Intel 12900K benchmarks leaked

i would have liked to gone threadripper for encoding but not for the price
 
Uhh so it's not faster than AMD except single core then? What shitty journalism again, probably get the good Intel shekels for marketing as usual.
 
JSHamlet234 said:
I would be interested in one with just the real cores and none of the potato cores.
Click to expand...
I would be interested if they split the smaller cores off into their own package and made them optional. Have like a secondary smaller socket for a slave CPU. Motherboards these days sure have plenty of space for that, for the past few years, motherboard manufacturers have been trying to hide the fact that there really isn't much on them anymore with big plastic shrouds, LEDs and logo plates. With how many things are integrated on the CPU these days, motherboards are more like a backplane than a traditional motherboard of the past crammed full of devices.

Back in the early 90s and prior, AMP was commonplace. You had a strong main CPU doing the heavy lifting and one or more slave CPUs doing more mundane tasks to free up the main CPU to get heavy lifting done faster. Arcade boards and game consoles were perfect examples, where something like a 68000 was paired with a Z80. The 68000 would run the main game logic and the Z80 would run the sound chips. There were even some x86 servers that had both 486 and 386 CPUs running at the same time, the 486 would do the heavy lifting and the 386 would run the more mundane tasks like bus arbitration, I/O, Networking, etc.

It would still be beneficial today to have a modestly powerful, but power efficient slave CPU to do things like network I/O. Even with the fast CPUs today, network I/O eats up a decent chunk of processor time, unless you have a fancy multi hundred/thousand dollar network card that can offload the processing to a dedicated DSP. Many of the NICs on motherboards today are just dumb software PHYs that do almost everything in the driver running on the systems CPU. The higher the throughput, the more CPU time it eats up.
 
I'll give ADL a spin. Haven't purchased INTEL (other than laptop) in about 3 years now.
 
I looked up my invoice so I can use the same CPU Noctua fan I picked up in Dec 2020. Which will have a different mounting kit. Or I could just use my 10850k for the next 4 years with the same results. It as ll depends if I can find one at launch and it's not overpriced.
 
VRay Bench.png


5900X / Gigabtye RX 5700 XT / X570 ASRock Velocita / 32gb G.Skill Ripjaws -3600 CL14 (No overclocking)
Score seems low to me, but I had about six programs loaded in background + bunch of TSRs. Also my old AIO is on near life-support and is about to be replaced this weekend
 
GiGaBiTe said:
I would be interested if they split the smaller cores off into their own package and made them optional. Have like a secondary smaller socket for a slave CPU. Motherboards these days sure have plenty of space for that, for the past few years, motherboard manufacturers have been trying to hide the fact that there really isn't much on them anymore with big plastic shrouds, LEDs and logo plates. With how many things are integrated on the CPU these days, motherboards are more like a backplane than a traditional motherboard of the past crammed full of devices.

Back in the early 90s and prior, AMP was commonplace. You had a strong main CPU doing the heavy lifting and one or more slave CPUs doing more mundane tasks to free up the main CPU to get heavy lifting done faster. Arcade boards and game consoles were perfect examples, where something like a 68000 was paired with a Z80. The 68000 would run the main game logic and the Z80 would run the sound chips. There were even some x86 servers that had both 486 and 386 CPUs running at the same time, the 486 would do the heavy lifting and the 386 would run the more mundane tasks like bus arbitration, I/O, Networking, etc.

It would still be beneficial today to have a modestly powerful, but power efficient slave CPU to do things like network I/O. Even with the fast CPUs today, network I/O eats up a decent chunk of processor time, unless you have a fancy multi hundred/thousand dollar network card that can offload the processing to a dedicated DSP. Many of the NICs on motherboards today are just dumb software PHYs that do almost everything in the driver running on the systems CPU. The higher the throughput, the more CPU time it eats up.
Click to expand...

The little cores are actually a great idea. We might look down on them because we think low power... wth do I need a low power core for. But yes in performance situations the low power cores can act in the same way as accelerators do in SOC for things like HVEC encode ect... only in a much more general purpose way.

What Intel will show with these chips is that as far as single core performance goes... Yes you can slave a smaller core to speed single thread performance. I have argued with a few people now that that is even possible.... yes it is. ARM has been capable of that for awhile, some ARM super computer chips use such tactics. Apple took the idea and made it mainstream. Yes the M1 achieves its single threaded performance numbers by using 1.5 cores not one.

The issue for Intel.... is as cool as that is. Multi threaded is more what basically everyone outside a handful of gamers chasing FPS in specific old titles that are still heavy Single thread dependent care about. In terms of Multi thread there is no way Intel is going to touch AMD with this gen. Even if they get is oh so close, or can show they are slightly faster in X or Y benchmark. AMD has already shown their hand. Lisa can drop 3D cache refresh Zen 3+ at any time... if that happens even if AMD just ships enough to satisfy a small chunk of the high end market. Intels new chips will get dominated in multi thread and probably singled threaded as well. As much as Single thread love high clocks.... single thread also loves boatloads of cache.
 
You must log in or register to reply here.
Back
Top