Nvidia shows ARM CPUs offer the same power as x86 CPUs in servers
Click here to post a comment for Nvidia shows ARM CPUs offer the same power as x86 CPUs in servers on our message forum
Kaarme
rl66
cryohellinc
It will be a bad day for everyone if Nvidia will be allowed to make this acquisition. For the good of the industry, I hope we will have an independent ARM, otherwise, ARM will quickly turn into one of Nvidia's tentacles...
JamesSneed
There is a complete lack of data backing this up. Sure if you toss 2-3x the ARM CPU's at a certain workloads you can get higher performance than x86 which I assume is what Nvidia did.
vestibule
Lol. I bet Intel are now blank cheque book lobbying the UK government not to let Nvidia get there hands on the prize.
Astyanax
rl66
AMD and Intel have already a step in ARM side,
they don't have wait that NVidia have an interest in it.
rl66
tunejunky
tunejunky
UK gov is in a very weird and inconsistent position.
after okaying the sale of ARM to Softbank (Japan), they are dithering over the sale to Nvidia (USA).
talk about shutting the barn door after the cows got out.
this is all political, and it's stupid politics at that.
this is all a reflection on right-wing nationalist politics post-Brexit, and has nothing to do with facts.
the fact is quite simple, Softbank could've at any time grant an exclusive license. they didn't because that wasn't their business model.
if Nvidia had fabrication i would worry about exclusivity.
but they don't so i won't.
there are basically two and a half ARM fabs - 1) TSMC, 2) Samsung, 3) Chinese (the .5 fab)
Intel does have an existing license, not that they are using it. AMD also has a license.
Nvidia needs these fabs who also have fat contracts with other ARM licensees like Qualcomm and Apple
Dribble
JamesSneed
schmidtbag
This isn't surprising but obviously cherry-picked results. Even MIPS or RISC-V can outperform x86 depending on how which instructions you use (or rather, omit).
The thing is, most GPGPU tasks don't require much from the CPU at all. You basically just want a lot of low-latency cores that can handle all the compiling quickly and then oodles of PCIe lanes. That's relatively cheap and easy for Nvidia to make. The thing that surprises me is: what took them soooo long?
rl66
rl66
zais101
theres now 1 year phone with ddr5, 240hz touch (even on 60hz) and 240hz+ oled screen, virtual mem swap storage, lets hope coders, drivers, codecs & os can be more unified 🙂
tunejunky
"IT might not have the knowledge (they should but they don't) or don't bother to do a radical change (and the financial departement is happy because this way they don't spend money) it's for that you still have old OS... And the security fail too... if not "user name: admin password: admin " 's lazy IT"
oh RL you made my ribs hurt from the laugh of pain
cryohellinc
fantaskarsef