LinuxQuestions.org - Why don't cpu's have a chipset inside?

- Linux - Hardware (https://www.linuxquestions.org/questions/linux-hardware-18/)

- - Why don't cpu's have a chipset inside? (https://www.linuxquestions.org/questions/linux-hardware-18/why-dont-cpus-have-a-chipset-inside-4175507408/)

Why don't cpu's have a chipset inside?

Why is the chipset always outside the cpu? Would't things be simpler cheaper and more productive if the chipset was in the same die as the cpu?

Quote:

Originally Posted by Ulysses_ (Post 5184534)

Why is the chipset always outside the cpu? Would't things be simpler cheaper and more productive if the chipset was in the same die as the cpu?

No. But you probably want more.

First of all, at the silicon wafer stage, if you get 15% good cpus out of a wafer that's a very good return. We already have cpu/gpu(=apu) combinations. Your success rate there would be 15/100 x 15/100 = 2.25%; or else take on the challenge and expense of manually connecting cpu & gpu. And at silicon level, you can just solder on a wire, because all to easily that junction becomes a schottky diode. Now you want to throw in a chipset as well (0.3375%)?

Secondly, Think software. Software has to control stuff. Banging them all in the same package isn't going to make that any faster, perhaps slower. And software writers require control over the data exchange.

Thirdly, think heat. You figure that one.
Fourthly, think size. The southbridge has all the peripherals and buses. connected. The Northbridge has all the memory, and the southbrigde and cpu & Gpu. If you came out under 600 pins, you be doing very well indeed.

They do make packages that contain more than one discrete part. It would be possible to create larger chips that have more on it and they do have that also. It is a computer on chip sort of design or very large scale integrated device.

Companies that make these parts have a huge investment in both plants and sales. They work at making profit as best they can. In some cases they might spend millions on a new design only to have no market for it. If they had a chip that would be sold in billions over years or decades then they could afford to gamble more.

With the same logic business_kid, if a cpu with one core has 15% success rate, then a cpu with 4 cores has 15/100 x 15/100 = 2.25% success rate. But this is not so. Because this is not separate dies but a single die with everything inside.

Heat is not an issue at all, almost all the consumption in a typical PC goes to the cpu and it's only a few watts for the motherboard.

Regarding pin count, what about putting just the north bridge in the same die?

Ulysses you talk about heat and watts in the same sentence. sorry but just because the CPU draws limited watts does not mean it does not generate massive amounts of heat. A typical modern CPU, be it AMD or Intel, dual, quad, or more cores, will run between 70-90C, sorry mate but thats hot. try running your CPU under load without a heatsink and see just how long it lasts. Guess what, its still only drawing a few watts, but the heat is there.

Watts is Joules per second, or in other words work done in a time frame.

you can also say that 1 watt = Volt Ampere or (kgm^2)/ (As^3) that would be kilogram meter squared over Ampere per cubic second. Again it is work done over time.

Please do not confuse those two. If I perform X amount of work in 1 second vs 10 seconds I've still done the exact same amount of work, but they do not produce the same Wattage, nor will they generate the same amount of energy, in the case of the CPU heat. A low watt consumption does not mean the CPU will not generate large amount of heat.

https://en.wikipedia.org/wiki/Northb...28computing%29

Quote:

On older Intel based PCs, the northbridge was also named external memory controller hub (MCH) or integrated memory controller hub (IMCH) if equipped with an integrated VGA memory controller hub (MCH). Increasingly these functions became integrated into the CPU chip itself, beginning with memory and graphics controllers. For Intel Sandy Bridge and AMD Accelerated Processing Unit processors introduced in 2011, all of the functions of the northbridge reside on the CPU,[2] while some high-performance CPUs still (2013) require northbridge and southbridge chips.

There is no reason to integrate the southbridge into the CPU, because it is slow anyway. Do you have a reason to ?

Quote:

Originally Posted by Ulysses_ (Post 5184534)

Why is the chipset always outside the cpu? Would't things be simpler cheaper and more productive if the chipset was in the same die as the cpu?

Both Intel's (Bay Trail or whatever the current "Celeron"s, in fact next-gen Atoms are called etc) and AMD's (A4-5000/5200, all desktop AM1 socket Sempron/Athlon CPUs etc) current lower power CPUs have integrated chipsets if thats what you referring to. Those contain SATA, USB controllers and memory controllers if im not mistaken.

Quote:

Originally Posted by Ulysses_ (Post 5184739)

The point about cores is a non sequitur. I believe 15% is a very good figure for cpu dies. Cpu dies can have more than one core. Yield is inversely proportional to die size, but not in the ratios you suggest.

As for the Northbridge, I believe they will have to integrate it as speed increases. That doesn't change my opinion about it being a good idea. It will be a difficult and expensive engineering challenge but they will be forced into it. In the end, it will cost us.

Hi,

Broadwell wiki;

Quote:

Broadwell is Intel's codename for the 14 nanometer die shrink of its Haswell microarchitecture due in late 2014, following Intel's tick-tock principle as the next step in semiconductor fabrication.[1][2][3]
Broadwell will adopt the Multi-Chip Package (MCP) design.[citation needed]
Broadwell will be used in conjunction with Intel 9 Series chipsets.[4]

Be sure to look at: Intel Tick-Tock;

Quote:

"Tick-Tock" is a model adopted by chip manufacturer Intel Corporation since 2007 to follow every microarchitectural change with a die shrink of the process technology. Every "tick" is a shrinking of process technology of the previous microarchitecture (and sometimes introducing new instructions as with Broadwell) and every "tock" is a new microarchitecture.[1] Every year to 18 months, there is expected to be one tick or tock.

A system on a chip or system on chip (SoC or SOC);

Quote:

is an integrated circuit (IC) that integrates all components of a computer or other electronic system into a single chip. It may contain digital, analog, mixed-signal, and often radio-frequency functions—all on a single chip substrate. SoCs are very common in the mobile electronics market because of their low power consumption. A typical application is in the area of embedded systems. The contrast with a microcontroller is one of degree. Microcontrollers typically have under 100 kB of RAM (often just a few kilobytes) and often really are single-chip-systems, whereas the term SoC is typically used for more powerful processors, capable of running software such as the desktop versions of Windows and Linux, which need external memory chips (flash, RAM) to be useful, and which are used with various external peripherals. In short, for larger systems, the term system on a chip is a hyperbole, indicating technical direction more than reality: increasing chip integration to reduce manufacturing costs and to enable smaller systems. Many interesting systems are too complex to fit on just one chip built with a process optimized for just one of the system's tasks.
When it is not feasible to construct a SoC for a particular application, an alternative is a system in package (SiP) comprising a number of chips in a single package. In large volumes, SoC is believed to be more cost-effective than SiP since it increases the yield of the fabrication and because its packaging is simpler.[1]
Another option, as seen for example in higher end cell phones and on the BeagleBoard, is package on package stacking during board assembly. The SoC chip includes processors and numerous digital peripherals, and comes in a ball grid package with lower and upper connections. The lower balls connect to the board and various peripherals, with the upper balls in a ring holding the memory buses used to access NAND flash and DDR2 RAM. Memory packages could come from multiple vendors.

Pathways will be an issue as the die size decreases for the substrate. Heat & power will be another issue for multi-core SoC.

Both Intel and AMD are working on SoC for the industry. SoC based Tablets are going to be the next Laptop as we integrate more on SoC devices. The future seems bright! :)

I am impressed by the new surface Tablet from a hardware perspective, not the OS.
:hattip:

Quote:

Originally Posted by business_kid (Post 5184557)

First of all, at the silicon wafer stage, if you get 15% good cpus out of a wafer that's a very good return.

Where did yuo get that number? I really doubt it.

While I've never had an Intel or AMD engineer confirm it, I've heard from a few souces that if they get less than 50% good CPU rate even in the early stages of a process its considered a failure.

By the end they expect a 90%+ rate of good CPUs.

Quote:

Originally Posted by cascade9 (Post 5185409)

I was wondering the same thing...if 85% of their work is getting wasted, surely they'd be almost completely focused on bringing the success rate up, not making newer and faster and smaller CPUs. Even if they only got the success rate to 30%, they would double the number of useable (and sellable) CPUs.

...keep in mind, I know nothing about the manufacturing of CPUs. Just my 2 cents.

For what it's worth, I think the money you pay for a cpu is not for the actual physical chip that they can duplicate for next to nothing, but for the R&D that went into it. So they don't care too much about die failure rate, whatever it is.

Gaming PC's have been such a large market for many years, why didn't they put the chipset inside the cpu much earlier for such systems, but this has remained a specialist design for so long?

The answer should be pretty simple: If there would be any commercial advantage for the chip manufacturers in doing so they would have done that already. So I think we can conclude that on high-end PCs (and high mid-range) having a separate chipset is commercially better for the chip manufacturers.
And I want to say, it is better also for the users: Just because I want to have a CPU upgrade doesn't mean at all that I also want to upgrade the chipset, so why should I pay again for something I already have?

Regarding the yield rate of chips, there were some example in the past, I remember them mainly for videochips, where a bad yield rate mad chips much more expensive. Producing high-end CPUs is not something that can be done cheap, so the yield rate indeed matters.

So it is an esoteric issue between manufacturers that cannot be guessed at by anyone outside manufacturing?

The actual percentage of chips is a trade secret in most companies. At one time it was rather low in the US compared to Japan. Some companies have improved it greatly but for this cpu example one has to realize yield may not be stated correctly. Some packages are sold that may in fact be defects. A 3 core cup may have one core that failed or a slower speed cpu may have failed speed so they down clocked it.