Language:

en_US

switch to room list switch to menu My folders
Go to page: First ... 14 15 16 17 [18] 19 20 21 22 ... Last
[#] Sat Aug 15 2020 22:42:24 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]


they might be binned a little bit differently, but they're the same part with a different TDP and PPT target (which happens to match the ECO mode on the 3900x)

actually they probably are binned a little differently because the review of the 3900x running in Eco mode shows it boosting a little higher than the 3900.


The base clock on the OEM, 65W part is only 3.1ghz and that's probably very conservative, I was able to hit a lot higher than that even in an almost fully all-core workload (kernel compile with -j24)

[#] Sun Aug 16 2020 09:27:01 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]


on closer examination the kernel compile appears i/o bound on this system now. explains the lack of difference between CPU settings.

I am running an older, SATA SSD.

you would think that you could get better IO throughput by using more concurrent jobs, but no. there's an optimum value between 18-24 jobs, which succeed in getting to 75% cpu utilization, and at anything more than that, CPU drops back down to around 50%...

it's all disk writes for compiler output, reads fit in cache.


journal_async_commit, maybe?

[#] Sun Aug 16 2020 22:15:22 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]


wow, nope. it's not remotely I/O bound. It's IPC bound.


Ran a whole kernel compile entirely out of a RAM disk (/dev/shm), and zero change in performance. The reason it fails to utilize more than about 80% CPU must be just bottlenecking on the master `make` process whose job is to spawn parallel children.

[#] Sun Aug 16 2020 22:44:06 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]


so results. linux trunk snapshot from yesterday, identical clean builds on ram disk. AMD 3900X maxes around 80% utilization with SMT enabled. make -j24.

10m34.743s 65W mode
10m18.668s 105W mode

difference - 2.5% for significantly greater power consumption

[#] Mon Aug 17 2020 20:50:28 UTC from IGnatius T Foobar

[Reply] [ReplyQuoted] [Headers] [Print]

Wowzers! It's totally not worth it to run at the higher speed. I'll bet the 2.5% difference becomes even smaller if the machine is doing a lot of different things rather than a single parallelizable job.

[#] Mon Aug 17 2020 21:08:51 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]


Yeah. I also noticed that having "C-States" turned on helps performance here; regardless of whether you're set for 65W or 105W mode, the CPU runs ~20% idle because of dependencies in the workload, yet it is throttled by the Package Power Target during all the more-parallelizable portions of the run. So it's a bit faster with C-States on, so that the idle cores do not contribute to the package power target when they're idle.

[#] Mon Aug 17 2020 21:11:30 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]


If one were to just raise the PPT in overclocking config and install liquid cooling, this chip might be a bit of a monster; but the payoff to 105W suggests that the payoff to 210W might be just another 2.5% ;

[#] Sat Aug 22 2020 00:41:28 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]


hmm... idle power consumption on this chip (according to PPT, via HWINFO) is around 15W

That's HIGH. On Haswell, it was measured in milliwatts.

Might have something to do with the multi-chip module design. Perhaps there's a reason you're not seeing this stuff in laptops yet.

[#] Sat Aug 22 2020 00:56:11 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]


Gamer desktops are one thing, but that's almost enough to call into question the datacenter suitability of this stuff, although it's still surely the best choice for heavy-duty workloads, a lot of stuff is bursty and runs closer to idle a lot.


And forget laptops. Wow.


This isn't being talked about much in the press, but some of the reviews I'm looking at seem to bear this out; they measure consumption at the wall and it might be 10-20W higher than some of the Intel stuff at idle.

[#] Mon Sep 14 2020 15:22:29 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]


https://www.anandtech.com/show/16080/nvidia-to-acquire-arm-for-40-billion

It.s Official: NVIDIA To Acquire Arm For $40 Billion

Can't say I'm happy about this; ARM looked a lot better when it was not owned by a dominant chipmaker, the IP has always been available to all on non-discriminatory terms.

[#] Mon Sep 14 2020 15:50:42 UTC from IGnatius T Foobar

[Reply] [ReplyQuoted] [Headers] [Print]

That was part of ARM's value proposition. Apple's move to ARM looks a bit less cool today. I wonder if this will breathe new life into RISC-V.

[#] Mon Sep 14 2020 15:57:06 UTC from Ragnar Danneskjold

[Reply] [ReplyQuoted] [Headers] [Print]

2020-09-14 11:50 from IGnatius T Foobar
That was part of ARM's value proposition. Apple's move to ARM looks a

bit less cool today. I wonder if this will breathe new life into
RISC-V.



I wonder if they'll slow down their transition. Maybe just make some ARM-based laptops..... See where things go.

[#] Wed Sep 16 2020 13:31:13 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]

2020-09-14 11:50 from IGnatius T Foobar
That was part of ARM's value proposition. Apple's move to ARM looks a

bit less cool today. I wonder if this will breathe new life into
RISC-V.

Does RISC-V even matter? IP is more about the hardware implementation than about the ISA, right? Doesn't the AMD vs Intel fight give us a clear precedent that you can't really defend the IP on the ISA itself?

[#] Wed Sep 16 2020 13:34:11 UTC from LoanShark

[Reply] [ReplyQuoted] [Headers] [Print]


https://riscv.org/exchange/cores-socs/

*wonders if any of those can really compete with ARM on performance.

[#] Mon Sep 21 2020 17:49:11 UTC from IGnatius T Foobar

[Reply] [ReplyQuoted] [Headers] [Print]

Today, no. The reason I brought it up is because both ARM and RISC-V are fabless publishers of ISAs (and reference designs?) rather than manufacturers of finished parts. If manufacturers feel that ARM is no longer neutral enough for them, RISC-V could be a consideration.

Admittedly, this is not likely to happen, unless Nvidia gets really stupid with the licensing fees.

[#] Fri Sep 25 2020 22:24:19 UTC from IGnatius T Foobar

[Reply] [ReplyQuoted] [Headers] [Print]


WTF?

I replaced the keyswitch for the Enter key on my Das Keyboard 4, which has Cherry MX Blue keyswitches. Sometimes the keypress would not register, other times the keypress would register multiple times. This was a few months ago.

Now it's happening again?! The exact same problem. I'm having a hard time believing that another keyswitch failed in the exact same way, just a short time later. I want to blame it on software, but the problem happens on two different computers.

I like this keyboard and I don't want to replace it.

[#] Tue Sep 29 2020 19:13:59 UTC from IGnatius T Foobar

[Reply] [ReplyQuoted] [Headers] [Print]

After a bit more testing I am now thinking that perhaps the problem wasn't in the keyswitch itself. It might be a hairline crack in the circuit board, or some other non-switch failure. I will miss this keyboard but it's still usable, so I'm going to move it to a NUC I have set up in the garage.

For now, my main rig has a mushy old rubber dome keyboard attached to it.
I suppose I could haul out one of my old Model M keyboards and find a way to make it work, but they are absolutely filthy. If I buy another mechanical keyboard, I think it will be a tenkeyless layout this time, because I never use the number pad and I'd rather have the desk space back.

[#] Fri Oct 02 2020 21:59:38 UTC from Ragnar Danneskjold

[Reply] [ReplyQuoted] [Headers] [Print]

My son has a tenkeyless.... I can barely type on it for some reason.....

[#] Fri Oct 02 2020 23:25:03 UTC from IGnatius T Foobar

[Reply] [ReplyQuoted] [Headers] [Print]

The keyboard I ordered to become my new daily driver is the "Ducky One 2 Horizon TKL".  For those of you viewing this in a web browser, a photo is shown here:

I ordered it through the Great Satan of Seattle so I can return it if it isn't right for me ... using a credit I had with them from the last thing I ordered and didn't like.

If you're wondering about the color scheme, the blue keys remind me of old Data General equipment.  That has a bit of nostalgia for me, and nostalgia makes us do weird things.

I went with a tenkeyless because I almost never touch the numeric keypad on my keyboard, so hopefully I can save some desk space.



[#] Tue Oct 27 2020 19:13:23 UTC from IGnatius T Foobar

[Reply] [ReplyQuoted] [Headers] [Print]

My son has a tenkeyless.... I can barely type on it for some
reason.....

After a couple of weeks with one, I think I know why.

I had some trouble landing keystrokes on it for the first day or two. I took a good look at what I was doing on both full size and tenkeyless boards, and I discovered that I was using muscle memory to measure the distance from the right side of the keyboard to where my right hand needed to be to hit the correct keys.

It took a couple of days to adjust, but it wasn't difficult. I'm back to my original typing speed, and my right arm is actually more comfortable because my trackball now sits where the number keys used to be.

What I find amusing is that tenkeyless keyboards still have a Num Lock LED even though there is no Num Lock key. There is no number pad, no "Fn" key to repurpose other keys as number keys like on a laptop, no purpose at all for it to be there. But it is there, and presumably it will illuminate if the host tells it to.

Oh, and there's still a PrtSc key, which is nice because after way too long, most operating systems finally do the right thing when you press it.

Go to page: First ... 14 15 16 17 [18] 19 20 21 22 ... Last