Question Why no on-board motherboard L3 or L4 cache with LGA775 Northbridge CPUs ?

[80251]

When super socket 7 was still around, the L2 cache was on the motherboard, controlled by the northbridge (which was also the memory controller). Some super socket 7 boards even had removable COAST modules so you could increase the L2 cache size. But when Intel released their LGA775 CPUs, there was no L3 cache on the motherboard anymore, even though the northbridge was still the memory controller.

Why was this?

Wouldn't it have made more sense to have a fast L3 cache attached to the northbridge MCU?
Surely SRAM even back then was faster than DDR3?

 

[Aeacus]

Some super socket 7 boards even had removable COAST modules so you could increase the L2 cache size. But when Intel released their LGA775 CPUs, there was no L3 cache on the motherboard anymore, even though the northbridge was still the memory controller.

Super Socket 7 is AMD socket. LGA775 is Intel socket. Two completely different companies, with different way of doing things.

Why was this?

On-die cache is much faster than cache on the MoBo. Just the trace alone is FAR shorter when cache is inside the CPU, compared to when it is on the MoBo.
This would be one of the reasons why cache was included with the CPU, rather than with the MoBo.

Another reason would be that having cache inside the CPU, one can fine tune the CPU to work optimally with the cache it has. Compared to when MoBo has some X amount of cache, which CPU then has to work with. It could lead to a situation where CPU needs more cache, but MoBo doesn't have enough (user hasn't increased the cache amount on MoBo), resulting either CPU not working at all, or at reduced performance.
And who you think the people would send their complaints in? Surely not the MoBo manufacturer, but instead to the CPU manufacturer since CPU doesn't perform as it was advertised as.

Wouldn't it have made more sense to have a fast L3 cache attached to the northbridge MCU?

This would have resulted in way too high latency. Far higher than was designed for CPUs normal operation.

 

[80251]

Super Socket 7 is AMD socket. LGA775 is Intel socket. Two completely different companies, with different way of doing things.

On-die cache is much faster than cache on the MoBo. Just the trace alone is FAR shorter when cache is inside the CPU, compared to when it is on the MoBo.
This would be one of the reasons why cache was included with the CPU, rather than with the MoBo.

Another reason would be that having cache inside the CPU, one can fine tune the CPU to work optimally with the cache it has. Compared to when MoBo has some X amount of cache, which CPU then has to work with. It could lead to a situation where CPU needs more cache, but MoBo doesn't have enough (user hasn't increased the cache amount on MoBo), resulting either CPU not working at all, or at reduced performance.
And who you think the people would send their complaints in? Surely not the MoBo manufacturer, but instead to the CPU manufacturer since CPU doesn't perform as it was advertised as.

This would have resulted in way too high latency. Far higher than was designed for CPUs normal operation.

Super Socket 7 wasn't just an AMD socket, the Cyrix 6x86 also used it as well as the original Intel Pentium. Surprisingly, the motherboard L2 cache worked w/all of them.

I can see your point that an on-die or on-chip cache would be faster, but back in the LGA775 days, didn't they have SRAM that was significantly faster than DDR3? When you think about, why not put a cache IN the northbridge, since it was the MCU?

 

[Aeacus]

When you think about, why not put a cache IN the northbridge, since it was the MCU?

There's little point discussing what happened a long time ago. I already gave you two reasons why not.

Historically, separation of functions between CPU, northbridge, and southbridge chips was necessary due to the difficulty of integrating all components onto a single chip die. However, as CPU speeds increased over time, a bottleneck emerged due to limitations caused by data transmission between the CPU and its support chipset. The trend for integrated northbridges began near the end of the 2000s – for example, the Nvidia GeForce 320M GPU in the 2010 MacBook Air was a northbridge/southbridge/GPU combo chip.

On older Intel based PCs, the northbridge was also named external memory controller hub or graphics and memory controller hub if equipped with integrated graphics. Increasingly these functions became integrated into the CPU chip itself, beginning with memory and graphics controllers. Since the 2010s, die shrink and improved transistor density have allowed for increasing chipset integration, and the functions performed by northbridges are now often incorporated into other components such as southbridges or CPUs themselves.

Source: https://en.wikipedia.org/wiki/Northbridge_(computing)