AMD goes fermi

[Grooveriding]

Crysis 2 ran better because of the tessellation of Fermi over Evergreen/Cayman.

Civ V ran better because of DX11 multi-threading support in NV drivers afaik.

7970 drivers have multi-threading support now ( I assume I didn't look through reviews for this) but the performance seen points to that. Whatever changes were brought about in Tahiti also look to have negated the tessellation advantage seen in Fermi.

I think most know that Hawx2, LP2 and C2 were all titles heavily influenced by nv to give results on their cards that put them in a better light. None of them really offered any sort of IQ of a level you were not getting in other games. See BF3 which supports and uses terrain tessellation but doesn't choke NV or AMD cards to make use of it vs C2 which arguably is not as nice a looking game as BF3 and hammers FPS on both cards when you use tessellation, moreso on AMD. Or Hawx 2 which looks pretty bad but has a large performance hit with copious tessellation used.

I never saw the tessellation thing as a flaw, because I saw nothing in the two heavily tessellated games, C2 & Hawx2, that made it look worthwhile or better than games that were not using excessive tessellation. Either way, it certainly looks like AMD cared enough to rectify things and then some.

Overclocked 7970 is as fast as a GTX590 in Crysis 2 now.

Being the speculator I am, I am looking forward to seeing what sort of disparities nvidia tries to highlight between the two once they have their new cards out middle of next year. It's something they do with every generational release, and I am wondering what will be the next 'advantage' they create out of thin air. It's never been anything worthwhile to the gamer and always only been good marketing.

 

[notty22]

For add in boards it's still about 60-40 for Nvidia. Those other announcements take in to account gpu chips. It's why Intel leads in those survey's.
Here are 2 John Peddie announcements. Taken from here.
Q3 graphics shipments up 16.7% over last quarter18.4% over last year


And this was from August , dealing strictly with Add In Video cards.
It shows quarter over quarter, year to year, and prior quarter.
Notice, Intel isn't shown there ?
Direct receipt of these surveys in their totality can only be had for costly subscriptions, in the thousands of dollars. It's why, we only see bit's and pieces from quarters, or certain angles of the whole market. Investors pay for this information, to help gain the real overall picture of companies worth.
Graphics Add-in Board Shipments Decline 15.2% from Last Quarter

 

[96Firebird]

Playing defense already, and they haven't even done anything yet!

So now that AMD is up to par with Nvidia on tessellation, it is back to being the greatest thing again, right?

 

[GaiaHunter]

For add in boards it's still about 60-40 for Nvidia. Those other announcements take in to account gpu chips. It's why Intel leads in those survey's.

Sure but AMD is also shipping APUs. I'm not sure if you can really compare those to IGPs.

So the waters are a bit muddied since now there is a new market for people that want more than traditional IGP power but don't really require more than the GPU on the APU.

This also makes IGPs shipments boost like crazy and AIB shipments drop substantially without the shipments of lower end gpus.

 

[notty22]

Sure but AMD is also shipping APUs. I'm not sure if you can really compare those to IGPs.

So the waters are a bit muddied since now there is a new market for people that want more than traditional IGP power but don't really require more than the GPU on the APU.

This also makes IGPs shipments boost like crazy and AIB shipments drop substantially without the shipments of lower end gpus.

k

It's going to give a Editor at one of these tech sites, good material to write a article about. Some survey's have included the IGP in chipsets. Those are almost completely gone. Now inside of both AMD/Intel CPU's.
It's why overall in some survey's Nvidia's numbers will continue to diminish.

The trend toward multiple GPUs in a single PC is clearly helping the graphics industry. Intel and AMD are driving this trend with embedded processor graphics CPUs (EPGs). According to JPR there were, on average, 1.15 GPUs in PC back in 2001, but there are 1.6 GPUs in every PC today.

 

[taltamir]

I didn't realize you guys were in some sort of academic debate over architecture.

Its not an academic debate. (I wanted one but not a single person discussed it with me)

People shouldn't get hung up about engineering decisions (256bit vs 512bit memory architecture) but the overall performance of a card. Performance is measured in its efficiency. Not in FPS, not in temperature, not in noise, and not in power draw. This is because all of the above are solely determined by the clocking of the card and can easily be changed.

Efficiency of an architecture is not an eingeering decision, it is the overall result of the quality of the architecture. Unlike memory controller size (an engineering decision) you can EASILY increase or decrease clockspeed to control power consumption and thermals, in fact people do it at home all the time. It is called overclocking and underclocking. Not only that, there are factory overclocked hardware and factory downclocked (made into dual GPU or fanless cooler)

You are saying "oh, they chose to clock it this fast consuming this much power so its all academic"... no it isn't, clockspeed and all its results are not architectural nor acedemic. And while comparing current products is the norm, the brand spanking new paper lunch of a 28nm product does not fairly compare to old 40nm products, especially when competing 28nm from nvidia is right around the corner. Yes, in three weeks (when it really arrives in stores) the GCN would be the best card on the market by virtue of having an overwhelming node advantage. For a very short time before more 28nm parts are released. And it might very well be that we will see factory overclocked versions. From what I have read the GCN cards overclock like crazy, but power consumption, and thermals, and noise all go through the roof when they do. Anandtech article clearly states that the limiting factor on GCN speed is power consumption.

 

[SirPauly]

Playing defense already, and they haven't even done anything yet!

So now that AMD is up to par with Nvidia on tessellation, it is back to being the greatest thing again, right?

I think it is wonderful to see stronger tessellation from both because developers may add more to improve immersion.

It's not a clean sweep though with higher levels:

http://www.pcgameshardware.de/aid,8...m/Grafikkarte/Test/bildergalerie/?iid=1607694

However, with the extreme tests it does wonderfully with Ungine Engine:

http://www.pcgameshardware.de/aid,8...Express-30-und-28-nm/Grafikkarte/Test/?page=5

And titles like Crysis 2:

http://www.pcgameshardware.de/aid,8...Express-30-und-28-nm/Grafikkarte/Test/?page=9

There is a lot to like about this chip when compared to the last generation.

 

[Lonbjerg]

I've been following this argument and I can't say I agree, especially when you look at it from several perspectives. First off, if you look at the charts from the last dozen posts, the difference between a GTX 580 and HD6970 isn't even 50-60W, it's more 29-33W from those charts. While that may seem like it makes this whole point moot, it's still more than significant. For example, if you add enough voltage and clock the HD6970 high enough to consume the same amount of power as a GTX 580, you know what you'll get? GTX 580 performance, give or take depending on the game. And there in is the answer to why power consumption and performance/watt is so important - they basically designate the quality of the chip. Now the above scenario can also be attested to the 40nm process in itself, but does any of that matter to consumers when the 6970 is $380 and the GTX 580 is $500? What does the extra $120 get you? Possibly +/- another 20&#37; performance, + more heat and power consumption? What about the needed PSU to put up with that? All these costs are multiplicative, simply because the chip was pushed too far from the start. If you throw in the fact that a <$250 6950 2GB can be unlocked to a 6970, and overclocked largely the same, now the GTX 580 is essentially double the cost for only the chance at ~20% higher performance, assuming it clocks well. And it has less vRAM to boot, but I digress. The take home point - the GTX 580 was never a good buy, and that just highlights what a joke the GTX 480 was.

This is why the 7970 is that much more impressive. To an average consumer, one might look at it and say "oh, it's only 20-25% faster than the GTX 580, that's not much." But then, once you start analyzing different metrics and become a better consumer, it really shines. Any consumer regardless of technical knowledge can do a quick price/performance comparison and say, "wait, that's 20-25% extra performance for only $50 at the enthusiast level - that's unheard of, and it has double the vRAM!" Then, if that consumer is an enthusiast, one can look at the technical merits of the chip itself, such as the lower power consumption and overclockability, and the picture just gets better and better. As I stated above, if you crank the 7970 to a GTX 580's power consumption level, it will slaughter it. Look at the power consumption chart Arkadel posted - even while consuming less power than the GTX 580, the overclocked 7970 is in the range of 40-50% faster. Yes, it's a process advantage, but I'd imagine NVIDIA will follow its same suite on 28nm - it'll "overclock" its chips until they have some performance lead over AMD's at the cost of heat and noise. Consumers will get wise to this (if not inherently, look at the Fermi release and what it did to NVIDIA's marketshare), and I hope posts like these help spread that message.

How much power does your 5.0GHz CPU gooble up?
And why should we care about that?

 

[Lonbjerg]

Imho,

Oh, do I agree with this. To assume that everyone that thinks of Fermi as a detriment or negative is very odd. Fermi, over-all, has been a very robust architecture that brings in strong revenue from three families: GeForce, Quadro and Tesla. There are reasons why AMD is going Fermi and it's called revenue potential.

Forum posts, are one sided and extreme at times, opinions are facts, some assume everyone thinks like they do -- part of its lore and fascination, one may imagine!

Also, GPU Processing is important for the future of gaming to me from improved dynamics to eventually RayTracing; it has to begin to evolve and mature.

Of course GPU Processing was/is important to ATI:

From 2006

http://techreport.com/articles.x/10956/1

If you look at the Steam numbers the GTX480 outsold the 5970 and the GTX580 outsold the 6970 by more than double...so the numbers speak a different tone, than that of some vocal minority on forums...

 

[GaiaHunter]

k

It's going to give a Editor at one of these tech sites, good material to write a article about. Some survey's have included the IGP in chipsets. Those are almost completely gone. Now inside of both AMD/Intel CPU's.
It's why overall in some survey's Nvidia's numbers will continue to diminish.

These reports also have breakdown by price bracket and occasionally one site or the other publishes it (although I guess it costs more to obtain/publish that kind of data). I would be interested to see what has been the movement on the lower end, see if that is where the AIB decrease is seen (and if so who lost more units shipped on those brackets) or if the loss in units is from the mid range and high end.

 

[MrK6]

I agree it's not ideal but what I like is over-all revenue, profits and margins -- market share numbers from Mercury Research and Jon Peddie.

Then don't use it, I promise I won't either :thumbsup:

Being the speculator I am, I am looking forward to seeing what sort of disparities nvidia tries to highlight between the two once they have their new cards out middle of next year. It's something they do with every generational release, and I am wondering what will be the next 'advantage' they create out of thin air. It's never been anything worthwhile to the gamer and always only been good marketing.

It's going to be tougher. Like I said, I think it'd be an absolute riot if AMD just started porting over all of NVIDIA optimizations if the code is that similar. I wonder what the legality of it all would be, but it would definitely be interesting.

For add in boards it's still about 60-40 for Nvidia. Those other announcements take in to account gpu chips. It's why Intel leads in those survey's.
Here are 2 John Peddie announcements. Taken from here.
Q3 graphics shipments up 16.7% over last quarter18.4% over last year


And this was from August , dealing strictly with Add In Video cards.
It shows quarter over quarter, year to year, and prior quarter.
Notice, Intel isn't shown there ?
Direct receipt of these surveys in their totality can only be had for costly subscriptions, in the thousands of dollars. It's why, we only see bit's and pieces from quarters, or certain angles of the whole market. Investors pay for this information, to help gain the real overall picture of companies worth.
Graphics Add-in Board Shipments Decline 15.2% from Last Quarter

I should have clarified, I'm discussing discrete, desktop GPUs. My point was that consumers get wise to the added costs of using inefficient GPUs. The article I linked to showed how NVIDIA is losing place within OEM's as an upgrade choice. Those articles you linked to include all AIB shipments by the companies, which includes mobile sales as well. I have no doubt NVIDIA is selling mobile chipsets well, especially with its Optimus system. But that's besides the point since we're discussing desktop cards.

Playing defense already, and they haven't even done anything yet!

So now that AMD is up to par with Nvidia on tessellation, it is back to being the greatest thing again, right?

Tessellation is what it is - a tool to give us more eye candy at less performance penalty, as long as it's used properly. Where did you read any of the rest into the discussion? I look forward to seeing tessellation properly implemented into games now, instead of being bastardized to NVIDIA's (short-lived) gain.

 

[SirPauly]

Imho,

The key with discrete is to look at the attach rates and if these drop drastically there is trouble in Santa Clara so-to-speak. nVidia doesn't offer integrated anymore, so, their over-all GPU Percentage market share may continue to drop.

 

[SirPauly]

Then don't use it, I promise I won't either :thumbsup:.

I couldn't help it though after reading extreme words like negative stigma and crippled, sorry!

 

[MrK6]

How much power does your 5.0GHz CPU gooble up?
And why should we care about that?

You clearly don't understand the discussion, so I'm going to simplify it. Here are three points to keep in mind: 1) Power consumption is a byproduct of an integrated circuit (IC) running. 2) Overclocking, our pastime, gives us increased performance from a part at the cost of power consumption (increases approximately linearly with clockspeed and quadratically with voltage) and its byproduct, heat. 3) PC's are built and run on constraints, including heat tolerance, noise, and more globally, budgets.

To begin, overclocking nullifies the importance of stock settings, and instead looks at performance on a spectrum based on the quality of the part, and not any predefined points on that spectrum. This changes performance comparisons from "this is my GPU speed" to "my GPU can do this speed, at this voltage, under these temperatures." This places a much greater emphasis on the quality of the chip itself. Therefore, those chips that offer the best performance/watt are the most desirable, since they have the most flexibility in the above described spectrum. This is why a 6950, the king of performance/watt in the enthusiast segment prior to the 7970, was such a steal. As I mentioned, for <$250, it could be overclocked to GTX 580 performance and had more RAM, for half the cost! Where as even the best GTX 580 samples could only squeeze out ~ another 20% performance with an overclock. This reduces the GTX 580's performance advantage from a stock ~30% down to only ~20%, while also adding in possibility that it might not even be needed at all if the performance of the overclocked 6950 is sufficient. That's the power of an efficient chip, and why performance/watt matters.

To answer your specific question, my entire system, with the CPU at 5.0GHz, only draws ~270W from the wall at full load in LinX. This is actually less than my previous i5-750 drew at 4.1GHz. CPU's are obviously different from GPU's, but can still be used as an analogy of the above concepts that I hope will help you understand them better.

If this still isn't clear, please highlight/point out specific statements in this post or my previous one that were confusing to you and I will try my best to explain it further.

If you look at the Steam numbers the GTX480 outsold the 5970 and the GTX580 outsold the 6970 by more than double...so the numbers speak a different tone, than that of some vocal minority on forums...

That's an erroneous statement, as the Steam survey doesn't work like that. You can't prove sales based on a sample survey in a niche market. This is Statistics 101.

 

[taltamir]

if you crank the 7970 to a GTX 580's power consumption level, it will slaughter it.

I would be very very concerned for the survival of AMD if that wasn't the case, since the 7970 is 28nm card being released in 3 weeks (softlunched now) and the 580 is 40nm and been on the market 13 months. And it is expected that the nvidia competitor to the 7970 being out within a month or two of the 7970.

The fact that 7970 cranked to GTX580 power consumption slaughters the GTX580 is not an indication of the quality of the GCN vs fermi architecture (as far as I can tell GCN is copying fermi left and right). It is merely an indication of how important it is to switch to a better manufacturing process.

I have yet to see anyone here dispute the claim that the 7970 is a better buy then the GTX580 (at current pricing)... the issues people raise are that the comparison between the two is rather flawed since they were released 13 months apart (14 if you count the softlunch, I just checked newegg and you can't buy a 7970 yet) and at different process nodes. And we have an upcoming competitor from nvidia in the same 28nm node. There was also discussion on other aspects of the cards beyond the "which is a better buy"

 

[MrK6]

I would be very very concerned for the survival of AMD if that wasn't the case, since the 7970 is 28nm card being released in 3 weeks (softlunched now) and the 580 is 40nm and been on the market 13 months. And it is expected that the nvidia competitor to the 7970 being out within a month or two of the 7970.

The fact that 7970 cranked to GTX580 power consumption slaughters the GTX580 is not an indication of the quality of the GCN vs fermi architecture (as far as I can tell GCN is copying fermi left and right). It is merely an indication of how important it is to switch to a better manufacturing process.

I have yet to see anyone here dispute the claim that the 7970 is a better buy then the GTX580 (at current pricing)... the issues people raise are that the comparison between the two is rather flawed since they were released 13 months apart (14 if you count the softlunch) and at different process nodes. And we have an upcoming competitor from nvidia in the same 28nm node. There was also discussion on other aspects of the cards beyond the "which is a better buy"

Were you worried when the GTX 480 was released 6 months after the 5870 and was only 10&#37; faster but consumed 50% more power (nevermind the cost, noise, etc.). NVIDIA's working with the same 28nm process that AMD is, and they already botched one new architecture/new node launch. Did they learn from it or will they make the same mistakes (or be forced to since they follow the large die strategy, among others)? Time will tell. And by all reports that will be for a while, not a month or two. Think, if NVIDIA was only a month or two away from launch, don't you think they'd be banging their drum about something right now in order to distract from the 7970's launch? I'd be more concerned about whether or not NVIDIA will deliver and in a timely fashion.

 

[SirPauly]

The performance per watt is fair game, very valid and good points, to me. The key is one may also believe that nVidia recognizes the need to improve performance per watt and having been banging this drum with Kepler. Will Kepler have impressive efficiency with performance per watt top-to-bottom? Time will tell!

 

[Lonbjerg]

That's an erroneous statement, as the Steam survey doesn't work like that. You can't prove sales based on a sample survey in a niche market. This is Statistics 101.

You can repeat that fallacy all you want to, dosn't make it any less a fallacy.
I only spoke of niche products, high end GPU's for wich Steam has ample numbers of convenience sampling making it statistical significant.

It's was "good enough" when AMD had a 6 months lead...but now it's not :whiste:

 

[MrK6]

You can repeat that fallacy all you want to, dosn't make it any less a fallacy.

If you think that's a fallacy, then you clearly don't understand statistics at all and have no place validating sources. For your benefit, do some reading and then you can join the discussion appropriately: http://www.amazon.com/Complete-Idiot...5198436&amp;sr=1-2

I only spoke of niche products, high end GPU's for wich Steam has ample numbers of convenience sampling making it statistical significant.

It's was "good enough" when AMD had a 6 months lead...but now it's not :whiste:

Steam is the niche market in this case, not the cards themselves. Anybody who knows even the slightest about statistics would have realized that. As we've already repeated and verified in this thread, the Steam survey still isn't a valid source, and hence we've been referring to professional surveys like those from JPR. Next thing I know you'll start referencing Wikipedia .

 

[notty22]

You can repeat that fallacy all you want to, dosn't make it any less a fallacy.
I only spoke of niche products, high end GPU's for wich Steam has ample numbers of convenience sampling making it statistical significant.

It's was "good enough" when AMD had a 6 months lead...but now it's not :whiste:

Agreed and in return, he is throwing up a Fudzilla article, and I don't know if he is seriously believing his own spin, that there was a 40% swing in gpu shipments. Market share or whatever value MrK is stating.

 

[SirPauly]

Steam may not be an exact or maybe even an objective metric but does offer some insight and shouldn't be ignored totally or dismissed totally and quite frankly, fun for spirited discussions. There are reasons why Steam data isn't used for financials or conference calls though. Steam may of been used in financials or conference calls but don't recall any.

 

[MrK6]

Agreed and in return, he is throwing up a Fudzilla article, and I don't know if he is seriously believing his own spin, that there was a 40&#37; swing in gpu shipments. Market share or whatever value MrK is stating.

The Fudzilla links to an article regarding the results of FBR's quarterly summaries. If you can't be bothered to read the source material, why are you participating in this discussion?

And since you both seem to assert that Steam is a valid source for marketshare information (which you curiously neglected to reference in any of your posts, notty22), please, either or both of you, write a proof as to why the Steam hardware survey is a legitimate source. This includes it's details in data gathering and the math/formulas behind it. Furthermore, please explain to the forum how you can insure its accuracy so that it may sit in the ranks of JPR, et al. that are the current gold standard.

 

[Lonbjerg]

Agreed and in return, he is throwing up a Fudzilla article, and I don't know if he is seriously believing his own spin, that there was a 40% swing in gpu shipments. Market share or whatever value MrK is stating.

Ah...so FUD is kosher now and Steam is not...lol

 

[notty22]

The Fudzilla links to an article regarding the results of FBR's quarterly summaries. If you can't be bothered to read the source material, why are you participating in this discussion?

And since you both seem to assert that Steam is a valid source for marketshare information (which you curiously neglected to reference in any of your posts, notty22), please, either or both of you, write a proof as to why the Steam hardware survey is a legitimate source. This includes it's details in data gathering and the math/formulas behind it. Furthermore, please explain to the forum how you can insure its accuracy so that it may sit in the ranks of JPR, et al. that are the current gold standard.

As to who can't be bothered. There is so much wrong with your tying in the value of Cayman, some huge swing in gpu share and discounting facts like I posted from August on add in boards . And information like AMD gpu division losing money in Q2, which would support low share in the Steam survey.
edit : fixed link

 

[Lonbjerg]

If you think that's a fallacy, then you clearly don't understand statistics at all and have no place validating sources. For your benefit, do some reading and then you can join the discussion appropriately: http://www.amazon.com/Complete-Idiot...5198436&amp;sr=1-2

All I see is a badly disguised personal attack...with not arguments

Steam is the niche market in this case, not the cards themselves. Anybody who knows even the slightest about statistics would have realized that. As we've already repeated and verified in this thread, the Steam survey still isn't a valid source, and hence we've been referring to professional surveys like those from JPR. Next thing I know you'll start referencing Wikipedia .

Steam is +30 millions users...if you call that niche, you should take your own advice...:ninja: