The SSD used by Pechinim is usually 0.5TB (512GB).

To become AI semiconductor G2
- Do well in planning the product.

Preparing 300TB SSD

The SSD used by Pechinim is usually 0.5TB (512GB). Making such a thing makes it impossible to export and the company's profitability deteriorates.

The SSD in the picture is 75TB. In fact, it should have come out like this before. If you were to be greedy about 100Gbps or 400Gbps, you should have designed the interface to be 400Gbps dual 800Gbps in both directions. In other words, memory and storage devices are connected through optical LAN at different server racks.

The same method should be added for main memory to make 1TB or 16TB HBM or rack size GDDR6 memory modules.

The server will consist of GPUs or NPUs and CPUs, while the primary main memory will go to the server side and the secondary main memory will go to the memory and SSD connected to the optical LAN. The optical interface will go from Tbps to Pbps. There will be a revolution in artificial intelligence semiconductors. Something like CXL, UCIe, PCIe, should revolutionize in this direction.

This becomes an artificial intelligence server equipped with a true high-speed interface. In this case, separating the storage, memory server, and operation server makes it easier to manage the server's heat. This idea is a $10 trillion product planning and invention, and I specifically open it on Facebook. I don't want to make it, but after what Samjeon or Hynix will do.

If this comes out later, it's good to treat them to a meal...

Do those who plan memory products have fun? It's frustrating. I have 40 years of experience in system design.

I think this article should be rewritten later by adding a connection system block diagram. I think the general engineer will misunderstand that it is not possible.

SOXX, QQQ / SOXL TQQQ
Why is SOXL stepped on TQQQ?

March 2016 is the day AlphaGo beat Lee Se-dol. In 2016, there was a need for artificial intelligence semiconductors to perform high-speed operations. How fast the calculation is performed has eventually become a way to improve the performance of artificial intelligence. Eventually, demand for semiconductors has soared since 2016. The semiconductor industry has been changing around 2016. From low growth to high growth... So, wouldn't data after 2016 be important?

In the graph, QQQ does not catch up to SOXX.
However, TQQQ is actually a higher price than SOXL.

A. Why is that?
1. If you look at the 2016-01-01 start value, TQQQ is high. That's why it's high so far.

2. There are many ball drags because it is a three-fold leveraged product.

B. Conclusion:
1. When SOXL rises sharply, it's better to be unfamiliar with the weight.

2. In other words, if you sell SOXL and change it to TQQQ... If the stock goes down, it's better to increase the SOXL weight.

3. When both stock prices fall and are at a low point, it is better to have a higher proportion of SOXL than TQQQ.

4. If index 2 of the sixth figure is exceeded, it should be cashed in because the market is overheated.
    If you buy a stock below index 1.5, it's going to be a big hit
    In eight years, there have been four times under Interx 1.5.

5. This is the beginning of an overheating situation, so it's time to enjoy the profits... but sooner or later the tide is coming.

6. If this works ... it's a market forecast.

7. It may be that the semiconductor leverage rises high in the sky, which means that the market is crazy.

8. If the semiconductor leverage becomes 2-2.5 times that of QQQ leverage TQQQ, the proportion of bonds with a high cash share should be increased.

I think that's what I expect.
Or not.

#SSD #HBM #GDDR6 #GDDR10 #Peta #Tera #Giga #Pbps #Tbps #Gbps #PCIe #UCIe #CXL #OPTIC