256-bit CPU WHEN? >256-bit fixed point values take the same

>>56834598
When we need to be able to address more than 2^128b (281474977000000YiB) RAM on a single machine.

>>56834598
>256-bit CPU WHEN?
Intel Sandy Bridge in 2011
AMD Bulldozer in 2011
https://en.wikipedia.org/wiki/Advanced_Vector_Extensions

>>56834894
By that logic we've been using 128-bit CPUs since the 90s

>>56834598
you would need so many pins on the cpu

>>56834907
Yes!
Muh DICK!!!

>>56834936
I'd live with it if it meant 256-bits

>>56834598
stupid kid xx-bit means many things, but it s used commercially to mean the width of the bus (main memory in specific), so what we have here in PC is 64 bit, that makes the cpu able to access 2^64 bits of physical memory. Not even the most powerful supercomputer need that much. So even today only 48 out of 64 bits are only practically used.

>>56834598

>8*32 = 256, so 256-bit programs are 8x faster than 32-bit

Found the liberal arts degree

72-bit when?

By xx-bitness I mean not the width of the bus but register width

>>56835381
this

the c64 was an 8-bit machine but had a 16-bit wide memory bus so It could address 64k of ram

>>56835025
How many for quantum computers then?

36 bits is all we need

>>56834853
>when need to be able to
what a meme

Did we NEED any of the revolutionary techs when they came out? No.
Only those who have the Courage™ can make it happen, and those who say "we don't need it" only stall progress

>>56835502
I like how you see things, anon but
There's no way something like this can happen. Not right now at least.
Maybe sometime in the future, but not now.

>>56835452
You can buy desktop PCs with 1 TiB of RAM today

i made one yesterday

>>56834853
>he fell for the 281474977000000YiB RAM meme

>>56834598
it wouldn't fit in phones anon, therefore nobody cares

File: um.png (10KB, 141x93px) Image search: [Google]

10KB, 141x93px

>>56835805
are you a dumbass?

>>56835873
1 Byte is not necessarily 8 bits.

>>56835886
are you from the 70s

>>56834853
this.

bits on cpu are only required for memory addressing reasons. otherwise there are already instruction sets to efficiently do operations on 128-bit 256-bit data. see FMA, AVX, SSE4.1 etc.

File: twosmart.png (29KB, 250x226px) Image search: [Google]

29KB, 250x226px

>>56834598
>256-bit programs are nearly 8 times faster than 32-bit

found the brainiac

>>56835886
That can be assumed as always the case even though not technically true

It would make 4 x 64 SIMD operations possible, which opens up a few very interesting avenues for linear algebra number crunching.
However, the amount of engineering required for a slight gain, nah.

>>56834598
Apple would be the first one to craft it.

>>56835886
Yes, a byte is always 8 bits in modern computing.

File: 1473802432457.gif (4MB, 261x261px) Image search: [Google]

4MB, 261x261px

>>56835373
>>56835452

File: grooves.gif (25KB, 320x132px) Image search: [Google]

25KB, 320x132px

>>56834598

When we have true 3d structures on silicon or whatever substrate. At the moment we cant fit a bus that wide in the dimensions we can do with microwave vapor deposition.

Trust me EE here.

Literally will never need it. The universe will literally end before we can count something tangible with 64 bits.

>inb4 people said the same thing for 32 bit
No. Year 2038 was always a problem.

>>56838036
what

>>56838102
https://en.wikipedia.org/wiki/Year_2038_problem

>>56838118
yeah, no, not that. the rest.

>>56838036
>The universe will literally end before we get (0.1 + 0.2 == 0.3) == True

>>56838134

Work it out yourself you dunce. Do you need mommy to hold your dick while you pee?

You would just have a bigger hole to slowly push data through. It doesn't make sense when silicon is not fast enough and fabrication has already made silicon more costly in resources. Essentially you'd make giant die's without reason.

>>56838153
what

>>56834598
>>256-bit programs are nearly 8 times faster than 32-bit
"no"

>>56838153
My mommy always holds my pee pee when I go pee pee.
I don't see the problem there.

>>56834894
Don't these work on 4x64bit or 8x32bit values? hence why they are SIMD.
I don't think you can actually work on 256bit values with this.

>>56836000
>What is SSE

>>56835778
Lmao

>>56835748
"""desktop"""

1 physical core = 3 logical cores when?

File: Power-architecture-logo.png (36KB, 325x411px) Image search: [Google]

36KB, 325x411px

>>56839000
As in SMT? IBM's already topped that.
>12 cores
>96 threads

>>56839017
"The POWER8® processors are capable of SMT-8 which means up to 128 cores can be used in SMT-8 mode which yields 1024 logical processor"

http://www.ibm.com/support/knowledgecenter/ssw_aix_61/com.ibm.aix.cmds5/smtctl.htm

wew lad

Guess I'm fine with 4 core = 32 threads at this moment. they must bring this down to mainstream desktop level. Not sure about the benefits that we can reap.

>>56838159
Non-silicon CPUs when?

>>56835381
afaiu, this exists already. some intel cpus are quad channel, i.e., 64b*4 wide bus

>>56839376
err... nevermind, I completely misread this

>>56835421
Quantum computation probably isn't going to function with logical gateways, since the processor can't describe it's computation or be observed in any meaningful way. So, there probably won't be a "bus" to speak of. There's going to have to be something else.

>>56839017
>>56839163
SPARC is also 8 threads/core now for a total of 256 threads per 4.1 GHz/32-core chip

>tfw I won't get to own one until they're saturating eBay like US T1/T2s are now

>>56840212
Why does SPARC have so many threads?

>>56834853
Why are dumb people allowed to make posts?

>>56840297
For heavily threaded workloads?

>>56840297
Because it's meant for applications that require high levels of multithreading. Duh.

>>56840336
What are those?

>>56834894
>>56834907
reminds me of how Atari's justification for falsely advertising the Jaguar was "it has two 32 bit coprocessors so it's 64 bit"

>>56834598
What do you mean by 256-bit?

>256 bit address space
We're not even using the full 64 bit address space right now

>256 bit/cycle memory accesses
We already have that and more

>256 bit registers
We already have that with SIMD instructions

>>56835937
No matter how efficient those IS may be, they can never match the speed of a native hardware implementation.