Welcome To A World Where NVIDIA Makes Chips And Intel Never Existed … Plus NVIDIA MGX
NVIDIA MGX Is Real; The Other Headlines Written By A Hallucinating AI?
Welcome to a strange new world, at least according to the headlines popping up everywhere that NVIDIA is the first chipmaker to be valued at a trillion dollars. Ignore the fact that the sharp spike in the GPU companies stock price is driven by the current AI craze and parse the rest. NVIDIA does not have any fabs, their chips are currently made by TSMC and have been fabbed at several other companies over the years so to call NVIDIA a chip maker is a bizarre statement. As to NVIDIA’s trillion dollar worth, that is only accurate if you are talking specifically about market cap, something with no bearing on reality at all.
What does reflect reality is that NVIDIA made $7.19 billion in total revenue this quarter, while Intel made $11.7 billion and that reflects a horrific quarter for Intel and a better than expected quarter for NVIDIA. There is no way for NVIDIA to scale up it’s production to come anywhere close to what the market cap suggests, they don’t have the people and more importantly TSMC doesn’t have the space on it’s production lines to make that many chips for NVIDA.
The new NVIDIA MGX open sever spec on the other hand is very real and could be very popular in the near future. The current designs for server chassis were designed many years ago and were built for chips both larger and less power hungry than today’s silicon. The design also couldn’t take into account HPC cards, seeing as how they didn’t exist. NVIDIA has stepped up with a solution, a modular design for racked servers which is designed specifically to handle large amounts of HPC hardware. The modular design will make it far less expensive to build, while offering a wide variety of designs to accommodate the specific hardware that will need to be cooled and powered.
Considering that ASRock Rack, ASUS, GIGABYTE, Pegatron, QCT and Supermicro have all signed on with NVIDIA MGX and will be producing chassis for customers later this year. It will be nice to see some new designs mounted on racks in the server room, the old designs are getting very dated.
Nvidia has created an open design for servers to house its accelerators, arguing that CPU-centric server designs aren't up the job of housing multiple GPUs and SmartNICs.
More Tech News From Around The Web
- Android apps with spyware installed 421 million times from Google Play @ Bleeping Computer
- This is the 2U 4-Node NVIDIA Grace Superchip ASUS Server @ Serve The Home
- Canon Develops Quantum Dot OLED Materials Without Rare Metals @ Slashdot
- HP printers should have EPEAT ecolabels revoked, trade group demands @ Ars Technica
- Intel says AI is overwhelming CPUs, GPUs, even clouds – so all Meteor Lakes get a VPU @ The Register
- ASUS Shows Off Concept GeForce RTX 40 Graphics Card Without Power-Connectors, Uses Proprietary Slot @ Slashdot
- Your Engineering Pad In Browser @ Hackaday
Who exactly is this article written for? The valuation of a company is entirely based on stock price. Intrinsic value is what you might be trying to convey here. Then you go on to say that Nvidia isn’t a Fab but talk about their lack of ability
to ramp up production. Of coarse not taking into account that chip makers all over the world are adding capacity. Yes that won’t come online for a few years, intel is not a fab either so I’m not sure where they will be getting the capacity to ramp up production. I could go on, but I think you see at this point there is no need.
Your comment that “the valuation of a company is entirely based on stock price” is obvious, and Jeremy mentioned this in the first paragraph:
Next, Intel IS a fab! What on earth are you talking about? Intel has their own fab, AMD and NVIDIA are fabless semiconductor companies, not technically “chipmakers”. I get that this term is just the accepted one by media to describe tech companies that design chips, so I’ll leave that alone. Apple doesn’t make their own computers, but they are called a computer company, for example.
“There is no way for NVIDIA to scale up it’s production to come anywhere close to what the market cap suggests, they don’t have the people and more importantly TSMC doesn’t have the space on it’s production lines to make that many chips for NVIDA.”
This relies on two incorrect asumptions:
First, that Nvidia is wholely reliant on TSMC for fabbing – they also work with Samsung, and intend to work with Intel to fab chips. Strike one.
Second, that ‘number of chips’ (or being charitable, square die area fabbed) equates directly and equally to revenue. Also not the case, as that relies on the assumption that price charged per die area is equal across all vendors. That is also not the case, and price charged per die area is not even equal within a single product line!
Samsung seems to be already running at 80-90% capacity, and that won’t change for at least a year. So the point still stands, there is no way in hell they can increase production 33 fold to hit a trillion dollars of income in a year.
Your second point is a straight out straw man, ain’t no one talking about dollar value per square die but you.