Three reasons.
1. Games don't push AMD's strengths, that being high bandwidth, massive shader and texel throughput, but opt for pixel fill/throughput which is Nvidia's forte, and pretty much the only objectively measurable scale Nvidia beats AMD at.
2. >>54812920 goes hand-in-hand with 1.
3. Their drivers have higher overhead than Nvidia's, meaning they lose out performance from the start.
In short, it's entirely a software problem.
>>54812900
>but the cards either end up being on par or slower than Nvidia
Just wait a couple of years when nVidia stops optimizing drivers for that arcitecture.
>Fuckin Kepler.
>>54812900
AMD is just plain shit.
FLOPS is a theoretical figure that has no bearing on real world performance. The calculation is :
ALUs x clock x operations per clock
IE 4096 x 1050mh x 2 = 8.6 TFLOPS
Bandwidth itself is analogous to how wide a highway is. If there is no traffic using the road, then all the additional lanes do nothing. One car traveling down an 8 lane road is not 8 times faster than if it were traveling down a single lane road. Peak bandwidth figures are very rarely reached in real workloads, the memory controller itself can be slow enough to have a measurable performance impact in certain ops.
Texel and Pixel fill rates are again not directly tied to any real world performance benchmark, they're a small part of a real world rendering workload. If you're dealing with a big scene and you're ROP limited then you might see where this would come into play.
Hardware is not so simple as to be reduced to a few bullet point numbers. Every GPU arch has its nuances, and they're so numerous that they fill entire programming guides for every revision.
>>54812979
Also I forgot to add, Nvidia is faster at peak
bilinear filtering for fp16, but not for int8(the scale might show up as gtexel/s in some tools, but I haven't seen the measure used in a long while)
>>54813031
>just wait a couple of years
The never ending song of the AMDCucks
>>54812900
AMD wins more at raw compute than in games, and they don't have their drivers as optimized, it's why they're pushing DX12 so hard and why their cards do much better under DX12
>>54812900
>How come AMD's hardware specs are always superior
Their chips are denser, more transistors per mm2, that's where the higher numbers come from.
>but the cards either end up being on par or slower than Nvidia?
Worse software.
>>54813077
All you have to do is look where the 7950/70 started and where it ended
>>54813077
The 7970 was no competition to the 780 ti, now its viseversa, high end nvidia gpu's are not a good investment.
Fermi gpu's commit suicide
Kepler got gimped
Only time can tell what maxwell will suffer.
I heard that the 980 ti and the titan x already got slightly gimped after pascal's release.
>>54813335
Maxwell already suffers it lacks DX12 support
I can't fucking wait for polaris 10 this shit is taking too long
>>54813335
Yea can confirm, drivers are more stable but gimped
Did you notice benchmarks:
>no sli 980 or 1080
980: impact on 1070 market position perf
1080: dual gpu min frame rate is the same as single gpu
There's a major issue on sli where the min frame rate is bottlenecked to single performance
4k results are very similar to maxwell all things considered
>relative perf multipliers
>>54813433
I can't wait to hear how AMD's benchmarks blow Nvidia's out of the water but all those benchmark websites mysteriously disappear from the domain registry and no YouTuber has benchmarks showing AMD is better
It's almost as if there's a conspiracy. Or AMD is put in place by Nvidia and Intel to be second-class and avoid monopoly fines (in other words it sucks)
>>54813465
Polaris 10 doesn't compete with nvidia stuff, it has a chance with DX12, but it's just the 480/X most likely,
AMD is waiting on HBM2 for Vega, and probably had to wait for GDDR5X on the 490/X
>>54813465
>Polaris
>http://videocardz.com/60561/amd-shows-off-radeon-rx-480-running-doom-at-1440p144hz
No chance it's competing with the 1070.
>>54813534
>doom
Why the fuck are they showcasing it with a game from 1993 that ran in total software mode?
>>54813534
It shouldn't be, the x80 part is meant to compete with the x60 part from Nvidia.
>>54813558
Are you not aware of the fact there was a new Doom released very recently?
>>54813578
I believe he is using wit.
>>54812900
> 300W
> Passive
wut
>>54813608
Server cases have huge amounts of airflow so graphics cards don't each need their own fan.
>>54813534
I'm glad delusional faggots will stop trying to claim that Polaris 10 is going to be the 490X now. That was just unbearably stupid.
>>54813608
Server hardware. Think about it.
>>54813608
firepro cards meant to be used in thin server rack mount cases that have their own fans designed for whole case airflow. even the CPUs in those are passively cooled
>>54813608
this stuff goes into servers or shit with propper case cooling
>>54813643
>>54813625
>>54813616
Big enough cabinets are actually climate controlled. They're hooked right up to the build's AC. I've seen plenty that have their own dedicated units.
Thats a big part of why HPC is so costly.
>>54812900
Off the top of my head, their architecture comes to mind.
Fury X has 4096 stream processors, 980ti has 2816 CUDA cores. CUDA cores tend to be more powerful than stream processors (note this is pretty much both marketing terms for a graphics processing core), whereas stream processors are more plentiful.
980ti has 96 ROPs, Fury X has 64.
980ti has 176 texture units. Fury X has 256.
etc.
>>54813643
>>54813625
>>54813621
>>54813616
Ah, all right. But why would a server need such a robust videocard?
AMD is hot garbage always fucking shit m8
>>54813678
render server
compute server
anything you don't want to do on your workstation that can use parallel computations
>>54813678
certain problems scale really well on a GPU architecture -- think of a GPU as thousands of really shitty CPU cores.
>>54813674
>Big enough cabinets are actually climate controlled. They're hooked right up to the build's AC.
I've worked with an entire server room that was hooked up straight to the AC. Was always fun opening and closing the door to the room because the pressure difference tried keeping it closed.
>>54813534
>running-doom-at-1440p144hz
Isn't that a good thing?
>purchase nvidia for longevity
>>54813717
many large buildings have full sized systems just for server room cooling, as it is more important to keep that system up on its own in an emergency than the whole building's AC
>>54813747
500 series card didnt get gimped, they litterally killed themselves
>>54813726
yes
Also should i wait for the 1060 or polaris?
I always buy the X60 ti from nvidia or the x80x/x870 from amd, those are always the perf sweetspot, but do you think that the 1060 wont suck at compute (i mean tflops compared to 480x)?
>>54813726
That's nonsense because even the 1080 can't run it at 90FPS
>>54813683
Okay.
>>54813884
>500 series card didnt get gimped
True, the 580 was the last *80 card that didn't have any artificial power limit or any missing pieces on the PCB.
>>54813903
Its a 144hz freesync monitor, the GPU is not running the game at 144hz.
Learn to read.
>>54813747
I bought a 670, I won't fall for the nvidia meme ever again.
>>54813912
It did have a hardware power limiter to prevent it from frying itself.
https://www.techpowerup.com/134460/disable-geforce-gtx-580-power-throttling-using-gpu-z
>>54813923
My 560 doesnt think so.
>>54813923
Whoops, was it the 480 the last one without the power limiter then?
>>54813941
Haha filthy binned peasant, only the glorious GF100 is allowed to have that.
>>54813954
GF110*
Whoops.
GF100 was regular Fermi
>>54812900
To my understanding every AMD card from the last (pre-Pascal generation) is faster than their nVidia Maxwell equivalent except Titan X (that loses to that monster 295x2 card or something)
Before you call me AMD fanboy I have owned two Titan X and currently own two 1080s.
>>54813979
And heck, even in bunch of notable benchmarks 'non-equivalent' AMD cards beat up their 'superior' nVidia cards.
I guess nVidia stays afloat because of retards like me.
>>54813979
No, Fiji is still faster than GM200. On paper, at least.
>>54813991
That's a hell of a thing if it applies outside theory.
>>54813944
Yeah, and the spectacular inferno that was GF100 is why the second generation of Fermi onwards had the power limiter.
I've applied my own theory to AMD/NVidia unreleased cards.
Nvidia needs 20% less TFLOPS than AMD for equal performance.
This is in DX11, might change with some lower overhead API.
>>54814036
Tflops wise amd's offerings are supperior, but that doesnt matter in gaming, just in dx12 titles where async matters too.
These new "use all the fucking system resources available" API's are pretty cool, old xeons and the fx lineup shine and even manage to beat current intel offerings.
>>54814057
The 480 was like 100C from factory though and wasn't it the 580s that literally caught fire?
>>54814075
480, 470, and 465 (all GF100 with various units disabled) spawned the Thermi memes, but I think it was actually the 590s that had the "burst into smoke and flames" issue. I know there was that one rather well publicized event where the guys were testing a 590, and when power was re-applied immediately following a reboot, the damn thing went pop.
There's also that one GIF floating around of the video card (I think its an MSI card) going up so fast and hot it melted the solder for the PCI-E power connectors (~220 degrees C) nearly instantly.
>>54813987
why are you running the 6700k at 4.0? at 4.3 it barely makes a temperature/voltge difference
>>54814109
I still have a 480 in my system (alongside a 970) but it doesn't do anything at the moment. Just warms up the water cooling
>>54814135
I have had no reason to OC it yet.
>>54814109
I would also like to add that for the GF100 Fermi, one site got their hands on a GF100 card where the last shader cluster was disabled via bios instead of fused off. They re-enabled the last cluster so the GPU was running full fat 512 shaders, and put a fuckhuge Arctic Cooling triple fan heatsink on it.
Even with that heatsink, activating the last shader cluster increased load temperatures to 97C and doubled card power to 600W. Whether it was because it required a massive voltage increase or what, the fact of the matter is that none of the consumer silicon could safely run full fat 512 shaders.
>>54814146
Why waste the energy running and cooling the thing? There are plenty of better options for a secondary GPU.
>>54814174
>I would also like to add that for the GF100 Fermi, one site got their hands on a GF100 card where the last shader cluster was disabled via bios instead of fused off. They re-enabled the last cluster so the GPU was running full fat 512 shaders, and put a fuckhuge Arctic Cooling triple fan heatsink on it.
>Even with that heatsink, activating the last shader cluster increased load temperatures to 97C and doubled card power to 600W. Whether it was because it required a massive voltage increase or what, the fact of the matter is that none of the consumer silicon could safely run full fat 512 shaders.
Can I have the article? Sounds interesting.
>>54814182
Just google gtx480 512 SP
>>54814201
Holy fucking shit HOW? I have two 1080s and they pull below 400 during games, jesus fucking christ.
>>54814210
>HOW?
Fermi is Nvidia's netburst.
>>54814182
The original article I think has been deleted (it has been a while) but >>54814201
has the gist of it.
>>54814210
The GF100 silicon was really really REALLY fucking bad. 1.7% yields on average (which ended up adding to the whole pile of shit we gave nvidia). Some of the finished wafers were so bad NONE of the chips on them were viable, which is like $40k-50k in hardware per wafer.
Nvidia tried to do too much too fast and when TSMC fucked up they got punished for it, which wasnt helped by the fact AMD had dropped the Radeon HD 5000 series which were superior to Fermi in many ways.
>>54814257
Ridiculous, lmao. I knew Fermi was a shit but I didn't know the level of shit it actually was.
>>54814257
Even with how bad Thermi was, AMD still barely outsold it with its far superior and cheaper Evergreen architecture.
Just goes to show you that you need far more than a good chip, marketing and brand name are 70% of sales.
>>54814305
Sadly this is true.
The most profitable companies invest more in marketing than in R&D
>>54814283
The architecture itself wasnt that bad. It was the silicon it was on that was absolutely terrible, because TSMC a shit. GF104 and the smaller dies were far better about the heat to the point where a pair of GTX 460s were the go-to cards over a single GTX 480 due to how damn well they scaled.
When the GTX 500 series came out, Nvidia had managed to fix the bulk of the problems with Fermi (some of which included disabling most of the FP64 performance on the consumer cards), TSMC had fixed their shitty process, and the series was generally better received.
Part of the issue was that Fermi was hot-clocked (the shader domain ran at 2x the speed of the rest of the die), which means the shaders needed more voltage and thus more power to run.
Nvidia eventually did away with hotclocking entirely in Kepler, where they doubled the number of shaders to counter the halving of the clock speed, then doubled again to get more performance. They also almost completely removed the FP64 hardware, leaving a token amount for the things that really needed it, and the FP64 hardware is really power hungry.
>>54814428
Thanks for the information, I guess. Are you a GPU historian?
>>54814459
Not really, I just read a whole bunch of articles and shit when I was considering an upgrade to a GTX 580, and when I instead acquired a used GTX 660, I did some reading on that as well.
I've also been putting some thought into bouncing over to the red team, as GCN as an architecture has aged really fucking well.
>>54814532
Heh, I was considering AMD before buying two 1080s, but these were quite cheap here so why not.
In retrospect, buying two Titan Xs in last generation was bit waste, should've gone for high end AMD dual setup like dual Fury X.
Hopefully AMD comes up with strong competition, it is healthy.
>>54814564
Well, we'll just have to wait and see. Hopefully Polaris proves to be a solid design, cause if not Nvidia is going to shove that thorny capitalist cock right up our asses without lube.
I kinda want a Fury Nano though. A FuryX wont fit in my rig because I dont have a space to put the rad(s), and I likely wont have the power plugs required for a Fury, or the space since those cards are huge.
>>54813483
they aren't going to use gddr5x at all.
>>54815122
With those narrow memory buses they sure are.
>>54815122
Plain old gddr5 or what?
>>54815151
GDDR5
>>54813884
nvidia doesn't have true dx12 support, that also means vulcan isn't fully supported, they can run those, but they gain little from it
meanwhile amd has dx12 support that improves performance meaningfully.
without seeing the 1060, i would say is likely going to be amds game, however nvidia has been known to cut a higher end card down to make it price competitive with amd or just out right better (see fury x and 980ti) so there is a chance that if the 480 is kicking nvidia's ass to hard, they will price adjust/cut down a higher end card and sell it competitively price wise.
but i despise nvidia's business practices ontop of more petty reasons, so i would say amd even if it wasn't obvious amd won this generation of gpus.
amd is shit and you are looking for rationalizations to help with your denial.
see: https://en.wikipedia.org/wiki/Denial
>>54815140
Not necessarily, narrow busses can be offset with fast memory or good cache management/large caches
>>54812900
Nvidia buys out devs, with the ones they don't buy out optimizing more on nvidia due to the higher market share (due the higher performance from buying out devs)
It's a cycle of anti-consumer bullshit.
>>54812900
if you are talking about consumer cards
1) you got nvidia gimping amd by getting devs to use their render paths instead of a natural one.
2) driver overhead on amds part
3) nvidia pushing retarded levels of things it does well (see tessellation)
4) no 20nm node, what the fury was made for, and due to it being re purposed for 28nm there isn't enough bandwidth to properly use all the cores.
now when you are talking about pro cards, this goes out the window, as they are signed drivers tested to work on pro programs, guaranteed to not crash and if they do you get dev time to fix the issue. here its less an issue of how fast the card is and more how much memory the card has and amd usually comes out ahead, the only reason to use nvidia was because they were first to gpgpu with cuda and allot of shit uses it.
>>54813127
Also this.
The 7970/280, gained something like 20% performance from drivers over the years.
>>54815242
I fucking hope the same gains will make it to my 380 but I doubt it.
>>54813678
You are now imagining how long it would take to encode, say, 4 hours of 8K video content with a normal workstation.
Or you could have access to 64 of those cards with a much quieter, more energy efficient workstation and have it be done in seconds.
>>54815140
i forget the reasoning, but its mostly because its at best a stop gap. nvidia likely wont put hbm2 on consumer cards this gen, but amd will on vega, and this is where the memory limits would become an issue too.
why use expensive gddr5x when gddr5 gets the job done but isn't fast enough to compete with hbm2?
>>54815151
on the cards that don't need the extra memory bandwidth, gddr5, on the cards that do, hbm2
>>54815275
>have it be done in seconds
I'll be the first to admit that I don't have a clue what to do with it but that makes my dick diamonds.
>>54815283
Won't that split the mid and high-end market massively in terms of performance?
>>54815293
Imagine the deliciousness, you could fill up your harddrives with 5gb x265 4K rips with excellent quality in a couple of hours.
>>54815306
AMDs Zen APUs are looking to be ridiculously power - they're the ones looking to the future, where there will be no market at all for low end cards because even mid tier APUs will do 1080p in most games.
Polaris 10 is a stopgap for the 470/480/X, vega will be the '490/X' (they'll keep the fury/radeon pro branding) as well as probably another 2 or 3 cards higher based on it.
Polaris 11 is for laptops which will have weaker APUs
>>54815352
I'm just worried that in the future hbm2 cards will be ridiculously expensive and gddr5 cards won't be fast enough anymore as an alternative.
I'm hype for Zen though senpai.
>>54815188
amd has done right by me far more then nvidia has.
still can't get over nvidia requiring dx10 and vista for 3d when i was looking into dropping the money to get a 120hrz 3d setup.
many of you will call it stupid, but you cant say that till you have played a game in 3d or vr. at that point in time (around 8-10 years ago), 3d was the closest thing to vr you would get, and they lost me because of that, and my next card after the 6800 ultra was a 5770 1gb and i refuse to look back till amd isn't in the ballpark performance wise any longer.
intel also fucked me hard with the p4, and along with its business practices i refuse to use them if amd is viable... zen is the deadline for amd to get viable before i get a 6 core intel, possibly an 8 core or dual cpu xeon setup, all depends on how much money im willing to throw at the issue.
>>54815270
they still use gcn as a backbone, so likely you will see gains, especially in the driver overhead departement when they fix shit like that.
>>54815306
not really, take a raw photo from an slr, that shit is 28+mb
toss data away that is over 8/10bit and you save a fuck load of space
then compress it losslessly and you save more
if you compress it lossy then you can make the file even fucking smaller with next to no degradation
thats kind of what videocards already do and new compression methods allow slower memory to act as though its faster then it really is.
and memory bandwidth isn't really a limiting factor for games till you hit 4k, and the 480/x isn't made to play 4k, its made for 1080 and 1440 if the demos are anything to go by
>>54812900
>How come AMD's hardware specs are always superior
Uh-huh.
>>54815352
My current prediction:
Polaris 11 - R7 460 (dropped with zen apu launch)
Polaris 10 (low) - R9 470
Polaris 10 (medium) - R9 480
Polaris 10 (high) - R9 480X (launched with vega)
Polaris 10 (maxed out) - R9 490
Vega (low) - Fury R5
Vega (medium) - Fury R7
Vega (high) - Fury R9
>>54815379
Because you're comparing Hawaii to GM200 you twat.
>>54812900
>>54815379
Territory performance
>>54815367
Only HBM2 matters for 4k 120hz, to pave the way for 8K. At the rate of adoption and sheer bandwidth consumption, market bottlenecks
AMD early HBM2 adoption vs Nvidia early release design
DisplayPort 1.4 supporting native FREESYNC is the end of gsync®
>>54815367
If you don't care about the noise then get a used dual socket Xeon server. They have an abundance of cores and processing power. It is possible to make them quieter too.
I feel the same as you do though. Either Zen is good or I'm sticking with Intel. My i5-2400 has served me fucking well.
>>54815366
HBM will inherently come down in cost as more and more gpus need it to achieve high bandwidth within a sensible power envelope. Right now HBM isn't really needed as you can still clock GDDR5 or GDDR5X high enough without starting a fire but that isn't going to be viable for long as heat density icnreases as gpus have more and more memory (and thus need more and more bandwidth).
HBM is the solution to this problem. HMC could've been as well but JEDEC said nyet on that.
>>54815407
I'm comparing two gpgpu pro cards released in 2015 in the same price segment.
And one of them outperforms the other with a good 25%.
What comparison do you want exactly?
I'm surprised AMD even has a pro line still, I've been working in the cgi industry for a good 5 years now and I'm yet to see a FirePro lmao.
>>54815534
They aren't comparable, GM200 is a far bigger GPU and its direct competitor is Fiji, both in FLOPS and TDP.
Hawaii is slower than GM200 at SPFP, but magnitudes faster at DPFP, its competitor isn't even Maxwell, but GK110, which again is a much larger GPU than Hawaii.
>>54812900
>>54815495
i'm on a phenom II 955, my cpu is my bottleneck more often than not.
i want to render video, but my 955 is just to slow to do it and other things, i sleep in the same room as the computer so rendering at night isn't doable and i cant render while im away just in case something catastrophically fucks up.
looking a xeons i can get large core cpus for fairly cheap, however they are also crap when it comes to single core performance, better then what i have, but still crap.
what i do all comes down to how much single core matters to me, as having 8-16 cores would open up new avenues for what i can and cant do more than high single core, though i would like to have the high single core.
>>54815589
I think you're missing the point completely.
Those cards were released at pretty much the same time.
If AMD isn't releasing cards with their best tech THAT IS THEIR FUCKING PROBLEM.
The real world isn't /g/. I don't care if AMD can be better than Nvidia in a synthetic bench. And I sure as hell don't care if it can be better with a hypothetical card that doesn't' exist. My management doesn't either.
Not that it matters for the professional segment anyway because AMDs drivers are still junk to this very day.
>>54815628
Even my old Xeon 1240v2 is pretty quick. I don't do 4k video encoding though. Isn't video rendering mutlithreaded well nowadays?
If anything you could just encode a different video and assign a core each couldn't you?
>>54815634
You compared two GPUs for a completely different purpose you illiterate halfwit, your underpaid job is irrelevant to the discussion, which was that AMD has more powerful hardware on paper, which it does.
>>54815661
>for a completely different purpos
Quadro and FirePro are used for the same purposes you inbred slav/indian/arab whatever retard AMD scum sucking retard you are.
Stop talking about shit you don't understand, huh? Ever thought you can do that?
>>54815685
Kill yourself you idiot, how is a GPU with next to none FP64 capabilities (1/32) comparable to a GPU with 1/2 FP64 ?
It's not, the GPUs target different fucking markets, one is for a HPC subset, the other is for workstations like the one you draw your render gay porn on.
Shut the fuck up and fuck off you dumb child.
>>54815654
720p 30fps takes about 1.5-2 times longer to render then it does watch and that's going as fast as i god damn can.
i want 1080p60 but even 1080 30 takes 3-5 times longer to render then watch.
this also maxes cores entirely, if i assign it 3 cores, its still only one core to use and even longer rendering times, and while single core doesn't matter to much, a fast single core, as in something twice as fast if not faster (current i7's) will out pace older 8 core xeons, not to mention this isn't just a render box, its a computer i use all the time and not everything will be multicore enabled, or optimized.
>>54815722
I've encoded a fair few 1080p files on my Xeon and it was always took less than the file was long. I'm sure I used to encode a 90min film in about 30 minutes or so.
Have you tried different software? I remember having to go through several before I found one that actually used more than one core.
When is Polaris gonna to be released I was gonna buy a 1080 but now I'm not so sure
>>54815748
Poolaris is shit don't even think about it Pascal is the far superior architecture
>>54813335
That's amusing - 980ti myself and my fps went up in all the games i play.
>>54813077
It will end with their bankrupt.
>>54815774
If you repeat that for 20 years it will come true.
You still got 14 more to go.
>>54812900
>itt kids who don´t know shit about tech
nvidia has more power because they go the way of bigger die(chips) their chips are always bigger than amds which gives them more space to work with.
Titan X for example is the biggest gpu chip ever made for consumers.
If nvidia and amd would have same die size amd would beat nvidia easy.
That´s also why nvidia cards are more expensive than amd cards.
Bigger chip = more failures and less chips per die so costs skyrocket. Btw big chips also mean that they can spread heat better which is one of the secrets nvidia has lower temps at least on maxwell.
Pascal is a housefire.
>>54815701
Quadros and FirePros have historically always been in the same segment. Nvidia opted not to use a chip with good double precision for the M series specifically and their reasoning was sound - Tesla exists for double precision.
I didn't see much of a reason to compare Tesla to AMDs shit, there is no point.
Just for your info during the K series when Quadro still had double precision it beat FirePros soundly in that as well.
So back to the original argument of OP - it's still wrong.
>>54815740
sony vegas and adobe premier both use more then one core, the codecs they have are another issue altogether, with some being unusable, some being very fast but almost lossless in size or even incompatible with most video sharing sites.
>>54815806
>Pascal is a housefire.
Pascal has aftermarket cooler 390 temps with a blower.
So it's safe to say it's nowhere near AMDs forest fire gpus.
>>54815806
Stop talking you fuckstick, AMD has a TSMC reticle limit GPU out as well, it's called Fiji and it's between GTX1080 and GTX980Ti in specs on paper.
But it doesn't perform like that.
Same thing for smaller GPUs, GM204 is also weaker on paper than Hawaii, this is an outlier though, they perform really similarly.
>>54815825
Instead of comparing cards with similar characteristics and performance segment/market you can't compare whatever the fuck you please and expect others not to take a shit:
GM200, Fiji, same performance bracket, same characteristics, die size and TDP, both can't do shit worth DP but excel at SP.
Most importantly, they're both from the same generation of hardware.
They're aimed at the same market, they can easily be compared, Fiji is faster on paper in pretty much 5 out of 7 measurements, but realistically it's neck-to-neck with GM200 depending on a bunch of factors.
Then you can compare GM204 to Hawaii, GM206 to Tonga, every time AMD will have a more powerful card in pure specs.
>>54815915
Not at the right time.
And only in specs.
I already told you assburger - my management doesn't care. It needs the performance and it needs it NOW. Not on paper. Not in an year. Not when AMD gets off their asses to write a non shit driver. NOW.
That's why the market share of Nvidia is even better in the pro sector. Which basically means they are a monopoly at this point. There is only AMD to blame for that, not the other way around, it's only natural for the consumer to choose the better product.
>>54815842
Pascal is 85°C and throttles itself, 390 doesn´t.
>we called amd housefire because they hit high temps with standard blowers, but now its ok because its nvidia. Top kek nvidiacucks
>>54815976
What your management needs and what your boss tells you is not my problem or the point of this fucking thread or discussion.
Take your frustrations elsewhere, we're talking about hardware, not your underpaid job.
>>54815976
top kek merica management failures at work. thats why americas managment is shit, everything is good aslong it gives shortterm wins
pajeet and sjw hipster gayum programmurrrs happened
>>54815991
You're talking about hardware you've never used and never will use. Armchair hardware analysts are fucking hilarious.
>>54816002
AMDs drivers for OpenGL and OpenCL have been shit for at least a good 6 years now. In my industry that's a lifetime.
Your long term "wins" only exist in gaymen and quite honestly this is one of the reasons AMD is doing jackshit in everything besides gaming consoles - game engine engineers are lazy and don't upgrade their shit more than once every 4 years. Not something to be proud or happy about.
>>54815982
>390 doesn´t.
Yeah. Sure.
>>54815982
Throttles? I have two 1080s and I play generally at 85-80c. and notice no throttling.. where do you read this throttling information? I don't see it in reality (I have FPS counter on in most games, if no counter, then Afterburner or EVGA thingy)
>>54816068
and what you posted says what exactly?
>>54816047
I don't need to use some hardware to read what it says on a piece of digital paper or calculate ALU*nominal clock*ops per clock
You're grasping at straws, what you do, your job and your boss has nothing to do with anything in this thread, the discussion is how AMD has more powerful specs on paper and why their cards can't fully utilize the hardware inside, not how you aren't able to buy GPU X or Y at a certain time period nor how Nvidia has higher marketshare.
>>54816090
>the discussion is how AMD has more powerful specs on paper and why their cards can't fully utilize the hardware inside
At least in video games AMD seems to come up with (recently, Fallout4 for example) better performance if you care to compare post-game launch statistics and drivers.
>>54816084
The 390 runs at 85C+ and doesn't' vent it's exhaust properly because of the retarded PCB design even with an aftermarket cooler.
This is a Strix, so one of the most expensive 390s out there.
>>54816098
Yeah, AMD is slower with software support, that's nothing new.
Some games don't even need that much software support from AMD when the engine plays to AMD's hardware strengths, but that's very rare.
>>54816090
>what it says on a piece of digital paper
No one cares what it says on a piece of paper.
Pretty much everyone cares what the actual performance is. Isn't that what you gaymers do when you choose your cards?
Why is it that the paper stats are important for AMD when it should be stronger, but irrelevant when it's doing better than an Nvidia card that should be better on paper?
Because you're fanboys that's why.
>>54816133
Hey hey can we stop with the attacks and ad hominems and discuss properly!
>>54816099
the guy said the 1080 throttles at 85c note that the 390 doesn't get to 85c
>>54816143
>note that the 390 doesn't get to 85c
It says 85.5 in plain numbers right there in the screenshot.
>>54816099
Depends on the model apparently.
http://www.techpowerup.com/reviews/Sapphire/R9_290_Vapor-X/29.html
http://www.techpowerup.com/reviews/Sapphire/R9_390_Nitro/27.html
Here, 66-77C
>>54816133
And no one cares about your disinterest in hardware either, I like talking about hardware, and I'm sure plenty others do, it's been told 20 times in the thread that AMD cards don't produce the performance their specs advertise, and explaining WHY is the reason for the thread.
I know you like skimming through benchmark sites for 5 minutes and making up your mind, but I don't.
People have different interests, whovudathunk
>>54812900
AMD is not working with game devs as active as nvidia do, thus their technologies are not implemented in most scenarios. Nvidia is also faster and better at tuning their drivers for the most recent card models for every single AAA game.
Nvidia does give a shit on how well the games are going to perform on their latest cards, while AMD don't.
>>54812979
And what about at pure computation? Why spend all that money on CUDA developers when OpenGL can use more processing power? inb4 it's more complicated.
>>54813077
>Using the wrong insult
>Not using AMDrones
>Using cuck like a retard
Literarally kill yourself if you cut shitposting right
>>54816252
It's OpenCL, not OpenGL.
Secondly, CUDA has been deeply entrenched with a fantastic toolkit for a very long time, it's the same reason why XP is still used in some business, managers don't want change.
![[NoobSubs] Monogatari S2 09 (1080p Blu-ray 8bit AAC).mp4_snapshot_15.32_[2016.01.31_01.52.30].jpg](https://i.imgur.com/eDZ6zUe.jpg "[NoobSubs] Monogatari S2 09 (1080p Blu-ray 8bit AAC).mp4_snapshot_15.32_[2016.01.31_01.52.30].jpg")
Oh AMD's got FLOPS alright
nVidia is literally the Apple of VGA cards. It's a superior option provided that you are willing to pay for the "premium" tax.
>>54816252
I don't have an explanation besides the fact a lot of professional applications use CUDA and in some cases ONLY CUDA.
I highly doubt it's Nvidia paying off developers (they have a better OpenGL performance as well) because even open/freetard devs are using CUDA preferentially now. I suspects it's simply easier to dev for.
AMD pays extra in silicon to compete with Nvidia, this is not long term sustainable.
If AMD has the same driver quality as Nvidia, they'd be enjoying a decent lead, but they don't.
I can think of two choices:
1. Produce much better hardware but stick with the high overhead drivers, short term gain, long term loss.
2. Rewrite the driver stack into something more sanitized and sane, this is a large short term loss, because it requires time and a lot of money, but in long term it would pay out.
Any normal company would go for the latter, the problem is that AMD doesn't have that much money to do the latter.
>>54816078
No one cares about your worthless opinions when nvidia and every "reviewer" says it throttles.
>>54816099
The strix is fucking awful for AMD. In fact every asus card is terrible for AMD.
>>54816515
I just want to read the review(s) that say it throttles since I only see throttling at above 90..
and it's hardly an opinion when I have the cards running and FPS counter logging information.
>>54816099
>AMD
>ASUS
You're in sapphire/xfx territory son.
>>54816540
learn to google then you fucking retard
>>54816068
>>54816099
still waiting for you to tell me were the 390 throttles like Pascal housefires
>>54816220
That's not quite true, they don't have the resources to do that. I'm sure they care, otherwise they'd be retarded.
>>54816598
spoiler alert: It doesn't.
>>54814257
>1.7% Woodscrews
Oh god it's like I'm really back in the old /g/
why do people get angry about graphics cards around here
>>54817047
You mean the good /g/?
Where did that /g/ go?
>>54816068
>Asus Strix cooler
>well known to be the worst aftermarket for AMD because they litteraly strapped the 980's cooler on and called it a day.
>that article and review is about how bad asus is, not the 390X
>Everyone knows this but you
>>54817085
Everyhing makes /g/ angry. Doesn't matter whether it's about CPUs, GPUs or the keyboard travel distance.
>>54817122
>keyboard travel distance
umm?
>>54817092
Everyone's house burned down.
>>54817146
https://mechanicalkeyboards.com/terms.php
Don't let the autism get to you.
>>54817155
If by house you mean dedicated nVidia coolant structure then I agree.
>>54817171
Jesus h fucking christ how can someone get angry over that
>>54816528
why the fuck do they do this
>>54817187
>>54799490
Take a look. Most /g/ threads end up being shitposting central after about 40 posts.
>>54817206
Save costs by not having to remachine the cooler. The 290/x versions of asus coolers were particularly horrible because kepler is a bigger die (hence why those two heatpipes don't make contact on hawaii).
When is announcement??
>>54817455
The event will begin on Wednesday, June 1, 2016 at 10:00 AM CST / 10:00 PM EDT.
Translate that to GMT and I'll tel you
>>54817482
In an hour or two I think.
>>54816099
>ASUS Strix
>AMD
Picked a cooler that was literally designed for Nvidia and slapped on an AMD card. Pick a cooler from a more reputable company next time.
>>54817620
If AMD wasn't so shitty shit it wouldn't even happen.
>>54817628
How is AMD responsible for what third party companies do?
>>54817649
Buttshot MADcuck detected.
>>54817660
>>54817660
You tried
>>54812900
Nvidia gameworks
>>54813683
objectively the best post in the thread
thread over
>>54817909
+1
Spread the word AMD is fucking unredeemable putrid shit
>>54815390
saving this because i get the feeling you might be right.
>>54815390
>>54819206
That would be retarded Polaris 10 is the 480/X and 470/X
11 is the 460 and 450 cards
>>54815634
You're literally retarded and I say this without the least bit of irony.
>>54819206
You're not right.
There's Polaris/Vegas 10/11
Polaris 10 = 480(x)
Polaris 11 = 470(x)
Vega 10 = Fury(x)
Vega 11 = 490(x)
There are 8 models lined up for AMD's new GPU. So 470,470X, 480,480X,490,490X,Fury,FuryX
>>54815534
Do you work in one of those hipster startups?
>>54813558
What a meme! Oh no she di'n't! Well le memed, m'sir!
>>54812900
>>54815379
What a fucking retard
>>54813726
Does that monitor look 1440p to you? Its a lie.
>>54816068
How does it feel to be the most illiterate retard on this board right now, anon?
>>54819358
this is correct
>>54820783
>Its a lie.
Why don't you inquire instead of calling everything you don't understand a lie.
Virtual Super Resolution:
http://www.amd.com/en-us/innovations/software-technologies/technologies-gaming/vsr
>>54821108
It is a lie, it's not the same as a monitor running 1440p 144hz natively.
Why the fuck would they not just put a monitor capable of 1440p 144hz natively there?
It smells like manipulation
>>54821127
>It is a lie, it's not the same as a monitor running 1440p 144hz natively.
Not him but the GPU ends up working just as hard. Also, anyone know what make and model that monitor is? Would like to see the specs for myself.
>>54821127
>1440p 144hz
You are aware the card will push out 1440p to a 1080 monitor? Both AMD and Nvidia so this. This is more taxing on the card and is done "to simulate SSAA in games that do not have native SSAA support" I don't think you don't know this and you're just another uneducated person on /g
>>54816322
i never heard of xnormal before, why does it use cuda?
