Gigabyte Leaks Ryzen 9 Specs

>24 threads @ 4.6GHz
>95W TDP

Intel struggles to make their 16t at 5GHz not exceed 140W, bitch please

What does the cache even do

why can't we have 16GB cache on CPU? hmm?

>300 seconds latency

>What is "die space"

Attached: aYJ0g2NG_700w_0.jpg (700x616, 45K)

because a 1gb die of memory is already the size of a cpu

Imagine if you basically had DDR5 RAM but inside your CPU so it was super duper fast to talk to. It's very expensive to put something inside the CPU, so you can only have small amounts.

what is prices

Since the 70's, there are two basic types of memory, dram and sram.
Sram is made out using digital logic, basically 8 transistors that can hold a bit logically.
Dram on the other hand use tiny capacitors (think on a battery that charges and discharge fucking quickly).
Sram is a LOT faster than DRAM, it's not even funny, because it is made exactly of the same shit CPUs are, and this is why they use it as a cache, it's a super duper hyper fast memory glued directly to the CPU.
Dram on the other hand, dram is sluggish, dram needs to keep refilling the capacitors, dram is slowish.
But a capacitor and shit to control it take the same space of ONE transistor, which means that Dram can literally have 8 times more data per size than Sram.
In the space you shove 8GB of DRAM, you can only shove 1GB of SRAM.

Very informative, thanks

Threadripper right now is decent for games. It's worse than mainstream CPUs, but it's not like it's so bad that if you have a Threadripper PC games will run like complete garbage and be unplayable. Threadripper right now is a bad buy if your main performance-hungry activity is gaming and it will remain so in the future as well. This won't change.

Games don't really benefit from large numbers of cores. Games on these new CPUs are likely to run best when all their threads are running inside the same CCX/die. This should provide 8C/16T for the game, assuming you have a CPU with completely functional dies, not cut-down dies. This should be more than enough for any game, so performance will be dictated by how fast that 1 die runs. TR won't really run any faster than the mainstream CPUs, so at most you'll get the same sort of performance as a mainstream CPU but at much higher price, which would obviously make it a bad deal.

If this architecture proves to be legit for mainstream, I really wonder what core-to-core latency will be like if the cores are on different dies. Do individual dies have to go through the I/O die in order to reach another CPU die or do they also have some direct link? It looks like there won't be a direct link. Certain loads like some video games or video game emulators really don't like high latency between cores, so performance could suffer if the OS schedules threads on different dies instead of the same die. I'm not sure whether Windows is smart enough to schedule threads from the same process on the same die.