I was surprised to hear CPUs only have a number of ALUs. What are most of the transistors in a CPU dedicated to?
How you spend your transistors really depends on the problem you're trying to solve.
If you open up a 1 billion-transistor x86 processor, you'll see a handful of very complex CPU cores, with many transistors dedicated to decoding and scheduling instructions, predicting branches, and managing virtual memory. You'll also see many transistors dedicated to very fast cache memories. This mix of transistors reflect the general purpose, varied workloads a typical Windows or Mac might encounter.
If you open up a 1 billion-transistor SPARC server processor, you'll find a bunch of fairly simple in-order SPARC cores and a ton of cache. These guys are optimized for many parallel database threads operating on huge database working sets.
If you open up a 1 billion-transistor GPU processor, you'll see hundreds of very specialized, deeply pipelined floating point compute datapaths, fine-grain thread schedulers, and memory transfer engines designed to stream texture, vertex and other data through the processor efficiently. There isn't much branch control logic, but a ton of matrix-friendly floating point compute hardware, along with RAM to buffer the data as it passes through.
If you open up a 1 billion-transistor DSP optimized for cellular base station applications (something I worked with in a previous job), you'll find multiple dedicated signal processing blocks for handling various over-the-air encoding protocols, a large network-accelerating block for handling millions of packets whizzing through the device at microsecond latencies, some highly capable general purpose DSPs dedicating most of their transistors to multiply-accumulate operations, and some ARMs to schedule everything. The transistors there are split maybe 50-50 between compute/data movement and memory. (Or maybe closer to 40-60.)
If you open up a 1 billion-transistor cell phone processor, you'll find the most varied mix of all. You'll see a few general purpose ARM processors (most likely), some GPUs, some video encode/decode accelerators, maybe some DSPs, some networking peripherals, and many other peripherals for managing USB, Bluetooth, WiFi, and interfacing to cameras, mics and speakers.
It all depends on what you're trying to do...
A billion transistors seems like a lot, but a trillion is now the record to beat, or at least a common ground to start from. Granted this is only for the most demanding applications. Analog and RF IC's have no desire to be so complex, not to mention the cost. The following blog is amazing that it happened so soon. From ExtremeTech.com
Cerebras Systems Unveils 1.2 Trillion Transistor Wafer-Scale Processor for AI By Joel Hruska on August 20, 2019 at 9:00 am
Modern CPU transistor counts are enormous — AMD announced earlier this month that a full implementation of its 7nm Epyc “Rome” CPU weighs in at 32 billion transistors. To this, Cerebras Technology says: “Hold my beer.” The AI-focused company has designed what it calls a Wafer Scale Engine. The WSE is a square, approximately eight inches by nine inches, and contains roughly 1.2 trillion transistors.
I’m genuinely surprised to see a company bringing a wafer-scale product to market this quickly. The idea of wafer-scale processing has attracted some attention recently as a potential solution to performance scaling difficulties. In the study we discussed earlier this year, researchers evaluated the idea of building an enormous GPU across most or all of a 100mm wafer. They found that the technique could product viable, high-performance processors and that it could also scale effectively to larger node sizes. The Cerebras WSE definitely qualifies as lorge large — its total surface area is much larger than the hypothetical designs we considered earlier this year. It’s not a full-sized 300mm wafer, but it’s got a higher surface area than a 200mm does.
The largest GPU, measures 815 square millimeters and packs 21.1B transistors. So the Cerebras WSE is just a bit bigger, as these things go. Some companies send out pictures of their chips held up next to a diminutive common object, like a quarter.