building new PC, emulator CPU question

[NaturalViolence]

kinkinkijkin Wrote:8 cores is small for what it's useful for. Servers usually have around 12-256 cores, video decoding/encoding machines usually have 16 or more if you buy them pre-built, professional workplace computers rely on the GPU now anyways because it's not possible with the CPU on most modern operating systems to have enough logical cores for what they do

AMDs server/workstation offerings are actually even further behind Intels than their consumer desktop line. With their desktop cpus they are able to mask some of their efficiency problems by overclocking and jacking up the TDP. Often times you will see 125 watt AMD cpus competing against 83 watt Intel cpus. They can't do that with server cpus because the cpus have to comply with very strict TDP requirements. OEMs need to be able to jam lots of cpus into small rackmount cases with small coolers running 24/7 for years without any maintenance and without them overheating. And businesses that buy from the OEMs are very interested in power consumption and cooling requirements since it costs a lot of money for them long term. As such predefined TDP levels have been layed out by OEMs that both Intel and AMD must comply with to keep everyone happy. So instead you will have for example a 110 watt AMD cpu competing against a 110 watt Intel cpu. When AMD is forced to compete with Intel at the same power consumption they lose their clock rate advantage and their architectures poor efficiency becomes very apparent. The 8 core xeons absolutely crush the 16 core opterons across the board in server workloads no matter how well multithreaded they are. The xeons have both a higher clock rate and IPC which easily outweighs their core count disadvantage. Whereas on desktops AMD holds the clock rate advantage and only loses in IPC since they are allowed to overclock their cpus so high as long as they provide a strong enough stock cooler and chipset VRMs to deal with it. As you can see AMD has a design emphasizing core count that was supposed to give them an edge in environments were multithreaded performance is more important. Yet it actually ends up doing the opposite.

kinkinkijkin Wrote:professional workplace computers rely on the GPU now anyways because it's not possible with the CPU on most modern operating systems to have enough logical cores for what they do

Workstation systems rely heavily on both. Depending on the application importance will lean more towards one of the two sides but most applications will tax both heavily. The second part of your statement makes absolutely no sense to me (as usual).

NathanielRH Wrote:I honestly don't know wtf is with AMD.

"here! more cores! No one needs all this and our individual cores are weak but buy them! FILL YOUR HOUSE WITH CORES!"

what kinda backwards logic is this

They have 1/10th of the income of Intel. Their R&D staff size and equipment budget is far smaller than Intels. At some point someone over at AMD decided that they could no longer compete with Intel in microarchitecture power efficiency (and rightfully so). Designing faster, larger, and more power efficient cores is extremely difficult and requires a ton of manpower from very intelligent engineers to make even the smallest improvements. Intel will always be able to outcompete them in this area. So instead they decided to design smaller, simpler, and weaker cores and pack more of them on the chip. This enabled them to design a chip that would at least be able to compete with Intel in multithreaded performance on a fraction of the budget. If you can't compete with them head on aim where they're not looking instead. "If we can't beat them in singlethreaded performance we'll at least make sure to beat them in multithreaded performance" was their original plan but the design had a number of efficiency issues that caused its performance to fall short of their expectations and prevented it from exceeding the multithreaded performance of Intels designs. I don't know if it was the right move but I can certainly see the logic in their decision.

Now it seems like they're abandoning multithreaded cpu performance having now realized that they can't beat Intel in that area either and are shifting the goal towards beating Intel in GPU performance instead. I think they've realized that both singlethreaded and multithreaded performance are so heavily impacted by power efficiency in modern cpus that because they will never be able to outcompete Intel in power efficiency (again fewer resources than Intel at their disposal) it's a hopeless goal. Of course Intel is placing greater emphasis on GPU performance each generation and if they ever decide to make it a priority in the future they might be able to outcompete AMD in that area too. We'll see.