Vice President and General Manager of AMD Radeon Gaming accuses NVIDIA GPP of monopolistic and anti competitive practices

[Sakkura]

2 hours ago, mr moose said:

Because basically consumers don't care.  Educated consumers and average Joe's will buy the product that best serves their needs regardless of how crappy the company is.   I can't think of a single company today that has suffered a drop in sales due to unethical behaviour (and there have been a few).   Fanboys are a minority, just like boycotts. 

Consumer boycotts do work once in a while. But it takes a lot to get enough people involved.

[79wjd]

1 minute ago, Sakkura said:

Consumer boycotts do work once in a while. But it takes a lot to get enough people involved.

A company has to do something so horrendously horrible for them to work.....The fact is that its virtually always in your best interest to prioritize your own interests (e.g. buy the thing that suits your needs best regardless of everything else) -- game theory strikes again. 

[leadeater]

7 minutes ago, Razor01 said:

Well the reason why I would have gotten a Windows phone would be different it would be something that I can directly sync up with my pc and have no issues with, well if MS does their work right lol.

 

I don't use my phone for stuff much outside videos, some games, apps wise is not a big deal for me. 

Yea setting up Exchange mail sync at the time was just so much better than iOS, also Skype for Business just works and just doesn't on iOS. Work phone is an iPhone but I never use it.

[leadeater]

5 minutes ago, djdwosk97 said:

A company has to do something so horrendously horrible for them to work.....The fact is that its virtually always in your best interest to prioritize your own interests (e.g. buy the thing that suits your needs best regardless of everything else) -- game theory strikes again. 

When that is the case though it's usually not the boycott that instigates the change, it's the bad press and the fear of the backlash that preempts any customer action so it's rather hard to tell if the boycott would have worked.

[79wjd]

1 minute ago, leadeater said:

When that is the case though it's usually not the boycott that instigates the change, it's the bad press and the fear of the backlash that preempts any customer action so it's rather hard to tell if the boycott would have worked.

Probably not, consumers have short attention spans. 

[Arika]

12 minutes ago, leadeater said:

You can't customize a damn thing on Windows phone either, nothing exists for it lol. Nokia 635 user btw lol.

 

P.S. I do actually like Windows Phone as I only use it for emails, txt, calls and hotspot. I don't give a damn about apps on a phone.

well to be fair there's a difference between a locked down OS and an OS with basically no worthwhile features

[mr moose]

3 hours ago, wasab said:

Problem is that what nvidia is doing is hindering the consumers from  switching brands. 

In what way?
 

3 hours ago, Sakkura said:

Consumer boycotts do work once in a while. But it takes a lot to get enough people involved.

Very rarely and usually they have to be heavily organised and the target business has to be very small.  One of the biggest boycotts in history was the Nestle one in the 80's/90's,  They only grew bigger.  

[laminutederire]

8 hours ago, Razor01 said:

 

Then live in your fucking monopoly and don't try to change anything then. I don't really know what more to say to you.  You dare to show me results based on a gtx 260? Seriously?

Then you get some graph out of context. Yeah the Titan X which is already faster than a r9 fury is still faster. What a surprise. That's the first thing. The second thing, which is said in the article and that you do not mention at all, is that with just a few optimizations, the fury performed 4-5 times better.

CUDA based cards may have some additional perks for deep learning, but the most important thing is that code running in deep learning is optimized for CUDA only in most benchmarks you showed. Starting to optimize those for OpenCL based architecture show huge improvement with just a few tweaks. And that's the important thing. Right now Apple to Apple comparison of CUDA vs OpenCL is not that relevant since OpenCL is in optimization phase which can yield 4-5x improvements, and I'm not sure that performance comparison on Nvidia cards only is fair, since we can't compare if drivers for CUDA and OpenCL are as optimized each. What was shown was that memory utilization tanks the performance if you do not use the natural way of handling memory on each architecture.

 

So basically what you are saying is CUDA beats OpenCL and that can't change. When there is evidence that OpenCL is slower now, but there is evidence as well that small tweaks to actually make them adapted to different architecture yield actually huge improvement that allow to catch up a bit with CUDA. So with Intel's push as well, we will have a baseline of CUDA optimized vs OpenCL optimized, and it should be around the same perf for most tasks. That can be seen for instance with JuliaGPU benchmarks over at phoronix for instance.

Have some perspective of what the future could be instead of looking at the past.

 

Good for you, i still wouldn't trust you on those simply because the way it should be seen is that there are a few probable outcomes, and not just one like you seem to suggest. And by the way, you display the "enthusiast" arrogance just now. "I am never wrong so I'm not wrong on that". One of the most important thing in life to learn is that you're never right all the time, so expression your view should be expressed as your view and not a universal truth.

 

[Humbug]

7 hours ago, mr moose said:

Very rarely and usually they have to be heavily organised and the target business has to be very small.  

The negative publicity and partial boycott can absolutely work. Unless by 'work' if you mean to make half of Nvidia's customers turn away and make Nvidia take a huge financial hit from lost sales in the short term... Nope that's not gonna happen.

 

But what it can very realistically do is swing 15% of the next purchasing decisions of traditional Nvidia customers. And that's a very big deal, it boosts AMD's profits, it gets a few more people into AMD's ecosystem which makes them likely to stay and repeat purchase etc.

 

The negative PR against Nvidia  absolutely does benefit AMD. Anyway AMD has no delusions about getting 40-50% discrete GPU market share overnight. They know it's not gonna happen now. But what they do want to do is continue to grow their market share and increase their profits. So every percentage point of market share matters.

 

All this bad publicity for sure turns off a small Chunk of Nvidia buyers. And AMD will happily welcome them. It matters a lot as AMD continues to become a stronger company...

 

That's the reason AMD tries to use this negative PR against Nvidia. They know it will reach a certain chunk of the buyers at least. And it costs no marketing money to send out a few tweets and expose the story to the press. Absolutely worth doing for them. 

 

Another example is all the negative publicity Nvidia has taken for gameworks and not being successful when they work with devs on graphical optimization, or associating their name with the wrong games...  None of this stuff is an absolute disaster for Nvidia. But it would be foolish to think it doesn't matter at all, it does matter and the PR from it does benefit AMD. It's just one more factor in a complicated market. 

[Humbug]

^^now somebody can read my above post and reply with 'AMD should focus on putting out better products' or something like that LOL. As if they don't have any good products or as if a company should halt everything else they do while they focus on NPD, as if other market factors do not matter...

[Razor01]

2 hours ago, laminutederire said:

Then live in your fucking monopoly and don't try to change anything then. I don't really know what more to say to you.  You dare to show me results based on a gtx 260? Seriously?

Then you get some graph out of context. Yeah the Titan X which is already faster than a r9 fury is still faster. What a surprise. That's the first thing. The second thing, which is said in the article and that you do not mention at all, is that with just a few optimizations, the fury performed 4-5 times better.

CUDA based cards may have some additional perks for deep learning, but the most important thing is that code running in deep learning is optimized for CUDA only in most benchmarks you showed. Starting to optimize those for OpenCL based architecture show huge improvement with just a few tweaks. And that's the important thing. Right now Apple to Apple comparison of CUDA vs OpenCL is not that relevant since OpenCL is in optimization phase which can yield 4-5x improvements, and I'm not sure that performance comparison on Nvidia cards only is fair, since we can't compare if drivers for CUDA and OpenCL are as optimized each. What was shown was that memory utilization tanks the performance if you do not use the natural way of handling memory on each architecture.

 

So basically what you are saying is CUDA beats OpenCL and that can't change. When there is evidence that OpenCL is slower now, but there is evidence as well that small tweaks to actually make them adapted to different architecture yield actually huge improvement that allow to catch up a bit with CUDA. So with Intel's push as well, we will have a baseline of CUDA optimized vs OpenCL optimized, and it should be around the same perf for most tasks. That can be seen for instance with JuliaGPU benchmarks over at phoronix for instance.

Have some perspective of what the future could be instead of looking at the past.

 

Good for you, i still wouldn't trust you on those simply because the way it should be seen is that there are a few probable outcomes, and not just one like you seem to suggest. And by the way, you display the "enthusiast" arrogance just now. "I am never wrong so I'm not wrong on that". One of the most important thing in life to learn is that you're never right all the time, so expression your view should be expressed as your view and not a universal truth.

 

 

Ah got ya you didn't even read what I linked, I purposefully stated Titan X had more flops, this Titan X was Maxwell, Not Pascal lol, it had 20% less flops than the Fiji lol

 

I knew you were just making things up.  It doesn't matter what people say to you because you are in a land of fairy tales and pixel dust and only see what you want to see.

 

Cuda Based cards have additional based perks you say lol

 

Like what?
 

Quote

 

Compared to the highly optimized machine learning cuDNN library, OpenCL Caffe still has a performance gap of 2x as it lacks those optimizations. The authors argue given the current performance, the OpenCL caffe is still competitive in terms of performance per dollar, considering the market price difference of AMD R9 Fury (about 560 dollars) and the NVIDIA TitanX (about 1000 dollars).

 

 

 

Yeah what did I say about extenstions that make things go faster on CUDA lol, Open CL libs don't have them!

 

OMG, I was right lol.  What did I say?

 

Quote

 

Just like you thought CUDA is bad because its closed, I explained that to you before, why CUDA itself isn't bad, lets go into this deeper,  first off CUDA just exposes features of nV cards (that is all an API does, any API), some of those features are not even there on AMD cards that is why Open CL is behind CUDA and also why programs are hard to port over, many of those enhancements are done for speed purposes for specific techniques, which AMD's hardware is not equipped to handle.  This is why we see nV cards with less Tflops still keeping up with AMD cards among other things as well.

 

How dare I, is this some atomic bomb laying ground where you want to change the history of what happened?

 

Nonsense, you don't even read what I typed, and make up fanciful crap to justify what I stated was wrong.

 

Next time actually do some research before you post and actually read what I post before you reply.

 

Now the rest of your rant, I don't want a monopoly, but its already virtually a monopoly as it is, and I don't see AMD changing in the graphics industry any time soon, so there ya go, take your emotions out of it and analyze the situation AMD is in and then post please.

[laminutederire]

35 minutes ago, Razor01 said:

 

You don't seem to have a clue of how software works though. I give up you're useless and you won't listen even when I present a nuanced view of the situation using the publication you gave me. You re the one not reading, and to be fair I have something else to do than trying to convince someone stubborn.

(Learn maths as well please, the fury being around 7 TFLOPS and the Titan X Maxwell being around 6.5, that's not at all 20 percent, it's not even 10 percent. Unless all reviewers are wrong in my conspiracy theory, huh? )

[Razor01]

28 minutes ago, laminutederire said:

You don't seem to have a clue of how software works though. I give up you're useless and you won't listen even when I present a nuanced view of the situation using the publication you gave me. You re the one not reading, and to be fair I have something else to do than trying to convince someone stubborn.

(Learn maths as well please, the fury being around 7 TFLOPS and the Titan X Maxwell being around 6.5, that's not at all 20 percent, it's not even 10 percent. Unless all reviewers are wrong in my conspiracy theory, huh? )

 

 

Really Fiji was 8.6 tflops, 6.5 tflops.

 

Hmm 2.1

 

What is 2.1/6.5?

 

Oh that is right that is its 32% or (24% less that FuryX), sorry I got that wrong on the benefit of AMD's side lol.

 

Look man you can pull fury X tflop numbers out of thin air of 7  if I remember correctly even the Nano wasn't even 7 tflops lol, what else can you do?  Talk about tipping scales in ones favor, just making up numbers now.

 

You need reviewers to do your work for you that you are so emotionally invested in?  OK good one

 

[79wjd]

4 minutes ago, Razor01 said:

 

 

Really Fiji was 8.6 tflops, 6.5 tflops.

 

Hmm 2.1

 

What is 2.1/6.5?

 

Oh that is right that is its 32% or (24% less that FuryX), sorry I got that wrong on the benefit of AMD's side lol.

 

Look man you can pull fury X tflop numbers out of thin air of 7  if I remember correctly even the Nano wasn't even 7 tflops lol, what else can you do?  Talk about tipping scales in ones favor, just making up numbers now.

 

You need reviewers to do your work for you that you are so emotionally invested in?  OK good one

 

The source you linked is talking about a Fury, not a Fury X. 

[Razor01]

7 minutes ago, djdwosk97 said:

The source you linked is talking about a Fury, not a Fury X. 

 

FuryX doesn't get much more performance over the Fury in these tests. 5 to 10% difference, and that doesn't account for the 300% difference in performance of the Titan X.

 

It all comes down to the optimizations of the open CL libs, which this was more than 2 years ago and they STILL haven't been done!, that tells me there is something more to it than just the library, AMD and Intel are both proponents of HSA, and Open CL, but neither of them are going to push low hanging fruit like a 300% increase in performance form an few extensions?  Now Direct Compute with SM 6.0 on the other had does have those extensions, I am not familiar about direct compute as much, but I don't know if those are cap bit add ons for just specific cards or not, I am presuming so because we have not seen changes in software for AMD and Intel hardware for these either, while Intel dropped making Phi altogether so we will never know now till they come back in to the GPU market.  But then again this won't be an open source API so @laminutederire will have a problem with this too.

 

How badly do these companies want into the HPC and DL markets?  Not at all then?  They don't give a shit apparently by their actions right?

[79wjd]

3 minutes ago, Razor01 said:

 

FuryX doesn't get much more performance over the Fury in these tests. 5 to 10% difference, and that doesn't account for the 300% difference in performance of the Titan X.

 

I don't care about the rest as I don't know enough about the market to comment there, I'm just pointing out that the source is using a Fury and not a Fury X -- a Fury, which has only 7 tflops compared to the 8.5 of the Fury X.

[Razor01]

20 minutes ago, djdwosk97 said:

I don't care about the rest as I don't know enough about the market to comment there, I'm just pointing out that the source is using a Fury and not a Fury X -- a Fury, which has only 7 tflops compared to the 8.5 of the Fury X.

Doesn't take from the fact that those optimizations are not present for AMD cards via extensions.  That was my original statement.

 

It also doesn't take away from what I stated, Open source isn't better than closed source, both have their benefits and downfalls, just because @laminutederire doesn't understand those differences, doesn't mean closed source is bad for the industry.  To the contrary, it has moved the industry forward and increased competition by diversifying the market, but now its gotten to an extent that the open source API can probably no longer compete and even with heavy investing, much more than nV has done with CUDA, might not be able to.

 

This is the same thing MS did with Windows.  Has Windows and all of MS's proprietary software and API's hurt the computer industry?

 

Does anyone remember Open GL and how it was the standard for graphics API?  What happened with DX9?  MS took that way from the Open GL.  As MS kept innovating and setting the standard for graphics API's, Open GL just got messy and weren't able to keep up.  Finally Vulkan came out and now its adoption rate is even slower than Open GL's adoption rate.  This is the whole story behind open source vs closed source.  If the parent company that has hardware supporting such open initiatives doesn't invest money and resources into driving the project, no other company will ever jump on board and try to fill those shoes, because they have no vested interesting in a project like that.  So AMD and Intel in due time (due time is now, because if they are going to have a GPU in 2 or 3 years time, its best to start now) needs to get off its ass and do something, if its not too late already.

[leadeater]

47 minutes ago, laminutederire said:

You don't seem to have a clue of how software works though. I give up you're useless and you won't listen even when I present a nuanced view of the situation using the publication you gave me. You re the one not reading, and to be fair I have something else to do than trying to convince someone stubborn.

(Learn maths as well please, the fury being around 7 TFLOPS and the Titan X Maxwell being around 6.5, that's not at all 20 percent, it's not even 10 percent. Unless all reviewers are wrong in my conspiracy theory, huh? )

Caffe is about the worst thing to use to compare OpenCL vs CUDA considering Caffe was explicitly developed for CUDA, it's native vs non-native comparison so it doesn't take a genius to figure out which will produce better performance.

 

Best to leave this conversation chain alone though, it's mightily off topic.

[Razor01]

3 minutes ago, leadeater said:

Caffe is about the worst thing to use to compare OpenCL vs CUDA considering Caffe was explicitly developed for CUDA, it's native vs non-native comparison so it doesn't take a genius to figure out which will produce better performance.

 

Best to leave this conversation chain alone though, it's mightily off topic.

They also tested clBlas, that is just a math library which was made for Open CL to begin with

[AlTech]

46 minutes ago, Razor01 said:

 

 

Really Fiji was 8.6 tflops, 6.5 tflops.

 

Hmm 2.1

 

What is 2.1/6.5?

 

Oh that is right that is its 32% or (24% less that FuryX), sorry I got that wrong on the benefit of AMD's side lol.

 

Look man you can pull fury X tflop numbers out of thin air of 7  if I remember correctly even the Nano wasn't even 7 tflops lol, what else can you do?  Talk about tipping scales in ones favor, just making up numbers now.

 

You need reviewers to do your work for you that you are so emotionally invested in?  OK good one

 

Teraflops don't matter. Real world performance matters.

[Razor01]

9 minutes ago, AluminiumTech said:

Teraflops don't matter. Real world performance matters.

Yeah that is true and nV kicks AMD's ass in that

 

It comes down to API and software implementation

 

https://liu.diva-portal.org/smash/get/diva2:909410/FULLTEXT01.pdf

 

While everyone was looking at what nV and AMD doing hardware wise, nV also was looking at what they can do software along with hardware wise to corner the market as a whole.

 

And they did just that for the past 10 years or so.  Invested tons of money in growing software and designed hardware to push that software development to give them a strangle hold

 

This is why Intel is even having tough times getting into the HPC/DL market.  Intel has the resources but too little too late comes to mind.

 

nV did this by innovating with hard work, money, and resources (which actually was a large risk for them but that risk is now paying off), and for them to just hand it over to a competitor is against our "free market" philosophies.  Not to mention our own laws for IP. 

 

The only way I can see this changing as of now is if nV slows down and do an Intel.

 

[leadeater]

20 minutes ago, Razor01 said:

This is why Intel is even having tough times getting into the HPC/DL market.  Intel has the resources but too little too late comes to mind.

They are getting rather decent share now though, 4 of the top 10 super computers are Xeon Phi accelerated, or 21 of the top 100, or 30 of 500.

[Razor01]

Yeah but that was because the latest Phi came out before the competing Tesla products, and now Intel dropped Phi altogether because they saw they can't keep going with their current architecture and remain competitive.

[KarathKasun]

4 hours ago, Razor01 said:

Yeah that is true and nV kicks AMD's ass in that

Actually, during the time period you are talking about, no.

Even now, Vega has some datacenter and workstation workloads where it is ahead of the Titan XP2.

[Razor01]

4 minutes ago, KarathKasun said:

Actually, during the time period you are talking about, no.

Even now, Vega has some datacenter and workstation workloads where it is ahead of the Titan XP2.

Like?  if you are talking about AMD's slides, sorry those were loaded