LegendKillerRG

It’s always the first number. But you do have to keep in mind, it’s the difference in what’s used that can give better performance. Usually Samsung b-die chipset will have something like 16-16-16-39 timing(which is better) and also the cheaper Hynix will be like 16-18-18-39. The main difference would be that b-die will have more headroom to be adjusted like overclocking or lower timing. Better memory that are binned good are more expensive which usually carries the b-die, and usually people will grab a cheaper b-die sets and overclock it to match that of a set that’s twice as expensive.
 
planning to build a SSF using 5900x so doing im due diligence on memory. I’m setting on a set of $240 trident z neo 3600Mhz 16-16-16-36 and overclock it to 4000MHZ. 

uATX, meaning Micro-ATX.
But check out this recent video, might help you with some ideas:
 

USB 2x2

Gosh! I certainly hope this doesn't hurt the CEO's credit scores. That's really hard to clear up. (/sarcasm).

No RGB in 2017.

For all your high speed porn file transfer needs

*clicks*
 
 *closes*

I guess you can say this news is pretty.... explosive!!!!... Get it? TNT?... 

We need some #CustomLinusCoolers from his scrapyard wars.

We need some #CustomLinusCoolers from his scrapyard wars.

Does anyone know anyone here on the forums that's well-versed in the deep learning, AI field?  It wouldn't hurt to get a simplified explanation of what is actually going on and what it can do (and why it matters)  

GPU do best in FP32 but except for FP64/FP16, a 1080Ti does 100-200GFLOPS(11.3TFLOPS in FP32 which is the acceleration you're talking about). Where as this USB can do over 100GFLOPS on FP16 on 1w of power and only cost $80.

GPU do best in FP32 but except for FP64/FP16, a 1080Ti does 100-200GFLOPS(11.3TFLOPS in FP32 which is the acceleration you're talking about). Where as this USB can do over 100GFLOPS on FP16 on 1w of power and only cost $80.

The new Mustang is pretty awesome (GT350R)
 

That's actually hilarious.
I mean, it's way easier to make something at least partially accessible than it is to make it a brick. That's some impressive dickery.

This is planned obsolescence in its raw form  Good job failing your customers Microsoft!

https://www.thurrott.com/mobile/microsoft-surface/118538/ifixit-gives-surface-laptop-lowest-possible-score
 
Teardown video
 
Louis Rossmanns views on the matter
 
So Microsoft one upped Apple, copying what is bad thinking it would sell better?

Meh.. ThreadRipper.

1) Then you should have written HBM or HBM1 not HBM/2
If you meant to say HBM + HBM2 then you write HBM(2) with the brackets implying that the 2 is conditional.
 
2) $1200 USD is not "cheap" neither is the $400-1500 of the Fury cards.
 
3) vega doesn't even exist in the market yet, everything is just rumours and suppositions.
 
4) having HBM does not magically make a GPU perform 100x better, the GPU core is what matters most by a long shot, HBM is just like a cherry on top that can get you a few more fps in certain games.

You guys realize that it's not like that?
You just can't replace one memory by the other and expect it to work the same. HBM type of memory have quirks which change the design of the core in order for it to take advantage of it. Slap a gddr5X memory on it and it doesn't work as well anymore because it's not meant to.

Meh.. ThreadRipper.

Meh.. ThreadRipper.

They're gonna pull a 34c/68t CPU out of their asses, duh.