"TSMC's 5nm node, featuring dual 128b Vector Units, 2MB of L2 cache per core, 3 GHz clock speed, eight-channel DDR5 memory controller, 128 PCIe Gen 5 lanes, and a 200-350W TDP" should be perfect for RandomX

2MB L2 per core, nice

they finally clued in

192 cores 350W TDP, is that a bit high TDP?

5nm process, this chip should beat most other chips for efficiency

350W is okay for 192 cores

less than 2W per core

IIRC they have 2 MB L2 per core, but shared L3 is small

would be fun to run some benchmarks on it anyway\

3 GHz sounds pretty good, what were the previous iterations like ? near 2GHz or close to 3 ?

ARMv8.6 instruction set, it most likely has AES too

8-channel memory means a single CPU can go up to 100 kh/s

not bad for 350W

8 channel DDR5, all the latest goodies

that's north of 250 h/J efficiency

it will be worse at the wall, but definitely better than what X5 can do

and probably it won't seppuku itself after 5 days

if we are comparing it to X5

and they have dual socket systems, those can probably match X5 hashrate, but with lower power :D

as soon as Linux gets patched for 384 cores

yeah with dual socket system single mobo will be more power efficient probably, maybe marginally but still

Waiting for proper Phoronix tests of this beast

I think they have XMRig in their test suite

but they're shit at tuning it :D

hasn't Amazon by buying up all of Ampere's chips? or I'm misremembering

Google has them too

main point is, mere mortals aren't going to be able to buy these.

what can we get our hands on that matches that power efficiency?

"The slowing of Moore’s Law has demonstrated a need for alternative processing architectures, which has led NVIDIA to become a $1.1 trillion company, greater than Intel and AMD combined ($161 billion and $170 billion, respectively). Arm will quickly catch up, with a rumoured $52 billion valuation after IPO."

had no idea Nvidia had outgrown AMD and Intel

wait, AMD is bigger than Intel?

161 < 170

I guess Intel has really fallen behind

The highest efficiency CPUs avaiable to retail buyers are probably 7950X, and Ze4 EPYCs

but they're all in 150-200 h/J range

7950X is <= 150 h/J after tuning

i wonder how the FPGA acquisitions affected those values, intel getting altera, amd getting xilinx

$16.7 billion deal underscores Intel CEO's plan to expand chip maker's business...

AMD said it has completed its $49 billion acquisition of Xilinx, marking the largest chip deal in history....

hmmmm altera so cheap

sech1: where were the zen2 cpu's efficiency wise for comparison?

I remember they were above 100 h/J, up to 120 h/J for well-tuned rigs

so still amongst the most efficient o_O

oh wait, 150 h/J ... not j/H never mind :D