I saw somewhere the memory price alone was around $20,000 depending on how much was used.

When are we expecting this new wave to hit consumer market?

"The more you buy, the more you save." - Leather jacket man.

No wonder the company is worth 2 trilion 😀 I myself contributed with a small ammount over the years since riva TNT 2 days

Yes, I'm sure the $10B comes from the report submitted to the taxation authority, for maximum tax deductions. Might as well then use the number in speeches and interviews.

people seem to not notice Nvidia's elimination of (A.I.) consumer choices by opting to only sell complete servers. to me that's 100% having your cake and eating it too. and a guarantee of profitability for Nvidia's weakest link - their ARM servers. not that those servers are bad, but that they aren't the strength of the company. imho, this is all about AMD's MI300 (and the inclusion of Zen cores)

Pryme:

"The more you buy, the more you save." - Leather jacket man.

The article says the lack of supply makes the price very high. More demand could mean even higher price!

I love statements like "it's X$ R&D money per GPU"... That's totally inaccurate and BS. Since I don't buy their 10 billion dev costs for it. Yeah, maybe since Nvidia exists, or since they first designed their CUDA core... if they spent 10 billion since the A100 was released they would never be profitable. So there's the first blatant lie. Second, the price doesn't even depend on R&D money only... if they sell a billion chips, R&D money is not even a reasonable split of the cost to be mentioned... if they sell 100 chips, of course it's a big part of costs per SKU. And to see that they already know the costs, no matter how many are sold, tells you outright that any R&D costs are not having a direct impact on the SKUs price. It's the market situation of competition, expected sales etc.

XenthorX:

When are we expecting this new wave to hit consumer market?

But more importantly I am really looking forward to NVidia's new AI drivers. Yes, one time only install drivers, no updates required The ultim8 Silver bullet for shity optimised games and game engines alike. Leading to double performance and dramatic energy consumption reductions as heavy lifting becomes a thing of the past. 🙄

fantaskarsef:

I love statements like "it's X$ R&D money per GPU"... That's totally inaccurate and BS. Since I don't buy their 10 billion dev costs for it. Yeah, maybe since Nvidia exists, or since they first designed their CUDA core... if they spent 10 billion since the A100 was released they would never be profitable. So there's the first blatant lie.

I don't see anyone in the interview making this statement? The cost of development and the cost that they plan on selling the GPU for is two completely different parts of the interview. Why is $10B a lot for them over 4 years? They did $60B in revenue last year. Nvidia' financials say they spend ~$8.7B a year on R&D currently. That's for the entire company though. Given that this product is a massive chunk of their revenues, I think it's probably safe to say that $10B is accurate for just this chip. I have to imagine that it's been in development for 3-4 years. $2.5B a year for this chip, probably scaled towards the backend due to software dev costs, doesn't seem that crazy.. that's only 4% of their revenue last year.

fantaskarsef:

Second, the price doesn't even depend on R&D money only... if they sell a billion chips, R&D money is not even a reasonable split of the cost to be mentioned... if they sell 100 chips, of course it's a big part of costs per SKU. And to see that they already know the costs, no matter how many are sold, tells you outright that any R&D costs are not having a direct impact on the SKUs price. It's the market situation of competition, expected sales etc.

Again, i don't see Huang, in any statement, justifying the price by simply dividing the number of chips by the R&D cost. Typically the BOM costs ($3500 or whatever) they give don't include R&D.

Denial:

I don't see anyone in the interview making this statement? The cost of development and the cost that they plan on selling the GPU for is two completely different parts of the interview.

They wouldn't mention it if it wasn't meant in relation to sales price. Might be wrong on my take, but why should they mention it then?

Why is $10B a lot for them over 4 years? They did $60B in revenue last year. Nvidia' financials say they spend ~$8.7B a year on R&D currently. That's for the entire company though. Given that this product is a massive chunk of their revenues, I think it's probably safe to say that $10B is accurate for just this chip. I have to imagine that it's been in development for 3-4 years. $2.5B a year for this chip, probably scaled towards the backend due to software dev costs, doesn't seem that crazy.. that's only 4% of their revenue last year.

How many engineers do you need for this product? Because 2.5 billion are a lot of money to throw out to engineers designing, and materials building prototypes etc. that's what I tried to convey.

Again, i don't see Huang, in any statement, justifying the price by simply dividing the number of chips by the R&D cost. Typically the BOM costs ($3500 or whatever) they give don't include R&D.

This is true, but as mentioned above, my understanding was that they mentioned it to "justify" the cost of the product or at least put the named sum (R&D costs) in relation to the product's price. Might be a misinterpretation though, I did not read the interview (only the news post here).

fantaskarsef:

They wouldn't mention it if it wasn't meant in relation to sales price. Might be wrong on my take, but why should they mention it then?

Jim Cramer asks him "How much is the product going to cost" then he gave a number. Jim Cramer then asks him "How much did it cost to develop" and he gave him a number lol He didn't say something like "It's a $30,000 chip because it cost us $10B to develop. Maybe that's what Jim was trying to convey but Jens didn't answer the question like that, it's two separate things. The cost is obviously way higher than R&D given sales - which is why R&D is only like 10% of Nvidia's revenue.

fantaskarsef:

How many engineers do you need for this product? Because 2.5 billion are a lot of money to throw out to engineers designing, and materials building prototypes etc. that's what I tried to convey.

It's also going to include software development costs for the platform, tape outs from the foundry, Nvidia's custom cell libraries from the foundry, hardware costs from companies like Cadence that probably costs nearly a billion itself to do the prototyping, etc. Keep in mind Nvidia has 26,000 employees. Most of those people, software/computer/electronic engineers are all making over 100k, some of them probably ~200k.

Denial:

software/computer/electronic engineers are all making over 100k, some of them probably ~200k.

most of them are starting at 100k. the cost of living is astronomical here you make good points but there is a level of Jensen being disingenuous

Denial:

Keep in mind Nvidia has 26,000 employees. Most of those people, software/computer/electronic engineers are all making over 100k, some of them probably ~200k.

Salary wise that would be the entry level for the engineers if looking at salaries for talented people in similar Big Tech companies.

Denial:

Jim Cramer asks him "How much is the product going to cost" then he gave a number. Jim Cramer then asks him "How much did it cost to develop" and he gave him a number lol He didn't say something like "It's a $30,000 chip because it cost us $10B to develop. Maybe that's what Jim was trying to convey but Jens didn't answer the question like that, it's two separate things. The cost is obviously way higher than R&D given sales - which is why R&D is only like 10% of Nvidia's revenue.

I did not read the interview, and I should have. Thanks for making me aware of this, since I am wrong on my assumption and it changes the context.

It's also going to include software development costs for the platform, tape outs from the foundry, Nvidia's custom cell libraries from the foundry, hardware costs from companies like Cadence that probably costs nearly a billion itself to do the prototyping, etc. Keep in mind Nvidia has 26,000 employees. Most of those people, software/computer/electronic engineers are all making over 100k, some of them probably ~200k.

Well I did think about software too, yes. But the suite to design this chip didn't have to be paid in full to be developed again (probably subscription costs for company licenses will be a big sum here too), so the actual cost is at best, 4 years since the predecessor's finishing it's design. Or a couple of more years, with maybe a few million to pay for software licenses. Maybe double digit millions. So we're in the single digit percent range for the given sum, I presume. The custom cell libraries wouldn't be too expensive, I think? I mean sure, millions of $, but I do think about the same as with software, single digit percentages? I do take that billion there, those 10% I can easily imagine going down the drain just to get the machines to build what you actually designed for the first time. As for employees, let's just assume that not everybody's work force and time is 100% going into this product, since at 26k employees x 4 years make 15.6 billion. But essentially, to sum up 10 billion, minus 1 for hardware, and say another 0.75 for software and tape out and libraries, leaves 8.25 billion. 2b a year for only human brain power, ~140k as a very conservative estimate of the average income would leave me with almost 3600 employees, full time working on the product, and exclusively. So yeah... I guess it mostly comes down to wages and how many engineers are working on it, exclusively. I just found it an exaggerated cost and I would not expect it to be 10 billion really, really, only, 100% for this product. I could imagine this to check out, now that you made me think it through. Thank you for making me reconsider.

Is this report about an interview in the future? I ask because it says the interview was "post-GTC 2024". Well GTC 2024 is happening in San Jose, CA right now so it can not possibly be "post-GTC 2024." Here is the schedule for it : https://www.nvidia.com/gtc/conference-schedule/

pharma:

Salary wise that would be the entry level for the engineers if looking at salaries for talented people in similar Big Tech companies.

Hold on, didn't people shed a tear when the redundancies hit a few months ago, yeah lets all feel sorry for the 100k a year new guys ha ha ha ha

pegasus1:

Hold on, didn't people shed a tear when the redundancies hit a few months ago, yeah lets all feel sorry for the 100k a year new guys ha ha ha ha

when you're paying +$40k a year for rent and you still have massive student debt i feel sorry for them. a lot of folks used to look at getting hired here like winning the lotto...when it's more like a reality show, once you're on it isn't all that much money compared with living in Hollywood - and interns get the shaft in both places because cost of living with no income

tunejunky:

when you're paying +$40k a year for rent and you still have massive student debt i feel sorry for them. a lot of folks used to look at getting hired here like winning the lotto...when it's more like a reality show, once you're on it isn't all that much money compared with living in Hollywood - and interns get the shaft in both places because cost of living with no income

Yep, 30 years ago it was worth moving for, not now. I trained there and it was costly. Parties in SanFran were great though 🙂

pharma:

Salary wise that would be the entry level for the engineers if looking at salaries for talented people in similar Big Tech companies.

I was confused, you wrote in the quote of my post, I haven't talked salary**