Why the small core and huge cache for Zen?

AMD has a stronger backend while Intel has a stronger frontend, Zen2 aims to fix the frontend while Icelake improves Skylake's backend.
AMD has a slower memory controller for client but a better one for server.
AMD also has a more scalable architecture but Intel has low latency UPI, at least until Zen2 comes out.
AMD has a more efficient core at normal clocks, way less efficient at housefire clocks, this will change with 7nm

That's the simple version.

Thanks, bit off topic but why are Intel products priced so high while Ryzen so low?

Because you touch yourself at night.

Because Intel needs to maintain 60% gross margins or their shareholders will tank the stock.

Because good goyim with stockholm syndrome will buy Intel regardless.

On-die L3 cache is integrated into the die and fabricated on the same process as the cores. You can't fabricate a single die with different process nodes.

CPUs spend a lot of their die space on cache because RAM has shit latency. DDR latency is about 10ns. A 4GHz Core that has a cache miss has to idle for about 400 cycles. Probably 10% of a CPU is dedicated to actually doing math. The rest is about keeping that math machine fed. An addition or comparison instruction costs about 1 cycle. Having the CPU pause for a RAM read costs 400 additions. So you can see why avoiding cache misses is prioritized over number crunching when allocating die space.

Attached: 1543513868353.jpg (460x321, 18K)

tfw nobody makes good laptops anymore so you will never have a Ryzen laptop.

When increasing the core count, the performance gain on that workload isn't strictly linear. AMD has to help scale various workloads up to 32/64 cores. An unprecedented amount in single sockets. In order to help reduce the dropoff when increasing workloads towards 32 cores AMD has provided clusters with big caches. those caches make sure complex cross CCX lookups and in flight data can be handled in a way you dont have to involve more RAM trips.

I do know you can higly parallelise different tasks on a single system but that's not always what these high end CPU's are for. They are for increasing the speed at which you can solve particular workloads in a scalable manner.

Thanks for the explanation

Why so many cores when the software can't even handle that many?

There's software that can handle that. Even if your software isn't able to use more than a one core, you can still take advantage of the extra cores if the workload is divisible in to independent units. A web server handling 32 requests is going to benefit from that many cores even if the software to handle a request is single threaded. A multitasking OS is software created to let you do this. That's it's big purpose.

When will AMD double the cores again after Zen 2?

Zen4 on 5nm EUV node...maybe. we just don't know is right now man. 5nm EUV ist the most promising i guess

cache is king, the less you have to exercise your noc to access remote memory the better

Huh
So that’s why pentium 4 sucked so much. The double pumped ALUs were useless and the pipeline was massive. Makes sense.

>finally replace athlon ii x4 640 with fx 6300
>everything runs the goddamn same
I'll definitely get around to a ryzen system as soon as i can.

>315mm squared
When will dielets ever learn?

Attached: amd_fd9590fhhkwof_octa_core_fx_9590_4_7ghz_desktop_1095662.jpg (2500x2500, 275K)

wut

Attached: Screenshot_Brave_20181216-111326.png (1080x1920, 284K)

> all processor elements have the same size
No, node size only refers to the smallest element on the die, it doesn't tell you that much and is pretty much just for marketing more

>thinkpad facebook edition
no thanks

the fuck are you talking about?

This, my sister uses a facebook tier Macbook air u class 1.8 chip and she runs VMs and does compsci work with it.

A 2700U is more than enough.

look at the Apple A12 to see the same strategy in action. big shared L3 that's 2x the size of the fast CPU cores (even including their massive L2 and L1 caches)