Pat Gelsinger: “The drivers for our GPUs were more challenging than expected”

Intel Innovation was held this week in San Jose, California. The event is the successor to the IDF. Due to this event, it was possible to ask CEO Pat Gelsinger questions today, the questions and answers can be found below.


Pat Gelsinger, CEO of Intel.

How do you think the processor will evolve in the next five to ten years?

That’s a big question. You can see it from four or five sides. The first is the construction, which will increasingly be based on chiplets. Some parts of the processor will be based on the latest technology, while the I/O technology will be based on older technologies. The CPU and GPU will use the latest technology.

We see it as a multi-architecture world. In addition to CPUs, you will have GPUs and accelerators in many areas such as AI.

You also need techniques to share working memory between the different GPUs, processors and IPUs.

RISC V: Keep pushing your X86 as you always have, or do you want to also focus on the soc-cpu with RISC V, for example.

In the world there are different architectures, but in the mobile world ARM has won. You can also see that some markets are interested in RISC V. At Innovation, we had a demonstration based on RISC V, as a development platform. As Intel, we deal with different architectures: X86, ARM and also RISC V. We have the most experience and possibilities in X86.

Do the current energy supply problems in Europe affect building factories in Europe?

We are still waiting for the final stage of the formal EU approval of the construction of the plant. We expect to make the formal announcement shortly. Energy is not the biggest part of the cost, but we are concerned about inflation. The availability of energy is essential for us to run a factory, it takes twelve weeks to make a wafer and you can’t just stop that. All in all, the energy cost of a chip mill is not the biggest problem.

When will Intel’s entry into the GPU market be a success for you?

It now finally looks a little different than we expected. We thought the software for our integrated GPUs could scale up to our discrete GPUs. This turned out not to be the case and the development of the drivers took us much longer than we expected.

There is now also a bit of economic headwind. We launched the A770 this week and we feel comfortable with this product now. It will take us a few more years to build up the market in a new segment for us. In my 40-year career, I have never managed to capture a large market share with a first-generation product, so we have to be patient.

A few years usually pass between different notes. Now you want to start a large number of nodes in a short time?

In the transition from 14 to 10, we introduced five different techniques which created the necessary challenges. It was also connected to the fact that we did not want to switch to EUV. TSMC has done it and it has worked well for them.

If we go to Intel 4 and 3 we use the ‘ticktock model’. 4 then becomes the technically complex step, 3 becomes easier. We often work together with the Dutch ASML. When we asked them to send their best technicians to Oregon, it was no problem for them. Because at that time Taiwan was closed due to Covid measures.

The shortage is diminishing, but is it still necessary to invest from chip factories?

The economic conditions have caused the deficits to fall. The lack of older hubs is still very large, which still has major consequences for, among other things, the automotive industry.

We had previously said that the deficits would disappear in 2024, we now expect that in the second half of 2023. However, it is still crucial to spread the production of chips more around the world. Not only in Asia, but also in Europe and the USA. We must move from just in time to just in case.

Now that we’re seeing 24 cores in mainstream desktop processors, will there still be a market for high-end desktop processors?

I previously ran this business unit of Intel. It is a question that has been around for twenty years. Then we even wondered if we shouldn’t just make mobile processors. I don’t really see the market changing. You will always find the gamers who want more performance and users who have unique needs and therefore need a heavy duty workstation. The cloud is too slow for many applications and therefore no alternative. The bandwidth is too low and the latency is too high. 20 years ago we were talking about thin clients and everything had to be in the cloud. I can’t see changing the fact that you have the best experience with a desktop.

How does Intel see the do-it-yourself computer market?

There is no fundamental change. Partly due to the availability of certain products, there has been an inflationary effect which has caused prices to rise. We don’t see the market changing and we still see it as a unit of productivity. I can’t imagine holding online meetings on a laptop.

It is not a market where computers are replaced by telephones. People want phones and computers to work together. That’s why this week we introduced Intel Unison, which lets you connect a phone to a computer.

« Previous post Next post »

Leave a Comment