What went wrong with Fermi: JHH

[thilanliyan]

http://translate.google.com/transla...=org.mozilla:en-US:official&channel=s&prmd=iv

http://www.xtremesystems.org/forums/showthread.php?t=259641

 

[Flipped Gazelle]

This gives me hope for seeing Kepler shipping in a year's time.

 

[RussianSensation]

I am glad that although Fermi was late, power hungry and hot, NV's management appears to have learned 2 crucial things:

1) Someone has to be in charge of specific complex parts of the design and make sure that things actually work cohesively with the rest of the architecture;

2) Understanding that theoretical 3D / physics models on the computer may not always align with the actual physical fabrication, as is often taken for grated with the use of such tools.

I am glad that a company actually acknowledges these mistakes/difficulties in detail at a press conference. I still haven't heard any concrete response from Mr. Jobs regarding the iPhone 4 antenna issues. While JHH may be hated by red roosters, it appears that he at least knows what he is talking about on a technical level.

Looking forward to heated competition in the next 3-5 years. JHH never backs down from a fight with AMD - even competitors have to respect that attitude!

 

[Flipped Gazelle]

... I still haven't heard any concrete response from Mr. Jobs regarding the iPhone 4 antenna issues.

"You're holding it wrong. - Steve"

 

[IntelUser2000]

Nah, its because Apple, despite the antennagate, is still selling trucktones of iPhone 4's, while Nvidia lost significant marketshare.

In this game, late, hot, and slow all mean the same thing. You screw up something, so it becomes late, then you cripple more "trivial" parts of the chip like managing power, then later it runs too hot so it has to be crippled performance-wise.

 

[RussianSensation]

In this game, late, hot, and slow all mean the same thing.

Not sure about that. You can be late, but if the performance is there, you can leverage the architecture for future designs. If you are late, hot and slow, then you might as well start over from scratch and do a complete redesign.

For example, Phenom 1/2 were late compared to C2D/Q, slow, and hotter. At least Fermi is late and hot but it's faster than the competition where it counts - DX11. So AMD's CPU division flopped much worse in the last 5 years. I still can't fathom how after the magnificant Athlon 64, they came up with PhII. I mean it took almost 5 years to come up with a processor that is still only about as fast as Penryn per clock.

This is getting a bit off-topic, but while people are still discussing the Fermi failure, there is never such an argument about Ph2 vs. i5/i7, but there definitely should be because Ph2 sucks. An X4 @ 3.5ghz still can't beat an i7 @ 2.66ghz. With SB 20&#37; faster than Nehalem, Bulldozer has its work cut out! While I still consider 450/460/470 decent buys from NV, there is not a single CPU from AMD <$130 that i would touch given that i5 760 is $170 at Microcenter and whips every X4/X6 to shreds (unless we are talking about video work).

 

[Flipped Gazelle]

This is getting a bit off-topic, but while people are still discussing the Fermi failure, there is never such an argument about Ph2 vs. i5/i7, but there definitely should be because Ph2 sucks. An X4 @ 3.5ghz still can't beat an i7 @ 2.66ghz. With SB 20&#37; faster than Nehalem, Bulldozer has its work cut out! While I still consider 450/460/470 decent buys from NV, there is not a single CPU from AMD <$130 that i would touch given that i5 760 is $170 at Microcenter and whips every X4/X6 to shreds (unless we are talking about video work).

PhII competed with C2Q. AMD has no competition for i7 (except for low-end i7)

AT's own benches suggest that your statement regarding i5 760 vs X6 is not entirely accurate.

Edit: not that I don't disagree that the i5 760 is an all-around better performer. It is. But to say that it "whips every X4/X6 to shreds" is just not true.

 

[RussianSensation]

AT's own benches suggest that your statement regarding i5 760 vs X6 is not entirely accurate.

Overall the X6 is a good CPU, but for gaming (i.e., minimum framerates) it's still inferior, while being priced higher than the 750/760. With cheaper AMD motherboards and Ph2 955 at $140, you can take the difference and invest it into a faster videocard. I'll give you that. But if I am keeping my mobo+cpu for 2 years or so, I'll spend the extra $30-50 for a faster CPU which will survive 3 or so videocard upgrades.

 

[Flipped Gazelle]

Overall the X6 is a good CPU, but for gaming (i.e., minimum framerates) it's still inferior, while being priced higher than the 750/760. With cheaper AMD motherboards and Ph2 955 at $140, you can take the difference and invest it into a faster videocard. I'll give you that. But if I am keeping my mobo+cpu for 2 years or so, I'll spend the extra $30-50 for a faster CPU which will survive 3 or so videocard upgrades.

Oh yeah, in terms of gaming i5/i7 has it all over AMD X4/X6, Intel C2Q. I think AMD's lagging behind Intel is brought up less because for many people - even hardcore gamers - the CPU has caught up more with software than the GPU has.

Sorry about going so far off-topic...

 

[RussianSensation]

Hey Flipped, check out this recent review at Xbitlabs. Ph2 still trails in very common tasks such as WinRAR and Adobe Photoshop.

 

[IntelUser2000]

I still can't fathom how after the magnificant Athlon 64, they came up with PhII. I mean it took almost 5 years to come up with a processor that is still only about as fast as Penryn per clock.

Considering how badly Intel missed every target with Prescott, I wouldn't be too sure about that. Phenom did deliver 15% better performance per clock than the predecessor.

Let's see, Prescott was supposed to arrive on mid-2003 with 4GHz launch clock speeds. Really, it came on Feb 2004 with 3.4GHz clocks and it never reached its 4GHz target. Even after the Prescott intro Intel was talking about quickly ramping to 4GHz.

About Nvidia. Their lead with the GTX480 is about as significant as the first Pentium 4 EE chips. Sure, they did take slight lead over the A64's, but at what cost?

 

[tviceman]

Nah, its because Apple, despite the antennagate, is still selling trucktones of iPhone 4's, while Nvidia lost significant marketshare.

In this game, late, hot, and slow all mean the same thing. You screw up something, so it becomes late, then you cripple more "trivial" parts of the chip like managing power, then later it runs too hot so it has to be crippled performance-wise.

How is Fermi slow?

 

[Flipped Gazelle]

Hey Flipped, check out this recent review at Xbitlabs. Ph2 still trails in very common tasks such as WinRAR and Adobe Photoshop.

Sadly (from AMD"s perspective) PhII has actually made inroads in Photoshop performance compared with Intel (Sysmark is 3 versions out of date, using CS2).

Here's AT's CPU bench, comparing 1055T and i5 760:

http://www.anandtech.com/bench/Product/191?vs=147

Most measurements are close enough, IMO, until gaming enters the equation, in which case X6 isn't even close.

 

[RussianSensation]

Considering how badly Intel missed every target with Prescott, I wouldn't be too sure about that.

Athlon 64 was magnificant. It took C2D to beat it. It's not just about Prescott's failure. A64 brought major innovations to the CPU arena, way ahead of Intel, such as an integrated memory controller.

About Nvidia. Their lead with the GTX480 is about as significant as the first Pentium 4 EE chips. Sure, they did take slight lead over the A64's, but at what cost?

In DX11? You are mistaken.

1920x1200 4AA
Lost Planet 2 - 65% faster
AvP - 25% faster
Dirt 2 - 23% faster
Metro 2033 - 39% faster
STALKER: CoP - 34% faster

When was P4 EE 25-65% faster than A64 in gaming?

 

[T2k]

It's been explained very well here on AT long time ago: What went wrong with NV's Fermi?

 

[T2k]

Considering how badly Intel missed every target with Prescott, I wouldn't be too sure about that. Phenom did deliver 15&#37; better performance per clock than the predecessor.

Let's see, Prescott was supposed to arrive on mid-2003 with 4GHz launch clock speeds. Really, it came on Feb 2004 with 3.4GHz clocks and it never reached its 4GHz target. Even after the Prescott intro Intel was talking about quickly ramping to 4GHz.

About Nvidia. Their lead with the GTX480 is about as significant as the first Pentium 4 EE chips. Sure, they did take slight lead over the A64's, but at what cost?

 

[Keysplayr]

It's been explained very well here on AT long time ago: What went wrong with NV's Fermi?

I don't think so.

 

[Skurge]

Hey Flipped, check out this recent review at Xbitlabs. Ph2 still trails in very common tasks such as WinRAR and Adobe Photoshop.

I know this is off topic, but I'm actually amazed how well dirt2 and RE5 scale so well with extra cores. The lower clocked X6 is faster than the higher clocked X4, and the X4 is twice as fast as the X2. Anyone know of a review where they test the multithreading of the Ego engine and the MT framework engine?

 

[PingviN]

So, basically, the "fabric" failed, resulting in a lack of communications between the clusterns. Also, the administrations failed, resulting in a lack of communications between crews of engineers. Oh the irony!

Maybe Nvidia should do what AMD did and re-evaluate the strategy. AMD went for a more scheduled orientation resulting in 2 highly sucessfull generations of GPUs - in a row.

 

[T2k]

I don't think so.

Ehh? JHH essentially says the same thing...

 

[IntelUser2000]

When was P4 EE 25-65&#37; faster than A64 in gaming?

It wasn't. Comparing the gains of next generation on a do-everything-fast general purpose CPU to a 3D-specialized app is really another world altogether. 20% in CPU world is sometimes more than a generation gap, but that same number in GPU might be SKU difference.

Athlon 64 was magnificant. It took C2D to beat it. It's not just about Prescott's failure. A64 brought major innovations to the CPU arena, way ahead of Intel, such as an integrated memory controller.

Right, but it was made far worse because Intel screwed up in addition. If we compare that to the more modern days, the original Phenom would have been competitive if it had high enough clocks, just like a 4.5GHz Prescott would have been against A64's.

How is Fermi slow?

Ok. Maybe it wasn't as bad as Prescott. By having 6 months delay with ~15% spec reduction, they lost quite a bit of marketshare. It's their major screw up since the 5800 days.

 

[Red Storm]

Maybe it's just me, but I think the reason why we see more video card fawning and flaming (this round it's Fermi) instead of Intel vs AMD comparisons is because... you know, this is the Video Cards and Graphics subforum.

Seriously though, there is a big difference. Intel is has many, many more resources to use on R&D and production than AMD. It would take a absolute failure in chip design to cede the performance crown to AMD in the CPU market. At the same time, AMD isn't going anywhere. The last thing Intel wants is for AMD to go out of business. The video graphics war is much more interesting as market share is much more evenly split (and hopefully will always stay around the 50/50 mark!).

 

[Flipped Gazelle]

Maybe it's just me, but I think the reason why we see more video card fawning and flaming (this round it's Fermi) instead of Intel vs AMD comparisons is because... you know, this is the Video Cards and Graphics subforum.

Seriously though, there is a big difference. Intel is has many, many more resources to use on R&D and production than AMD. It would take a absolute failure in chip design to cede the performance crown to AMD in the CPU market. At the same time, AMD isn't going anywhere. The last thing Intel wants is for AMD to go out of business. The video graphics war is much more interesting as market share is much more evenly split (and hopefully will always stay around the 50/50 mark!).

I think a large part of the difference between Intel & AMD fanboys is that there is some clearly defined segmentation right now. Intel owns the high-end, and even the mid-high end. PhII X6 is recognized as the top choice for AMD upgraders, but competes with Intel's mid-range products, in sort of a tit-for-tat battle. There's decent parity in the mid-range segment, and AMD has an edge in the budget category. So, except for trolls (there's one who is both Intel & Nvidia ) everyone pretty much recognizes the pecking order.

Nvidia & AMD are fighting for the same ground in the graphics market, though, so for those ridiculous enough to choose sides other than some aspect of performance, there is ample opportunity for "fawning and flaming".

 

[electroju]

AMD Phenom or K10 processors are mainly designed to handle multi-threads and not single threads. People here need to look at the benchmarks more closely. A lot of programs still does not utilize multiple processors, so single thread processors like the Intel's Core processors work well. AMD processors starts winning the benchmarks when programs send out more than four threads.

The Athlon64 or K8 is great, but people rarely notice it. Thanks to Dell and their advertisement "Dude get a Dell" made it hard for AMD to sell any K8 systems. Dell sells and only sells Intel systems. Manly people bought Dell computers during those days. Advertisements brainwash people, so people buy Intel systems. I have not seen any advertisement for AMD. The only way for K8 processor to be advertise is from the DIY community. The DIY community for computers is very, very small, so AMD did not do well during the K8 days. Now the DIY community is advertising the Core processors even though there is plenty of commercial advertisements from Intel.

The Fermi graphic processor just got delayed. There is nothing to cry about it. The K8 processors got delayed too and it came out great. AMD did not get their luck when they delayed the K10 core, so you win some and you lose some. Fermi actually won after long delays. The Fermi is the most complex chip than what AMD and Intel can create, so everybody should applaud of a small company like nVidia having a successful product that is the most complex. Intel lost when they first made the Pentium 4. Pentium 4 suck when they first came out because it can not compete against its predecessor the Pentium III. This is not the first time Intel made their latest processor to not be faster than the predecessor. IMHO, Intel is a yo-yo company. The Core i7 did the yo-yo. Intel does better at their tock than their tick. Core 2 Duo processor were the tock because it came from Pentium M.

AMD these days are doing good at the low end or budget. Most people do not care for the feeling of warm and mushy when they bought a computer that contains a high performance processor. A lot people here are missing the point in computers. There are a lot people buy computers based on price. If they see a system cheaper than the other, they will get it. They do not care what is inside. If they pick a computer that has an AMD and it is the cheapest of them all, that person will get it. The majority of the people that buy computers are just using it to browse web sites, look up email, office, social networks, so an AMD system is best at this since it is the cheapest. Most gamers are in this forum, so they do not see this until they help someone out to buy a computer.

 

[tweakboy]

I think for me what killed it was the 335watts requirement. Only 512 shader streams alto I believe this is X 2 anyhow.

They blew it because nothing could touch a 5970 flagship AMD. While nVidias flagship is a single gpu on pcb design and takes up 335watts, I bet the AMD takes up half as many watts.

They need to fix the wattage bs and the heat. better pcb design.

They should introduce the 5xx line and have a dual gpu card or else I think AMD wins again with their 6xxx dual gpu.. Thanks,,