AMD goes fermi

[notty22]

I found a old AMD slide before AMD went Fermi, LOL

http://www.siliconmadness.com/2009/10/amd-prs-counter-paper-launch-with.html

 

[taltamir]

I hope so, my only consern is that AMD has no ECO system, unlike what CUDA has.

CUDA is just native C code and fortran running ability on GPU. All AMD has to do is make their GPGPU run native C code and fortran and all the GPGPU applications for fermi could be easily recompiled for it. And due to the extreme similarity of the architecture it should also utilize the same performance optimizations out of the box.

The question is, where is the C code and fortran compilers for GCN.

Also whether I am correct about the assertion that the two architectures are very similar. If I am wrong then they will have an optimization issue. Can any expert chime in with their opinion on the similarity or lack thereof of the two architectures?

So far only one person other then me even touched that issue

They aren't. GCN still favors high throughput over fast complex memory access, compared to Fermi

Which I would love some clarification on from someone more knowledgeable.

 

[Cerb]

Nope, 32nm IS the 28nm process. 32nm was so heavily delayed they changed the name to 28nm process, as the method used to measure this is subjective and not absolute.

The official word, at least, went like this: 32nm was going to be awesome, then go into 28nm, and so on; then 32nm got canceled, with more focus on trying to make the 28nm that was already in the pipe work, instead.

Either way, where was AMD at 65/55 and 45/40? NVidia was adding GPGPU features, and creating a nice set of markets for themselves, while AMD was blowing hot air, and praying that companies might start developing with OpenCL, before they even had support bundled in their drivers. Hopefully RR will show enough leadership that they won't be left adrift (or in a permanent rut, like VIA's x86 businesss). Intel also tends to act that way, too, developing base technology, and then basically just leaving it there to do whatever with. They have had good opportunities to mimic NV in that regard, and hopefully they will. They don't just need to have good hardware, but need to lead the market towards making use of it.

Which I would love some clarification on from someone more knowledgeable.

I swear I had something good about their memory management bookmarked, but I'll be damned if I can find it now. For ow, <fake-accented southern lawyer in a movie>I retract my statement</fake-accented southern lawyer from a movie>.

 

[Eisenstein]

Yeah AMD bitched a lot about Fermi being compute and now they're adopting it 🙂 It's clear the performance hit AMD has suffered with 7970 (I mean 10&#37;-18% total average better performance than GTX 580 at 1080p and 1600p res) is because of this move. but NVIDIA has already suffered through compute so I'd expect Kepler to be faster.

 

[taltamir]

Yeah AMD bitched a lot about Fermi being compute and now they're adopting it 🙂 It's clear the performance hit AMD has suffered with 7970 (I mean 10%-18% total average better performance than GTX 580 at 1080p and 1600p res) is because of this move. but NVIDIA has already suffered through compute so I'd expect Kepler to be faster.

what is the current latest rumor for kepler ETA btw?

 

[NIGELG]

Kepler is not available now to the public or anytime soon.Maybe Midyear.

 

[MrK6]

I found a old AMD slide before AMD went Fermi, LOL

http://www.siliconmadness.com/2009/10/amd-prs-counter-paper-launch-with.html

Where on that slide does it say "GPGPU is stupid and we're never going to do it." Please, point it out, because I see a slide that says:

When compared to the ATI Radeon 5870:
~40% more transitors
~60% bigger die area

Given that the slide was made for 10/13/2009, that's pretty damn accurate info. I think the GTX 480 was actually 63% bigger, so they undershot it.

Large chip design primarily targeted for HPC applications
-No Graphics Information
-Not ready for production

Again, both of these points are correct - a large portion of the chip is dedicated to compute work, no graphics production, and it wasn't read for production on 10/13/2009. In fact, it wouldn't be ready until 5 1/2 months later. So again, AMD is correct here.

Targeted for an emerging niche supercomputing market
-ECC memory support
-Double precision support
-Costly for Graphics and Consumer Market

Once again, all these statements are correct, as the GTX 480 was expensive, hot, and loud, but only 10% faster than the 5870.

So what was the point you were trying to make? What was so funny about the slide. Please, explain to the forum the point you were trying to make since you thought it was worth the forum's time.

 

[Keysplayr]

So what was the point you were trying to make? What was so funny about the slide. Please, explain to the forum the point you were trying to make since you thought it was worth the forum's time.

You might consider the possibility that what he thought was funny, was perhaps that the once/long touted small die strategy may not have been the right one, as seen by AMD's entry into the foray of GPGPU computing with the 7xxx series. I of course could be wrong, I'm not Notty, but it seems a simple logical train of thought brings us there.

-Keys

 

[MrK6]

You might consider the possibility that what he thought was funny, was perhaps that the once/long touted small die strategy may not have been the right one, as seen by AMD's entry into the foray of GPGPU computing with the 7xxx series. I of course could be wrong, I'm not Notty, but it seems a simple logical train of thought brings us there.

-Keys

You'd think he'd post one of the hundreds of articles or slides about the "small die strategy being king" if that was the case, wouldn't you? That's where the simple logical train comes from, not the "failed troll" train we're on now. Although it figures you'd jump in to the defense though, way to go Keys. 🙄

 

[Chiropteran]

At the moment, yes. But you have to realize that 7970 is AMD's 1st GPGPU architecture, while Kepler will be NV's 2nd. HD7970 came out on 28nm, not on 40nm. The only way to fairly compare the efficiency of both architectures is to place them on equal nodes. When Kepler arrives, we'll be able to do that. For now, we are comparing a 2-year-old Fermi architecture on 40nm to a brand new GCN architecture on 28nm. It was obviously expected that GCN would be far faster than GF100/110 in compute and in games. The question is, did AMD do enough with 7970 to position it well against NV's 28nm GPU? We'll find out in 2012. Exciting times ahead.

Fair way to compare? So I guess we should only compare AMD CPUs to 2 year old intel chips, since the new intel chips are on a newer node? No, obviously not. You compare the best available to the best available.

There is no fair in business, products compete as soon as they are released, and if nvidia doesn't release it's next GPU for 6 months or a year it'll suffer in the meantime, that is nvidia's fault not AMD.

 

[Keysplayr]

You'd think he'd post one of the hundreds of articles or slides about the "small die strategy being king" if that was the case, wouldn't you? That's where the simple logical train comes from, not the "failed troll" train we're on now. Although it figures you'd jump in to the defense though, way to go Keys. 🙄

Hey, ease up there my friend. I just offered a point of view.

 

[notty22]

Don't Explain the Joke

In short, explaining the punchline of a joke just makes it not funny, whether or not it would be otherwise. Jokes can be hard to do, and sometimes not everyone will get it, but while explaining the context might help, the punchline should stand on its own.

 

[tincart]

I am confused. Are some people saying that since AMD, at one time, adopted the "small die strategy", they cannot change their strategy without being inconsistent?

Perhaps the small die strategy was once a very convincing approach to GPU design, and now the situation has changed and a different approach is warranted? That would be, y'know, a fairly responsible thing to do: Change your strategy given the market conditions and potential sources of revenue (and diminishing opportunities from past sources).

 

[Arkadrel]

I am confused. Are some people saying that since AMD, at one time, adopted the "small die strategy", they cannot change their strategy without being inconsistent?

Perhaps the small die strategy was once a very convincing approach to GPU design, and now the situation has changed and a different approach is warranted? That would be, y'know, a fairly responsible thing to do: Change your strategy given the market conditions and potential sources of revenue (and diminishing opportunities from past sources).

my gawd man... that almost sounds like logical thinking!
Prepaire to get trolled! and flamed for that post.

 

[Keysplayr]

I am confused. Are some people saying that since AMD, at one time, adopted the "small die strategy", they cannot change their strategy without being inconsistent?

Perhaps the small die strategy was once a very convincing approach to GPU design, and now the situation has changed and a different approach is warranted? That would be, y'know, a fairly responsible thing to do: Change your strategy given the market conditions and potential sources of revenue (and diminishing opportunities from past sources).

In all honesty, I think AMD is only turning onto a road that has existed for a while now. I don't know that anything has "changed" as you put it.
And to try to reduce your confusion, as you say, I think that there isn't anything wrong with adopting a new strategy when an old one doesn't pan out.
The "funny" part might only be a rub at the people who in the past might have poked fun at Nvidia for large dies geared toward GPGPU. People would then say, "Look how much smaller AMD's dies are compared to Nvidia's, yet still seem to compete with them." This slogan was used quite a bit, but while sort of ignoring the little detail that Nvidia's GPU was geared for not only gaming, but for substantial GPGPU performance. As we all know.

So in other words, to compare size and performance was kind of a farce.
I hope this alleviated some of your confusion? I "think" I got the gist of it right.

-Keys

 

[SirPauly]

Where on that slide does it say "GPGPU is stupid and we're never going to do it." Please, point it out, because I see a slide that says

Imho,

Pr is PR on both sides and cause more division than anything else. AMD tried to create the illusion they were the "gaming" company and some-how nVidia has forsaken the gamer for compute. WTF?

That was utter hogwash and insulting actually based on nVidia does a lot for their gaming customers, too. There was no doubt that AMD was going compute in my mind when they offered that PR division to sell product now. PR is disingenuous at times and part of its nature -- all about selling product now.

For me its simple: Sure, I listen when AMD or nVidia speaks based on their immense talents but it's not just what they say but what they're not saying, too, that I desire to know about.

 

[MrK6]

Don't Explain the Joke

Fail troll is fail.

I am confused. Are some people saying that since AMD, at one time, adopted the "small die strategy", they cannot change their strategy without being inconsistent?

Perhaps the small die strategy was once a very convincing approach to GPU design, and now the situation has changed and a different approach is warranted? That would be, y'know, a fairly responsible thing to do: Change your strategy given the market conditions and potential sources of revenue (and diminishing opportunities from past sources).

Yes, it would, and that's why they did it.

In all honesty, I think AMD is only turning onto a road that has existed for a while now. I don't know that anything has "changed" as you put it.
And to try to reduce your confusion, as you say, I think that there isn't anything wrong with adopting a new strategy when an old one doesn't pan out.
The "funny" part might only be a rub at the people who in the past might have poked fun at Nvidia for large dies geared toward GPGPU. People would then say, "Look how much smaller AMD's dies are compared to Nvidia's, yet still seem to compete with them." This slogan was used quite a bit, but while sort of ignoring the little detail that Nvidia's GPU was geared for not only gaming, but for substantial GPGPU performance. As we all know.

So in other words, to compare size and performance was kind of a farce.
I hope this alleviated some of your confusion? I "think" I got the gist of it right.

-Keys

I disagree. Market conditions and technology change all the time, and it seems AMD saw now as the right time to deploy a GPGPU design. Despite changing designs and moving to a new process, AMD was still able to release a product that has staggering performance/watt and very competitive price/performance ratios months ahead of its competitor. Everything in the card is in check and balanced and reviews so far have been very positive. There's no heat issues, noise issues, performance issues, etc. I think this a testimonial to wait until the time is right to release such a revolutionary product rather the rush it out the door. Even then, they're still first to 28nm by a long shot (if the rumored Kepler release dates are accurate).

 

[MrK6]

Imho,

Pr is PR on both sides and cause more division than anything else. AMD tried to create the illusion they were the "gaming" company and some-how nVidia has forsaken the gamer for compute. WTF?

That was utter hogwash and insulting actually based on nVidia does a lot for their gaming customers, too. There was no doubt that AMD was going compute in my mind when they offered that PR division to sell product now. PR is disingenuous at times and part of its nature -- all about selling product now.

For me its simple: Sure, I listen when AMD or nVidia speaks based on their immense talents but it's not just what they say but what they're not saying, too, that I desire to know about.

But I think some of you read too much into PR and get incited when there isn't any reason to be incited. But that said, the above is an internal slide that's confidential, it was never part of PR nor should have been released. Even with that understood, which pretty much deflates your argument, does it say anywhere or even imply what you said above? I think like Keys, you're obscuring your personal opinion with logic. AMD has its gaming slogan "Gaming Evolved" - is it anywhere on that slide? The gist of the slide is "here's what NVIDIA's doing with their GPU, it's not being released anytime soon, it has all these features that won't add to gaming performance but will add to cost and production time, we're good." In summary, it's saying "we're not going GPGPU now and here's why." And you know what, they were right. The GTX 480 was a joke.

 

[Grooveriding]

I am confused. Are some people saying that since AMD, at one time, adopted the "small die strategy", they cannot change their strategy without being inconsistent?

Perhaps the small die strategy was once a very convincing approach to GPU design, and now the situation has changed and a different approach is warranted? That would be, y'know, a fairly responsible thing to do: Change your strategy given the market conditions and potential sources of revenue (and diminishing opportunities from past sources).

Well they are still using a smaller die, it's a little bigger than Evergreen(5870) and a little smaller than Cayman(6970). The door is likely being left open for the refresh and a small increase to die size to bring more performance.

Going on the across the board 200Mhz overclock reviewers are getting with the excellent overclock scaling on the 7970:

I think they managed to get away again with a smaller die and huge performance.

So the smaller die strategy looks intact, not shifted.

 

[NIGELG]

Imho,

Pr is PR on both sides and cause more division than anything else. AMD tried to create the illusion they were the "gaming" company and some-how nVidia has forsaken the gamer for compute. WTF?

That was utter hogwash and insulting actually based on nVidia does a lot for their gaming customers, too. There was no doubt that AMD was going compute in my mind when they offered that PR division to sell product now. PR is disingenuous at times and part of its nature -- all about selling product now.

For me its simple: Sure, I listen when AMD or nVidia speaks based on their immense talents but it's not just what they say but what they're not saying, too, that I desire to know about.

Has Nvidia ever spoke 'hogwash' or create 'illusions' in your eyes?Were they ever 'insulting'?Or disingenuous?Funny how you Nvidia types are now attacking AMD for being 'aggressive,pro active,setting the fastest card's price at 550 bucks etc...but it was okay when Nvidia was doing it.

You always preach 'proactive,aggressive' etc.Both companies have their PR.It's up to us to filter out the nonsense.

 

[SirPauly]

Of course! PR on both sides.

 

[Idontcare]

I use this on a near daily basis:
http://www.vreveal.com/


GPGPU is not something just for professionals.

Just yesterday I cleaned up a 20 minute video of my daughters Jule-ending at her school.

Took 10 minuttes to clean that video op...I wouldn't like to think on how long I would have spent if I were using my CPU...

Thanks for posting that up. I downloaded and am going to check it out. I used TMPGEnc but this app looks to be a lot more suited to my needs :thumbsup:

 

[SirPauly]

Well they are still using a smaller die, it's a little bigger than Evergreen(5870) and a little smaller than Cayman(6970). The door is likely being left open for the refresh and a small increase to die size to bring more performance.

Going on the across the board 200Mhz overclock reviewers are getting with the excellent overclock scaling on the 7970:

I think they managed to get away again with a smaller die and huge performance.

So the smaller die strategy looks intact, not shifted.

Huge performance? I don't consider that huge -- what is are the leaps like the 9700 pro or 8800 GTX over their respective generations.

Small die is some-what intact but on the rise but one thing that is not in debate is their pricing is on the rise.

It's a good chip and since there is huge and fantastic performance now for you, there is hope for you to garner an AMD platform.

 

[Idontcare]

Nope, 32nm IS the 28nm process. 32nm was so heavily delayed they changed the name to 28nm process, as the method used to measure this is subjective and not absolute.

:thumbsup:

 

[Idontcare]

And it's right here in my link:

Plus, take advantage of the fastest encoding to H.264 using NVIDIA CUDA (up to 6 times faster than the CPU).

Plus, take advantage of the fastest encoding to H.264 using NVIDIA CUDA (up to 6 times faster than the CPU).

Plus, take advantage of the fastest encoding to H.264 using NVIDIA CUDA (up to 6 times faster than the CPU).

Not via OpenCL...:biggrin:

This type of post might not violate any rules, and I am only posting this as a fellow member of this community and not as a mod, but this style of posting with super-big font is annoying as all hell. 😡 Why do it when all its going to accompish is the annoyance of anyone reading your post?

Surely there are more convincing, more productive ways to communicate your position?