What went wrong with Fermi: JHH

[Grooveriding]

I think that's the statistics misleading (and I'm not pointing you out RussianSensation, but just some of the comments that have been made in this thread). I'd be more willing to be that this is due to the different sample taken this month than it is due to people buying 5970's.

Agreed, I can't see many people buying 5970s right now. It's such a terrible buy for the price. As well the 5970 is really an enthusiast buy, I'd think someone who would pick one up would know what is going on in the video card market and not blow $600 right now with the 6 series less than a month out.

 

[jvroig]

I saw this video from golem.de about 24+ hours ago, and checked the forum to see if a thread had been created, but no thread existed yet.

Now that this thread is here, I'd just like to drop by to say I actually liked the video. It greatly humanized nVidia in my eyes. Jen-Hsun Huang came across as a nice guy, who picked up the blame, a "the buck stops here" kind of guy, and not only answered the question (what broke for Fermi), but also managed to squeeze in a lessons-learned segment.

(As an aside, I cannot understand how the S|A people can try to interpret it as another "nVidia is arrogant and says it can do no wrong" moment, when in fact JHH said it plain and simple that they were, indeed, at fault. Sure, he said "the model, the tools, and reality didn't match up", and one can possibly misconstrue that statement as "it is TSMC's fault", but he also added that what actually led to such a scenario was a management problem - ergo, he was saying it wasn't TSMC, but it was nVidia.)

While JHH may be hated by red roosters, it appears that he at least knows what he is talking about on a technical level.

This is obvious to anyone who knows nVidia's history. Even just a moderate effort at learning about nVidia's origins, the backgrounds of the three founders, and how they made the company work, will reveal as much.

However, I doubt anybody who lacks the knowledge of nVidia's (and JHH's) history and believes JHH to be one of those annoying MBA-type CEOs (instead of a person with a highly-technical background and industry experience) will watch this video and change their opinion. It wasn't very technical at all, and it is something that is easily an "engineering gave me this explanation so I'll give it to you as well" type of answer that any MBA-type CEO could accomplish after due diligence in preparing for a press con (they don't go to press conferences buck naked, they make a list of possible questions and prepare for it, and in this case "Fermi" and "delay" are definitely high on the list of possible questions). Those "red roosters" you mention will just enjoy taking the video apart and taking statements out of context and find some way to somehow make JHH seem like an arrogant SOB yet again - exactly like S|A with their article.

 

[Seero]

I saw this video from golem.de about 24+ hours ago, and checked the forum to see if a thread had been created, but no thread existed yet.

Now that this thread is here, I'd just like to drop by to say I actually liked the video. It greatly humanized nVidia in my eyes. Jen-Hsun Huang came across as a nice guy, who picked up the blame, a "the buck stops here" kind of guy, and not only answered the question (what broke for Fermi), but also managed to squeeze in a lessons-learned segment.

(As an aside, I cannot understand how the S|A people can try to interpret it as another "nVidia is arrogant and says it can do no wrong" moment, when in fact JHH said it plain and simple that they were, indeed, at fault. Sure, he said "the model, the tools, and reality didn't match up", and one can possibly misconstrue that statement as "it is TSMC's fault", but he also added that what actually led to such a scenario was a management problem - ergo, he was saying it wasn't TSMC, but it was nVidia.)


This is obvious to anyone who knows nVidia's history. Even just a moderate effort at learning about nVidia's origins, the backgrounds of the three founders, and how they made the company work, will reveal as much.

However, I doubt anybody who lacks the knowledge of nVidia's (and JHH's) history and believes JHH to be one of those annoying MBA-type CEOs (instead of a person with a highly-technical background and industry experience) will watch this video and change their opinion. It wasn't very technical at all, and it is something that is easily an "engineering gave me this explanation so I'll give it to you as well" type of answer that any MBA-type CEO could accomplish after due diligence in preparing for a press con (they don't go to press conferences buck naked, they make a list of possible questions and prepare for it, and in this case "Fermi" and "delay" are definitely high on the list of possible questions). Those "red roosters" you mention will just enjoy taking the video apart and taking statements out of context and find some way to somehow make JHH seem like an arrogant SOB yet again - exactly like S|A with their article.

+2

 

[amenx]

I wonder why no one made a big stink about ATI's big failure when it happened, the 2900 series. Their first DX10 card that was several months late and when it arrived it barely manged to hold against Nvidias mid-range, the 8800gts cards. Now THAT was a fail. Did their CEO come out and admit or explain what happened with that?

 

[SlowSpyder]

High bar set by whom? I expected GTX 295 performance. How did that turn out?

Well, they met your expectations than. But for many hardware enthusiasts they did not. The GTX480 is not a slow part. But it's slower than AMD's fastest part that was out months sooner and manages to use less power (even if only a small bit less). That's why I think some people were a bit let down.

I wonder why no one made a big stink about ATI's big failure when it happened, the 2900 series. Their first DX10 card that was several months late and when it arrived it barely manged to hold against Nvidias mid-range, the 8800gts cards. Now THAT was a fail. Did their CEO come out and admit or explain what happened with that?

Lots of people here made a big stink about it. Anandtech has a nice story on what went wrong with the 2900, and what they changed in the 4xxx series, talking to AMD engineers.

 

[Flipped Gazelle]

I wonder why no one made a big stink about ATI's big failure when it happened, the 2900 series. Their first DX10 card that was several months late and when it arrived it barely manged to hold against Nvidias mid-range, the 8800gts cards. Now THAT was a fail. Did their CEO come out and admit or explain what happened with that?

A little bit of a different situation. When Fermi was introduced, there were already DX11 titles. When 2900XT was introduced, IIRC, there were no DX10 titles. Also, the Radeon had features that Nvidia didn't.

Some sites, like AT, did take ATI to task a bit for the underwhelming part.

And, perhaps everyone was still a bit high on the awesomeness that was the G80.

Anyway, I don't see what ATI's release of the 2900 series has to do with the topic, What went wrong with Fermi: JHH

 

[Martimus]

I wonder why no one made a big stink about ATI's big failure when it happened, the 2900 series. Their first DX10 card that was several months late and when it arrived it barely manged to hold against Nvidias mid-range, the 8800gts cards. Now THAT was a fail. Did their CEO come out and admit or explain what happened with that?

I seem to remember a pretty big stink about it. In fact the reaction was pretty much the same as the current reaction to Fermi.

 

[ShadowOfMyself]

Lots of people here made a big stink about it. Anandtech has a nice story on what went wrong with the 2900, and what they changed in the 4xxx series, talking to AMD engineers.

Indeed... 2900 was almost considered the second FX series even, it was just plain bad

I still remember how AMD said the 2900 was competing with the 8800 GTS, all because it just failed to deliver :awe:

Really, no matter how you see it, there is no conspiracy here, there are only a bunch of members who state the facts regardless of the company... Right now, and since the 4xxx series, AMD has had the upper hand, so of course you can expect the forum to be filled with mostly pro-AMD people...

But wait until Nvidia releases another excellent card like the 8800 series, and the forum will suddenly become green

The problem is when you are wearing green glasses, you see neutral as red, and vice versa

 

[Genx87]

Well, they met your expectations than. But for many hardware enthusiasts they did not. The GTX480 is not a slow part. But it's slower than AMD's fastest part that was out months sooner and manages to use less power (even if only a small bit less). That's why I think some people were a bit let down.

Why would enthusiasts expect more? Go look at Nvidias track record. Their new flagship typically performs on par with the previous generations x2 part. Sounds like a bunch of people either had way too high expectations or made those expectations up in order to feign outrage at how the 480 performed out of the gate.

 

[SlowSpyder]

Why would enthusiasts expect more? Go look at Nvidias track record. Their new flagship typically performs on par with the previous generations x2 part. Sounds like a bunch of people either had way too high expectations or made those expectations up in order to feign outrage at how the 480 performed out of the gate.

But they've also been out before AMD for the last several generations. Their part also left enough flexibility with power use to create a best of the best performing x2 part the last few generations. They also haven't had JHH explaining how things went wrong the last few generations.

Even thought it's hardly what I'd call a slow part, there were obviously some let downs with Fermi.

 

[happy medium]

Did their CEO come out and admit or explain what happened with that?

NO

 

[T2k]

A little bit of a different situation. When Fermi was introduced, there were already DX11 titles. When 2900XT was introduced, IIRC, there were no DX10 titles. Also, the Radeon had features that Nvidia didn't.

Some sites, like AT, did take ATI to task a bit for the underwhelming part.

And, perhaps everyone was still a bit high on the awesomeness that was the G80.

Anyway, I don't see what ATI's release of the 2900 series has to do with the topic, What went wrong with Fermi: JHH

Exactly. Besides these obvious reasons everybody knows about 29xx-series Orton also DID come out and tell the story about their R520 flop much clearer (ie lame bug in the library) than this blurry admission by JHH, let alone doing it earlier IIRC: http://forum.beyond3d.com/showpost.php?p=729882&postcount=1 (Cannot seem to find the original front page article.)

BTW NV's biggest flop was The One and Only GeForce FX 5800 Hair Dryer by Nvidia... and they tried explain it back then too - with all the tricks of the world pulled in "reviews" and on "reviewers", causing the most despicable times in the graphics business.

 

[T2k]

NO

Wrong - they did, of course and as a matter of fact it was much better than marching out some BS-friendly CEO - it was Eric Demers, the architecture lead on R600: http://www.beyond3d.com/content/interviews/39/

 

[bryanW1995]

We are picking and choosing what we want to use to validate the 480 as slow. As a single piece of silicon AMD doesnt have anything right now that can dethrone it. In their respective markets Nvidia is faster vs the competition.

480>5870
470>5850
460>5830

Nvidia has no product that competes with the 5970. If they had a 480x2 it would be > 5970.

That said not having the single card crown hardly makes the 480 slow.

the 480 isn't necessarily SLOW, it's just not fast enough to compete with a dual gpu card from amd. 8800gtx was the same way, it was very fast but also loud/hot. it just took amd over a year to actually come out with the dual gpu card that could beat it, instead of releasing 6 months AFTER the dual gpu card that was also faster. also, the 480 is nearly as loud/hot as the 5970 but doesn't provide the same performance. if gtx 480 was extremely cool and quiet then they could have used that angle to sell some of cards, but amd has them in that category right now as well. as jhh has stated on numerous occasions, nvidia is not in this game to be #2, they shoot for the absolute performance crown with each generation. note that they don't shoot for the "single gpu" crown, they want the fastest card. their consumers expect them to produce the fastest card. gtx 480 wasn't the fastest card, so it's power/heat/noise became more of an issue than it was in earlier generations.

I've been predicting lately that nvidia will change to amd's small ball strategy, but I wonder now based upon jhh's comments the other day. They can still make it work but they will have to develop two different architectures in the future if so. I almost see him admitting that they're going after the pro market (much like amd in cpus) and just doing the best they can with the pro cards in the consumer market. hopefully I'm wrong and they'll be back with a competitive lineup soon because amd has shown many times in the past that they're just as greedy as anybody else when they have a stranglehold on the high end.

 

[exar333]

A little bit of a different situation. When Fermi was introduced, there were already DX11 titles. When 2900XT was introduced, IIRC, there were no DX10 titles. Also, the Radeon had features that Nvidia didn't.

Some sites, like AT, did take ATI to task a bit for the underwhelming part.

And, perhaps everyone was still a bit high on the awesomeness that was the G80.

Anyway, I don't see what ATI's release of the 2900 series has to do with the topic, What went wrong with Fermi: JHH

A handful of titles hardly translates to something that was already adopted and commonplace. I love my 5870, but even now DX11 is pretty much irrelevent.

 

[bryanW1995]

I saw this video from golem.de about 24+ hours ago, and checked the forum to see if a thread had been created, but no thread existed yet.

Now that this thread is here, I'd just like to drop by to say I actually liked the video. It greatly humanized nVidia in my eyes. Jen-Hsun Huang came across as a nice guy, who picked up the blame, a "the buck stops here" kind of guy, and not only answered the question (what broke for Fermi), but also managed to squeeze in a lessons-learned segment.

(As an aside, I cannot understand how the S|A people can try to interpret it as another "nVidia is arrogant and says it can do no wrong" moment, when in fact JHH said it plain and simple that they were, indeed, at fault. Sure, he said "the model, the tools, and reality didn't match up", and one can possibly misconstrue that statement as "it is TSMC's fault", but he also added that what actually led to such a scenario was a management problem - ergo, he was saying it wasn't TSMC, but it was nVidia.)


This is obvious to anyone who knows nVidia's history. Even just a moderate effort at learning about nVidia's origins, the backgrounds of the three founders, and how they made the company work, will reveal as much.

However, I doubt anybody who lacks the knowledge of nVidia's (and JHH's) history and believes JHH to be one of those annoying MBA-type CEOs (instead of a person with a highly-technical background and industry experience) will watch this video and change their opinion. It wasn't very technical at all, and it is something that is easily an "engineering gave me this explanation so I'll give it to you as well" type of answer that any MBA-type CEO could accomplish after due diligence in preparing for a press con (they don't go to press conferences buck naked, they make a list of possible questions and prepare for it, and in this case "Fermi" and "delay" are definitely high on the list of possible questions). Those "red roosters" you mention will just enjoy taking the video apart and taking statements out of context and find some way to somehow make JHH seem like an arrogant SOB yet again - exactly like S|A with their article.

yes, jhh is one of the most successful people in the history of the tech industry. this did not happen by accident, he did a fantastic job of realizing a successful vision. I am highly confident that the fermi fiasco won't be repeated any time soon and nvidia will be back on top or at the very least highly competitive again soon.

isn't it funny how when amd took a nose dive we all knew that hector ruiz was a big part of that but now that nvidia is having some hiccups they're going to be ok? It's amazing to me that people criticize ceo salaries, a truly effective ceo will make or break most companies in the long run. dirk meyer seems to have a good head on his shoulders, but I'd take jhh over him in a heartbeat.

A handful of titles hardly translates to something that was already adopted and commonplace. I love my 5870, but even now DX11 is pretty much irrelevent.

yeah, we're just now getting some decent dx11 games out. it won't be truly useful for most games for another year or two. heck the only two dx 10 games I play are DAO and civ5.

 

[T2k]

A handful of titles hardly translates to something that was already adopted and commonplace.

It WAS already commonplace among all new games when Fermi came out.

I love my 5870, but even now DX11 is pretty much irrelevent.

Ehh?

 

[Keysplayr]

It WAS already commonplace among all new games when Fermi came out.



Ehh?

Well, how many full DX11 titles do we have out now? Probably still relatively easy to count them all.

 

[PingviN]

I wonder why no one made a big stink about ATI's big failure when it happened, the 2900 series.

R600 has been bashed so many times it's just a pulp lying next to the dog crap on the sidewalk at the corner of Broken Dreams Street and Boulevard of Failure. Next to the FX-series, which has by now almost but washed away into the drain.

Their first DX10 card that was several months late and when it arrived it barely manged to hold against Nvidias mid-range, the 8800gts cards. Now THAT was a fail. Did their CEO come out and admit or explain what happened with that?

There wasn't much to admit. Everyone knew R600 missed the target by a mile. ATi went for big, Nvidia went for bigger and RV520s delay affected R600 aswell. ATi ended up sitting with a power-hungry, low efficiency chip that just couldn't compete. Since then, ATi has done right and RV770 simply dominated and Evergreen didn't get any real competition until GTX460 showed up. In some way, one wouldn't be wrong to say the R600 in itself was a failure, but in the long run? R600 arch. can be found in Evergreen. Optimized and changed a lot, but it's still fundamentally there.

Don't really see the point of your post though, must've missed it somewhere. How does ATis CEO not commenting their flaws relate to JHH:s interview? We all know Nvidia can't be overjoyed with the situation. We know somethings went wrong. We know about the delay. We don't need the CEO saying that they've failed. I would like to know why, because hardwares interesting, but that's pretty much it. Save PR for the masses, give the nerds the nerdy bits.

 

[T2k]

yes, jhh is one of the most successful people in the history of the tech industry.

Not even close, let me tell you. I hate Steve Jobs but even he's far ahead of JHH & his buddies by any measure (wealth, influence, industry etc) and SJ didn't invent crap so far and actually 80%+ of his wealth is his Pixar stock package.
Of course, I could also bring up Gates or people of Intel & Fairchild like Grove, Noyce or Kilby of TI and I don't even want to go back further like Neuman etc.

Don't get fooled by their PR: on the grand scale of things NV or ATI will be small episodes in the history of technology, their CEOs will be in footnotes if they mentioned at all.

this did not happen by accident, he did a fantastic job of realizing a successful vision. I am highly confident that the fermi fiasco won't be repeated any time soon and nvidia will be back on top or at the very least highly competitive again soon.

Yeah, riiiiight - you know that they famous about rewriting history? Take a look at the last part of this interview, the reply to "Where do you see NVIDIA expanding next?": http://www.beyond3d.com/content/interviews/7/4

It's only 5 years old... do you see a single word about being a "software company" they claim they always wanted to be?

Every successful company is always in a constant state of change - it's just mostly not public and not overnight. As I wrote 100x NV's future is really dark unless they can convince enough people to port their apps to CUDA instead of x86, even if they provide the best and fastest GPUs for the next 10 years.

isn't it funny how when amd took a nose dive we all knew that hector ruiz was a big part of that but now that nvidia is having some hiccups they're going to be ok? It's amazing to me that people criticize ceo salaries, a truly effective ceo will make or break most companies in the long run. dirk meyer seems to have a good head on his shoulders, but I'd take jhh over him in a heartbeat.

It's more than funny you think CEOs worth their money and then you bring up JHH as an example, right when he's explaining how he just completely mismanaged an entire architectural/generational transition.

CEOs make or break companies is a silly urban legend - TEAMS matter, not a single person. A very effective CEO knows how to delegate people and make them work together, that's the key but they alone don't worth jacksh!t.

yeah, we're just now getting some decent dx11 games out. it won't be truly useful for most games for another year or two. heck the only two dx 10 games I play are DAO and civ5.

STALKERs, Metro, BF2 or Dirt 2 are out since end of last year, what are you talking about?
Ahh, and FYI Civ5 IS DX11...

 

[T2k]

Well, how many full DX11 titles do we have out now? Probably still relatively easy to count them all.

Define "full DX11 title", please.















HINT: there's no such thing.

 

[RussianSensation]

CEOs make or break companies is a silly urban legend - TEAMS matter, not a single person.

HuH??????????? A major role of a CEO is to create the corporate vision and make sure it comes to fruition. If that vision/sense of direction is incorrect in the first place, then the teams/subordinates which execute on that vision aren't going to accomplish anything for the firm if they are following a flawed path.

Carly Fiorina of HP, Bob Nardelli of Home Depot, CEO of Nokia are just some examples of failed CEOs who were unable to create a good vision for their companies, which has proven detrimental to each of these firms. Similarly what would Apple be without Steve Jobs at the helm before it became what it is today? You think Steve Jobs had nothing to do with Apple's vision?

CEOs can easily make or break an organization because a CEO sets the direction and oversees the operations of an organization. Imagine having a poor general at war commanding the most skilled and experienced army. Without a great tactical general, even a well-equipped and trained army can lose a battle.

 

[Ares1214]

We are picking and choosing what we want to use to validate the 480 as slow. As a single piece of silicon AMD doesnt have anything right now that can dethrone it. In their respective markets Nvidia is faster vs the competition.

480>5870
470>5850
460>5830

Nvidia has no product that competes with the 5970. If they had a 480x2 it would be > 5970.

That said not having the single card crown hardly makes the 480 slow.

While some of that is true, i can do the same thing:

5970>480
5870>470
5850>460 and 465
5770>450

So thats not really a fair way to put it.

 

[RussianSensation]

While some of that is true, i can do the same thing:

5970>480
5870>470
5850>460 and 465
5770>450

So thats not really a fair way to put it.

I guess you missed the phrase "in their respective markets". Sorry to disappoint, but 5870 is about $90-100 more expensive than a GTX470, while GTX460 is much cheaper than a 5850 is on average. Finally, GTS450 is priced $10-15 lower than most 5770s. Therefore, what you posted has little relevance with "respective market positioning".

 

[Silverforce11]

Market positioning changes all the time. Price cuts here and there.

Fermi isn't slow. So on that measure it's not a failure. Consumers only really care about perf/price. Some may care about perf/w.

On a business level, Fermi is a failure. Why? It requires >50% die size to be competitive in performance. Unless you can operate with much lower cost, the increased die size and exponentially poor yield as a result means you are not as profitable.

gf104 (~340-360mm2) is the best hope for NV because it's very close to the competition (HD5850) in terms of perf/size.. but alas, it was very late. With Barts XT @ ~240mm2, this is going to be tough for NV.