NVIDIA Announces Earnings of $2.6 Billion for Q3 2018

[Yotsugi]

By all means, enlighten me and my bad memory. When did AMD have the crown in all three of performance, power and price? As far as I can tell, they have only ever had the crown in 2 out of 3 at most.

5870 was absolutely perfect.

 

[antihelten]

5870 was absolutely perfect.

Looks like I'm not the one with the bad memory. The 5870 certainly won on power and price, but lost on performance (the 480 being about 10-15% faster), so no trifecta there.

 

[Yotsugi]

Looks like I'm not the one with the bad memory. The 5870 certainly won on power and price, but lost on performance (the 480 being about 10-15% faster), so no trifecta there.

It was out half a year earlier than 480.
On launch, it won the trifecta.

 

[antihelten]

It was out half a year earlier than 480.
On launch, it won the trifecta.

Incorrect again. At launch the 5870 was still slower than the 295x2, so even if we move the goalposts to the rather ridiculous notion of comparing against last gen products (and I think it was fairly obvious that this was not what MajinCry was refering to), AMD has still never held the trifecta.

Here's a direct quote from AT review of the 5870:

The catch however is that what we don’t have is a level of clear domination when it comes to single-card solutions. AMD was shooting to beat the GTX 295 with the 5870, but in our benchmarks that’s not happening. The 295 and the 5870 are close, perhaps close enough that NVIDIA will need to reconsider their position, but it’s not enough to outright dethrone the GTX 295. NVIDIA still has the faster single-card solution, although the $100 price premium is well in excess of the <10% performance premium.

So again, the 5870 was a great card, but it didn't get the trifecta. Or as AT put it:

As it stands the 5870 is the greater value, even if it's not the fastest card

 

[Yotsugi]

So again, the 5870 was a great card, but it didn't get the trifecta.

295x2 was a dual GPU board.
5970 crushed it if we're comparing dual GPU solutions.

 

[antihelten]

295x2 was a dual GPU board.
5970 crushed it if we're comparing dual GPU solutions.

So what if it was a dual GPU card, it's still comparing a single card to a single card (the best available from either vendor at the time). And yes the 5970 obviously crushed the 295x2 (although not by as much as one might except from a next gen card), but guess what, the 5970 didn't have the power and price advantage that the 5870 did, so still no trifecta.

 

[Head1985]

5870/6970 were last cards where AMD have performance/mm2 advantage.If they choose make 500mm+ GPU like Nvidia they would be faster than nvidia GTX480/580.After that bad GCN came out and nvidia started selling those midrange gpus for 500+usd(it end up with GTX1080 for 700USD)thats how they make money.Not from Titans but from overpriced midrange cash cow/milking cards.

You guys should ask why GTX1080 and GTX1080Ti cost same at launch.
GTX1080 is only 314mm2 with 256bit and 8GB Vram and super cheap PCB
GTX1080TI is 471mm2(50% larger)have 354bit and 11GB Vram and good PCB
Nvidia making insane money on those overpriced midrange GPUs like GTX1080 and 1070.Same it will be with gaming volta.first overpriced midrange in disguise GTX2070/2080 to make insane amount of money and after that normal high-end 2080TI card.

 

[caswow]

5870/6970 were last cards where AMD have performance/mm2 advantage.If they choose make 500mm+ GPU like Nvidia they would be faster than nvidia GTX480/580.After that bad GCN came out and nvidia started selling those midrange gpus for 500+usd(it end up with GTX1080 for 700USD)thats how they make money.Not from Titans but from overpriced midrange cash cow/milking cards.

That is not true.
Hawaii was 438mm² released okt. 2013 vs. the 780 that was at 561mm² released may 2013 or the 780ti released shortly after hawaii also at 561mm². so no the perf/mm² was just one or two generations ago in favor for amd.

 

[Head1985]

That is not true.
Hawaii was 438mm² released okt. 2013 vs. the 780 that was at 561mm² released may 2013 or the 780ti released shortly after hawaii also at 561mm². so no the perf/mm² was just one or two generations ago in favor for amd.

Only 780Ti was full GPU and it was also faster than 290x.AIB 780TI was around 20% faster than 290x uber.So they were pretty close in perf/mm2
https://www.techpowerup.com/reviews/MSI/GTX_780_Ti_Gaming/24.html
+ hawaii was only 2 cards not entire product stack.

 

[caswow]

Only 780Ti was full GPU and it was also faster than 290x.AIB 780TI was around 20% faster than 290x uber.So they were pretty close in perf/mm2
https://www.techpowerup.com/reviews/MSI/GTX_780_Ti_Gaming/24.html
+ hawaii was only 2 cards not entire product stack.

why are you comparing an aftermarket ti to an ref. 290x. point still stands. amd had an perf. mm² advantage even if they were almost the same in speed. it gets even worse if you look today.

 

[Cloudfire777]

Talking about dual GPUs...

Watch AMDs answer to Volta in entire 2018 is dual Vega and refreshes on new node with a little higher clocks on the GPUs today.

That is what we can expect from AMD next year imo

 

[Head1985]

why are you comparing an aftermarket ti to an ref. 290x. point still stands. amd had an perf. mm² advantage even if they were almost the same in speed. it gets even worse if you look today.

because 290x uber =AIB 290x
https://www.techpowerup.com/reviews/MSI/R9_290X_Gaming/24.html

 

[crisium]

It's really a damn shame that AMD chose deliberately to allow Nvidia to have the most powerful GPU during the Tesla and Fermi years. That's how insane the leadership of that organization has been.

Realizing they had an advantage in performance per watt and an even greater advantage in performance per mm^2 they chose the "Small-die" strategy that ultimated just cost them sales on the high end and damaged their mindshare into making them the "budget" company. So they leveraged their cheaper to make, faster, more efficient GPUs only up to a certain size, conveniently (for Nvidia) only to where they got to within 15% or so of their competitor's gargantuan top end chips.

As Head1985 said, if AMD make a big GPU from 2008-2011 they would have had the performance crown for all of those years. I understand their first attempt at a mammoth GPU was a relative failure; the 2900XT didn't cut it. I understand why they made the 3870 small. But the success of the chip should have given them pause for thought, and there was no excuse for 3 more generations of deliberately tiny chips.

As it was, they only had the temporary performance crown by launching their gen first a couple times.

It's as if Nvidia in current years never bothered releasing bigger chips than their x04 line, even if Hawaii/Fiji/Vega were around. But they instead fully capitalize on their advantages, top to bottom, and take a look at those earnings.

 

[tviceman]

Talking about dual GPUs...

Watch AMDs answer to Volta in entire 2018 is dual Vega and refreshes on new node with a little higher clocks on the GPUs today.

That is what we can expect from AMD next year imo

I doubt AMD will shrink their existing chips. Neither Nvidia nor AMD has done a direct shrink of a pre-existing chip for quite some time, probably due to scale of economy and complexity.

 

[Cloudfire777]

I doubt AMD will shrink their existing chips. Neither Nvidia nor AMD has done a direct shrink of a pre-existing chip for quite some time, probably due to scale of economy and complexity.

True, but what other choice does AMD have? The Volta train wont be pretty for them if they dont do anything.
Well unless Navi gets released in 2018 though, but when was the last time AMD delivered on the GPU front? Its been a trainwreck for many years now

 

[tajoh111]

It's really a damn shame that AMD chose deliberately to allow Nvidia to have the most powerful GPU during the Tesla and Fermi years. That's how insane the leadership of that organization has been.

Realizing they had an advantage in performance per watt and an even greater advantage in performance per mm^2 they chose the "Small-die" strategy that ultimated just cost them sales on the high end and damaged their mindshare into making them the "budget" company. So they leveraged their cheaper to make, faster, more efficient GPUs only up to a certain size, conveniently (for Nvidia) only to where they got to within 15% or so of their competitor's gargantuan top end chips.

As Head1985 said, if AMD make a big GPU from 2008-2011 they would have had the performance crown for all of those years. I understand their first attempt at a mammoth GPU was a relative failure; the 2900XT didn't cut it. I understand why they made the 3870 small. But the success of the chip should have given them pause for thought, and there was no excuse for 3 more generations of deliberately tiny chips.

As it was, they only had the temporary performance crown by launching their gen first a couple times.

It's as if Nvidia in current years never bothered releasing bigger chips than their x04 line, even if Hawaii/Fiji/Vega were around. But they instead fully capitalize on their advantages, top to bottom, and take a look at those earnings.

They were worried for the risk in the engineering process. Big chips are harder to design and always have lower yields. The gtx 480 shows this. A big monolithic die that did well in gaming would have got AMD the performance crown at the time, but because the architecture wasn't good at compute unless it was very simple operations, it wouldn't have done well in the professional market. And to make these big monolithic chips worth it, you need some professional market presence. AMD's Vega is proof of this.

The thing that allowed this risk to be more affordable to Nvidia was the fact they had a monopoly in the professional space and this is one of the reasons why AMD has not nearly generated the profits they could have during the 5870 series.

Nvidia professional cards in particular their gx100 series has been generating 200-250million a quarterly during 2008-2011. Because of the massive marketup, most of this is profit. Quadro p6000 = $5000-6000 = gp102.

This is why Nvidia's losses during the gx100 minimized. Nvidia's gaming lines have been cashflow drivers to pay for most of their expenses and the professional market is where the profit is generated.

This is why AMD's profits have not been where they could have been even during their prime years.

https://seekingalpha.com/article/22...folio-to-increase-its-footprint-in-the-market

In 2009, AMD's professional marketshare was only 12%. Since they numbers are usually in terms of cards sold and Nvidia higher selling price in the professional market, Nvidia was selling likely generating 10x the revenue of AMD in this market.

So if AMD professional market were like Nvidia's at the time, they would have been making 150-250 million in net profit quarterly. The lack of revenue from this segment is what prevented AMD from having big quarters from their graphic division. Take aways Nvidia's professional market profits and they have been doing worse than AMD for a great number of years prior to maxwell.

Nvidia's explosion in profits recently has been the culmination of succeeding in both the consumer graphic space and the professional one as well. Considering 1.1 billion minus the switch sales, comes from IP licencing and professional markets, it should be no surprise that Nvidia's profits are 800+ million.

 

[tviceman]

True, but what other choice does AMD have? The Volta train wont be pretty for them if they dont do anything.
Well unless Navi gets released in 2018 though, but when was the last time AMD delivered on the GPU front? Its been a trainwreck for many years now

What is out now doesn't mean what is in development will come sooner. Vega already sucks and it was obvious to everyone that had common sense this would be the outcome (although not even I thought it would be this bad). Volta is going to completely wreck AMD on the performance and performance per watt front and it will probably be out 6+ months before AMD can respond with a mid-range Navi product (even longer for a high end Navi product).

 

[tential]

What is out now doesn't mean what is in development will come sooner. Vega already sucks and it was obvious to everyone that had common sense this would be the outcome (although not even I thought it would be this bad). Volta is going to completely wreck AMD on the performance and performance per watt front and it will probably be out 6+ months before AMD can respond with a mid-range Navi product (even longer for a high end Navi product).

Volta already destroyed Navi, the funny thing is investors don't know it yet so we'll see a MASSIVE uptick of Nvidia stock again when AMD has no response and AMD'sGPU division will have terrible reports for all of 2018

 

[caswow]

there wont be any volta for consumer. consumers get ampere.

 

[Cloudfire777]

there wont be any volta for consumer. consumers get ampere.

All we know is that Huang is presenting Ampere in March. It could be the architecture after Volta presented in a roadmap too.

Sounds weird to have two architectures out at the same time if you ask me.

 

[caswow]

All we know is that Huang is presenting Ampere in March. It could be the architecture after Volta presented in a roadmap too.

Sounds weird to have two architectures out at the same time if you ask me.

it might be volta -hpc and a new name for the marketing buzz.

 

[Qwertilot]

All we know is that Huang is presenting Ampere in March. It could be the architecture after Volta presented in a roadmap too.

Sounds weird to have two architectures out at the same time if you ask me.

They already do that - V100 and P102 & friends - and have done for the best part of a year now

With how competitive and massive the compute/deep learning market is, you can be sure that 'son of V100' isn't that far away. More or less the moment there's a viable further die shrink from TSMC and friends I suppose.

 

[tviceman]

there wont be any volta for consumer. consumers get ampere.

What in the world makes you say that? Ampere succeeds Volta and is probably 18-24 months from coming to market, unless you think Ampere is coming in early-mid 2018 in consumer form and won't get a compute-only chip a la GP100.

I'm guessing Volta consumer cards comes in Q1 / early Q2 2018, and Ampere won't be out until sometime in 2019.

 

[amenx]

Seems gaming/consumer products are no longer the main focus of Nvidias business. Rapidly evolving company with new areas as significant sources of new revenue. Wonder if they will still be committed to gaming hardware a few years from now.

 

[HurleyBird]

A big monolithic die that did well in gaming would have got AMD the performance crown at the time, but because the architecture wasn't good at compute unless it was very simple operations, it wouldn't have done well in the professional market. And to make these big monolithic chips worth it, you need some professional market presence. AMD's Vega is proof of this.

Nvidia definitely had a compute advantage for Fermi. It's a bit harder to compare the Terrascale versus Tesla, but AMD did have a massive advantage in double precision throughput for that generation.

Of course, GPGPU was a significantly smaller market than it is today. Compute didn't subsidise the gigantic GT200 die in any substantial way.