NVIDIA Announces Q4 and Fiscal 2024 Results, Clocks 126% YoY Revenue Growth, Gaming Just 1/6th of Data Center Revenues

[pk67]

That is a quote from September 2022, way before any AI hype started.


And about the logic it's okay to pay the same price increase as the preformance increase - by that logic the preformance increase would now bring is to GPUs costing tens of thousands of dollars, starting from the early 90'.

Don't believe me? Just take a simple approx. 50% performance, 50% price increase from now on, the same we have this generation:

2020, RTX 3080 - $700
2022, RTX 4080 - $1200
2024, RTX 5080 - $2040
2026, RTX 6080 - $3468
2028, RTX 7080 - $5896
2030, RTX 8080 - $10022
2032, RTX 9080 - $17038
2034, RTX 1080 - $28965

At what point will you stop?

If you dont like these prices above you can always go for AMD or Intel stuff or start design your own like chinese developers do now. Are Nvidia stick the gun to your head an asking you for paying these prices ? It is not the case - it is calling free market instead. If you dont need discrete GPU in 24/7 regime you can still buy APU with integrated iGPU and NPU. These toys will be very powerfull way earlier before 2034. If you think these prices are unjustified you can still desingn your own hardware and software and put your orders to leading manufacturers - it is easy as 1 2 3. If you think manufacturers have unjustified high prices - you can build your own fab. But very high NVidia gross margins telling now otherwise.

 

[kaktus1907]

As Raja Koduri Says. "Developers on PC GPUs are the key enablers to DC GPU success. So all the dev tools need to work flawlessly on PC GPUs. Currently this is largely true with Geforce. I still get dirty looks from my engineers when I ask them to swap away their Geforce card. Engineers cost way more than any GPU pricing delta, so not worth messing with their productivity.

Gaming GPUs are so accessible globally. Just to give an example, engineers in smaller cities in places like India that don't have a path to access DC GPUs, can walk to their local PC reseller and buy a Geforce card and start coding!

Consistent architecture and driver stack across PC and DC GPUs will be a huge advantage. This is much overlooked disadvantage of dedicated AI accelerator eco-system. "

Gaming market is the hook for younger talent and marketing for your hardware to any pc on earth. Nvidia knows the importance of both seeding academia/institutes & making it as widely accessible as possible via unified code/software with gaming GPUs & datacenter accelerators. And they don't do this for charity, still earning steady and sizeable revenue of income.

 

[A&P211]

Nvidia's stock price 15% today. This is thing is insane.

 

[cvaldes]

NVDA's $250 billion increase today is almost the entire amount of AMD's market capitalization.

NVIDIA Gains Almost the Entire Market Cap of AMD Following Stellar Earnings and a Cascade of Stock Price Upgrades

NVIDIA has added nearly $250 billion in market cap, which just about equals AMD's entire market-based valuation of $265 billion.

wccftech.com

 

[Random_User]

Don't be that happy about gaming branch being stable. That's all due they had the stocks of cards made pre-AI-fever. Giving the trend, the next generation of GPUs will be even more enterprise oriented, so will be the margins. The avarege gamer Joe and Janes might not be able to buy even xx60.

Now, is there any doubt's literal gouging? The sad thing AMD continues this trend, and keeps their consumer prices to be as high as possible as well. If there's no collusion, then what is this?

They prefer maximizing profits from old silicon/Gpu's

China’s special graphics card RX 6750 GRE is too cheap! AMD takes action: fines, suspension of goods

At the end of January, AMD issued an internal notice to AIB brand manufacturers and dealers, requiring that the prices of RX 6750 GRE 10/12GB must be strictly controlled, with the lowest prices being 2,149 yuan and 2,379 yuan respectively.

That's interesting, how lightningly fast AMD reacted to the "cheaper" prices, while there are hundreds of thousands cards in the world, being sold with 200-300% margins above MSRP, on daily basis, for years. Since the first major mining boom in 2017-18. And they did not make a single move. Disgusting.

 

[Bwaze]

Don't be that happy about gaming branch being stable. That's all due they had the stocks of cards made pre-AI-fever. Giving the trend, the next generation of GPUs will be even more enterprise oriented, so will be the margins. The avarege gamer Joe and Janes might not be able to buy even xx60.

We don't even know if "Gaming" even means selling gaming cards - let alone selling gaming cards to gamers.

Nvidia was caught redhanded when they transferred revenue during cryptocraze so that it showed growth in all sectors, even Automotive, but then it all plummeted when the cryptoreality struck. But when the shareholders tried to sue Nvidia judges were quick to tell them it's not illegal.

 

[pk67]

Don't believe me? Just take a simple approx. 50% performance, 50% price increase from now on, the same we have this generation:

2020, RTX 3080 - $700
2022, RTX 4080 - $1200
2024, RTX 5080 - $2040
2026, RTX 6080 - $3468
2028, RTX 7080 - $5896
2030, RTX 8080 - $10022
2032, RTX 9080 - $17038
2034, RTX 1080 - $28965

At what point will you stop?

OK. Lets try some thought experiment. Let assume Huang is right, More Law is dead for good and whole industry almost reached the "wall" and any progress is really slow.
Let assume hipotetical price scenario without AI rush.
2020, RTX 3080 - $700
2022, RTX 4080 - $1200
2024, RTX 5080 - $2040
2030, RTX 6080 - $2468
2038, RTX 7080 - $2896
2050, RTX 8080 - $4022

Again lets compare it to yours scenario.
2020, RTX 3080 - $700
2022, RTX 4080 - $1200
2024, RTX 5080 - $2040 ================= --- 4080 Super - $1000
================= 2026, RTX 6080 - $3468 --- 4080 Super - $800 --- 5080 Super - $1740
================= 2028, RTX 7080 - $5896 --- 4080 Super - $700 --- 5080 Super - $1540 --- 2028, 6080 Super - $2918
2030, RTX 6080 - $2468 --- 2030, RTX 8080 - $10022 - - End of life 4080 - --- 5080 Super - $1340 --- 2030, 6080 Super - $2518 --- 7080 Super - $5196
================= 2032, RTX 9080 - $17038 --- 8080 Super - $8422 - - End of life 5080 - --- 2032, 6080 Super - $2218 --- 7080 Super - $4696
================================= --- 8080 Super - $7022 --- 9080 Super - $14438 --- 2034, - End of life 6080 - --- 7080 Super - $4296 ================================= --- 8080 Super - $5922 --- 9080 Super - $12038 --- ============== 2036, - End of life 7080
2038, RTX 7080 - $2896 =================== - - - End of life 8080 --- 9080 Super - $10040 --- 2038
2050, RTX 8080 - $4022
2065, RTX 9080 - $5538

My simple question is which one scenario is better for gamers ?

In slow progress scenario first availability of RTX 8080 is year 2050 at moderate price level $ 4022 but in AI rush scenario it is available 20 years earlier !! - the price is really high at the moment of premiere but 6 years later it dropped to $5922 for Super incarnation.

IMHO slow progress scenario is much worse for gamers than fast pace scenario even at exploding prices as in your example.

 

[cvaldes]

The best scenario for gamers is to buy some NVDA shares and sell one or two when you need to buy a new graphics card.

 

[pk67]

The best scenario for gamers is to buy some NVDA shares and sell one or two when you need to buy a new graphics card.

I'm not sure if they are considerably more safe than BTC now. Renting some computing power is even better and safer option for gamers to get their money back.

 

[cvaldes]

I'm not sure if they are considerably more safe than BTC now. Renting some computing power is even better and safer option for gamers to get their money back.

There's no infrastructure or system for that type of activity. You can't just post to Craigslist saying "RTX 4090 for rent: AI workloads welcomed when I'm not playing CoD".

It's far more primitive than cryptomining which was essentially for peasants.

And some of these AI training models can be massive. There's a lot of other resources that need to be set up to provide AI workload services.

And remember, most of the world's fortunes came from investments. Buying NVDA is a better investment than buying Nvidia hardware and trying to pawn off some GPU/Tensor core cycles. You don't get rich driving a delivery van for Amazon, you get rich trading AMZN shares.

 

[pk67]

There's no infrastructure or system for that type of activity. You can't just post to Craigslist saying "RTX 4090 for rent: AI workloads welcomed when I'm not playing CoD".

It's far more primitive than cryptomining which was essentially for peasants.

And some of these AI training models can be massive. There's a lot of other resources that need to be set up to provide AI workload services.

And remember, most of the world's fortunes came from investments. Buying NVDA is a better investment than buying Nvidia hardware and trying to pawn off some GPU/Tensor core cycles. You don't get rich driving a delivery van for Amazon, you get rich trading AMZN shares.

You can rent your rig but in 24/7 regime. After year or two you can use your rig for gaming. It is not secretive knowledge.

 

[cvaldes]

You can rent your rig but in 24/7 regime. After year or two you can use your rig for gaming. It is not secretive knowledge.

You're still footing the bill for retail electricity rates which takes a major bite of profitability. Consumers don't have access to the commercial electricity rates that the pros get.

That was one of the biggest reasons why cryptomining ended up being for peasants. It wasn't the cost of the hardware, it was the cost of the electricity.

Even today, if you look at the Distributed Computing section of the TPU forums, you'll see that participants are struggling with electricity costs. I mentioned that donating money to the charity is a more effective use of cash than burning up CPU/GPU cycles folding proteins, whatever. Because again, a large organization can buy electricity at favorable rates.

By renting out your rig for AI work, you are essentially destroying margin by shelling out for consumer electricity rates.

What you are talking about is for peasants. Yes, it can be done but there are better things to do with money that'll generate more return than renting out your rig.

 

[pk67]

You're still footing the bill for retail electricity rates which takes a major bite of profitability. Consumers don't have access to the commercial electricity rates that the pros get.

That was one of the biggest reasons why cryptomining ended up being for peasants. It wasn't the cost of the hardware, it was the cost of the electricity.

Even today, if you look at the Distributed Computing section of the TPU forums, you'll see that participants are struggling with electricity costs. I mentioned that donating money to the charity is a more effective use of cash than burning up CPU/GPU cycles folding proteins, whatever. Because again, a large organization can buy electricity at favorable rates.

By renting out your rig for AI work, you are essentially destroying margin by shelling out for consumer electricity rates.

What you are talking about is for peasants. Yes, it can be done but there are better things to do with money that'll generate more return than renting out your rig.

It depend if you are lanlord with your own electricity generation or just tenant. Sure you can selling drugs if you need more money faster. But renting your rig is very good option if you have access to affordable electricity to get your money back. If you are rich enough to buy dedicated rig for gaming but still crying gpu is too expensive - i'm showing you it is not the case. Maybe electricity at your location is too expensive but not GPU - cos other guys are ready for that price.

 

[cvaldes]

Well if it works for you, you are definitely part of a minority. Most people who can afford high-end consumer graphics cards live in major metropolitan areas where the electricity isn't particularly cheap. This is even more the case for Europeans with the energy crisis resulting from the invasion of Ukraine.

Expensive electricity will just destroy gross margin on your GPU rental activities. Again, this is all peasant stuff.

This is like saying that farming corn is good because diesel for the tractor is cheap where you live. It's still better to trade corn futures than drive a tractor.

And Nvidia stock is pretty volatile with a beta of 1.68 according to Yahoo Finance. So opportunities to buy low and sell high happen very frequently. If you bought NVDA on Tuesday and sold yesterday (Thursday), you could have walked away with +15% easily. Ten grand in investment capital would have grossed $1500, and even after taxes, you would have enough for a 4080 SUPER. In less than 48 hours.

 

[pk67]

Well if it works for you, you are definitely part of a minority. Most people who can afford high-end consumer graphics cards live in major metropolitan areas where the electricity isn't particularly cheap.

If you are REAL gamer just change your location into more friendly place like Argentina or Brazil. In Brazil you will have a lot of cheap solar energy and hydro as well. No problems with freezing in winter season.
Metropolitan areas are fine for busy people who dont have a lot of time for gaming or for awfully rich ones who dont care electricity bills or pennies they have to spend for latest most powerfull GPU available for sale.

 

[A&P211]

The best scenario for gamers is to buy some NVDA shares and sell one or two when you need to buy a new graphics card.

Each share will cost around a 4080. It sits at $788 atm.

Well if it works for you, you are definitely part of a minority. Most people who can afford high-end consumer graphics cards live in major metropolitan areas where the electricity isn't particularly cheap. This is even more the case for Europeans with the energy crisis resulting from the invasion of Ukraine.

Expensive electricity will just destroy gross margin on your GPU rental activities. Again, this is all peasant stuff.

This is like saying that farming corn is good because diesel for the tractor is cheap where you live. It's still better to trade corn futures than drive a tractor.

And Nvidia stock is pretty volatile with a beta of 1.68 according to Yahoo Finance. So opportunities to buy low and sell high happen very frequently. If you bought NVDA on Tuesday and sold yesterday (Thursday), you could have walked away with +15% easily. Ten grand in investment capital would have grossed $1500, and even after taxes, you would have enough for a 4080 SUPER. In less than 48 hours.

I started buying nvda some time ago in the $250 range, my last buy in was at $493 in December. I just sold everything on Thursday after their earnings call. NVDA has gotten too rich for me. As of right now, NVDA cut 2.5yrs off from early retirement. Before earning, it was 2yrs early, now its 2.5yrs. Thanks Nvidia, its been fun but time to put the gun back into its holster and safer stocks like index funds.