General Chat Thread

[VegetableStu]

link

hahahaha oh joy



EDIT: so apparently it went from that to this without bloodshed ._.

 

[VegetableStu]

third post in a row ,_,

so I had a look at the next image, and my first thought was "why don't motherboards...?"

 

[AleksandarK]

While showing the A11 CPU, i can say that i was surprised with its performance. It is just so damn powerful.

Only if it could be scaled it perfectly to desktop...

 

[Phuncz]

If you don't need all the parts to be industry-standard, you can take a look at the Mac Pro 2013-Present to see what can be achieved with focus on efficiency, on the x86 platform.

 

[EdZ]

While showing the A11 CPU, i can say that i was surprised with its performance. It is just so damn powerful.

Only if it could be scaled it perfectly to desktop...

ARM cores really don't scale up well. Clock them up and efficiency ends up well below modern x86 (and x86 isn't too far off when scaled down) while not gaining close to the performance of a modern CPU in non-mobile benchmarks (all the "A11 faster than Desktop CPU X!" headlines stemmed from a single test in a single mobile benchmark, Gekbench). You can try and scale out by putting a whole pile of them on a die, but various companies have been trying that for many years and nobody actually want to buy them, mainly because there's not really any use case that isn't suited better by either a GPU (if you can scale out to a big pile of slow ARM cores, you can scale out even better to a vastly bigger pile of CUDA cores) or a Xeon-D (a whole bunch of cores, but the individual cores are faster and can run your existing x86 stack).

 

[Soul_Est]

ARM cores really don't scale up well. Clock them up and efficiency ends up well below modern x86 (and x86 isn't too far off when scaled down) while not gaining close to the performance of a modern CPU in non-mobile benchmarks (all the "A11 faster than Desktop CPU X!" headlines stemmed from a single test in a single mobile benchmark, Gekbench). You can try and scale out by putting a whole pile of them on a die, but various companies have been trying that for many years and nobody actually want to buy them, mainly because there's not really any use case that isn't suited better by either a GPU (if you can scale out to a big pile of slow ARM cores, you can scale out even better to a vastly bigger pile of CUDA cores) or a Xeon-D (a whole bunch of cores, but the individual cores are faster and can run your existing x86 stack).

Precisely this. The thing that makes x86 what it is are the longer and wider pipelines, multiple caches, and dedicated circuitry for various desktop and server workloads. Scaling up ARM can be done but you'll add complexity, size, heat which go against the markets that ARM and their licencors go for and can support. Perhaps if certain companies were interested in collaborating to create these large ARM chips, it could be done. Maybe.

 

[jØrd]

I'm not saying I agree or disagree w/ anything being said here but there are some substantial efforts being made in the ARM server space.

Qualcom 48 core ARM CPU
Cavium Thunder X2
System 76 Starling Pro ARM

Not ARM and maybe the last of its kind but the Sparc M8 is just obscene. Also some other tidbits, Intel are pushing out new FPGA's, Nvidia are pushing new workstations w/ quad Tesla V1000's in, Intel also has new neural network processors. IMO outside of x86 things are actually getting really interesting.

My $0.02

 

[Soul_Est]

I'm not saying I agree or disagree w/ anything being said here but there are some substantial efforts being made in the ARM server space.

Qualcom 48 core ARM CPU
Cavium Thunder X2
System 76 Starling Pro ARM

Not ARM and maybe the last of its kind but the Sparc M8 is just obscene. Also some other tidbits, Intel are pushing out new FPGA's, Nvidia are pushing new workstations w/ quad Tesla V1000's in, Intel also has new neural network processors. IMO outside of x86 things are actually getting really interesting.

My $0.02

Definitely. Always held a soft spot for DEC Alphas, SUN Microsystems UltraSPARC Niagara, and IBM POWER. I think that future CPUs will go 3D just like NAND did.

 

[VegetableStu]

Can't remember which thread was it, but I vaguely remember someone wanted a watercooled power supply?

Just not this one, IMO.

 

[Phuncz]

Power supplies are quite difficult to cool without a liquid medium between all the components. The only way I see this work effectively is if they seal the power supply, fill it with non-conductive fluid and attach a copper waterblock in there that can extract heat. That way it could also be completely silent. Now it's just an air-cooled PSU with a waterblock attached to some flat components.

 

[VegetableStu]

going down this rabbit hole a bit: I was about to say "why not heatpipe over the areas where the radiators are required and weave them through the PCB and solder/press onto the waterblock below?", then a bit more made me realise that other components need cooling of some form as well, especially kilowatt PSUs (thread related)

At this point it became an exercise of One Radiator-Canal To Cover/Cool Them All, so I stopped ._.

Also TIL Silentmaxx watercooled 500W PSU (still no pictures of inside)

 

[VegetableStu]

whoa interesting:

▲ that's the PC.

the PCIe ports are here:

https://www.anandtech.com/show/12014/colorful-c-j1900a-btc-plus-v20-mining

 

[CC Ricers]

What is up with some Ryzen owners reporting higher specced CPUs than the model they bought for? Here's a Cinebench result showing 8 cores for a Ryzen 3 1200 with Ryzen 7-like results. Did they produce these from the same die, but disable some features like SMT?

 

[AleksandarK]

What is up with some Ryzen owners reporting higher specced CPUs than the model they bought for? Here's a Cinebench result showing 8 cores for a Ryzen 3 1200 with Ryzen 7-like results. Did they produce these from the same die, but disable some features like SMT?

Ryzen R models are made out of two CCX modules with 4 cores each. So they just disable some of the cores, or the CCX module hasn't passed the test with all 4 cores working and they for example use that one CCX with 3 cores and and they find another one like that, so it can create a 6 core ryzen 5.

From what i have read and heard, yields are 80%+, which means more than 80% of the CCXs actually has all 4 cores working, so a few chips labeled as R5, R3, can "slip". That is completely normal. It takes time to sort them all, and the demand is high.

 

[VegetableStu]

is there a thread limiter for xenforo?
(also is there recaptcha for xenforo?)
((also also is xenforo able to generate pages taking into account for ignored users?))

 

[confusis]

@VegetableStu What you're seeing is the result of a brutal spambot attack. It should now be back to normal.

 

[cmyk78]

Firefox updated. Everything looks different. I am not OK. Send help.

EDIT: Quick google check, and I like everything except the new look.

 

[jØrd]

BitScope Raspberry Pi Cluster for 3000 Cores in 30U
https://www.servethehome.com/bitscope-raspberry-pi-cluster-3000-cores-30u/

mind. blown.

 

[jØrd]

Tesla hype!

 

[Phuncz]

10.000 Nm means a tectonic plate will shift if two of these do a drag race.
I wanted an EV sports car ever since the original Tesla Roadster, which still costs about € 60.000 (USED) here.