sliken

159 ms·

631.
by sliken 4y ago
Sadly AMD/Intel iGPUs (which AMD calls APUs) are limited to 83GB/sec or so, of which you'd be lucky to see 50GB/sec or so. Apple has 100, 200, 400, and 800GB/sec flavors that fit in laptops (100-400), mac mini (100-
632.
by sliken 4y ago
I was working on a backup system, the usual. Walk a directory tree, track new/changed files, then queue them for sha256 and encryption, then upload to a server with gRPC. I hit the GIL, switched to multiprocessing, which helped. Stil
633.
by sliken 4y ago
Depends on your use case. Sure you can cap a 7950X at 105 watts and fit a RTX 3070 in the case, but it's not going to be particularly small or quiet. The RTX3070 is physically large, does have a nice 448GB/sec or bandwidth, but o
634.
by sliken 4y ago
Agreed. Sadly Intel didn't make the NUC a standard. There are decent SFF around, but sadly they all use 128 bit memory interfaces which is severely limiting for an iGPU. Sadly Apple can build low power small systems with 2x, 4x, and
635.
by sliken 4y ago
Yes, sure, at somethings. The 7950x does really well at some cache friendly workloads. But it also burns way more power and generally lives in a large ATX case with at least 7 fans (2 intake, 1 exhaust, 1 power supply, 2 on the GPU, and o
636.
by sliken 4y ago
Nice. Any ML inference or training by chance?
637.
by sliken 4y ago
Interesting didn't know that, figured it worked like a home loan. Banks often have similar insurance requirements and/or PMI for home loans.
638.
by sliken 4y ago
Apple makes nice hardware. Nice case (milled aluminum), nice heatsinks/airflow, and killer memory systems. Not a fan of their OS though. Generally the competition from AMD, Intel, and Nvidia generally has much less memory bandwidth (
639.
by sliken 4y ago
Use the accelerated GPU at all? I've been watching the progress with async, batching, and related. But I've not seen any benchmark numbers or any idea of performance.
640.
by sliken 4y ago
You haven't bought a car till you pay for it. If you buy it with a loan the bank owns it, and can restrict your use of it as they see fit.
641.
by sliken 4y ago
Heh, well driving a car at 60 mph during on the wrong day on the wrong road would kill 100x more bugs than a entomologist does in their entire career.
642.
by sliken 4y ago
Right, so the 7800X3D (due in April), the old 5800X3D, or any future threadripper or Epycs with the X3D. Best L3 per core outside of the X3Ds it the 7600 or or 7900 CPUs that come with the full size L3 (32mb PER ccd) but only 6 cores per CC
643.
by sliken 4y ago
Or just buy the X3D that have cache on each CCD instead of half of them.
644.
by sliken 4y ago
Indeed, but the much nicer/faster/better PCIe and faster CPUs in the RK3588 for another $25, with the option to get double the ram of the top RPi 4.
645.
by sliken 4y ago
Dunno, got some numbers to back that up? So far I'm pretty impressed with my Rk3588 board, 8gb ram, and dual 2.5gbe. It's a bit more than half as fast as my quad core/8 thread xeon desktop, but that's abotu where I'
646.
by sliken 4y ago
Or any of the Rockhip 3588 boards. Better CPU, better PCIe, some have dual 2.5gbe, 4, 8 or 16gb ram, etc. I got the NanoPi R6s. Seems silly to save $25 on the slower CPU and half the ram.
647.
by sliken 4y ago
I was in a relationship. Had 2 single friends, took a dance class and started dancing. A month later they both had girl friends and stopped dancing.
648.
by sliken 4y ago
I found a bunch of ML benchmarks, only one had a comparison to a RTX3070. Seems quite a bit better than "not able to run anything". In particular you can use up to 96GB ram, 4x the 4090. Granted it's m2 max is approximatel
649.
by sliken 4y ago
Agreed on deceptive marketing. I dug around and found some M2 updates. Seems that generally, Nvidia laptops drop to a small fraction when on battery. Examples: 3dmark wild life. M2 max = 150 fps, 3080 ti 120 fps, 3080 Ti unplugged = 27fps
650.
by sliken 4y ago
You mean like PyTorch or TensorFlow?
651.
by sliken 4y ago
Like ML training and inference?
652.
by sliken 4y ago
People seem generally unaware of the difference between number of channels, width of channels, and number of dimms. In fact this question is wrong, or at least misleading. The OP went from one DDR5 dimm (with two 32 bit channels) to 2 DDR5
653.
by sliken 4y ago
That fits. The p4 had a long pipeline, high clock speed, and poor perf per MHz. AMD pushed the general performance score, forget the name, and seems like the succeeded. Doubly so since they shipped x86-64 first, which people did care abo
654.
by sliken 4y ago
The code literally checked for CPU = AMD, and ignore the CPU feature bits what show which accelerations are available. So sure Intel shouldn't tune for AMD, but if the AVX2 is listed as available a compiler should use it. This was pro
655.
by sliken 4y ago
Like say going from 80GB/sec memory bandwidth to 400GB/sec? Or maybe having the iGPU performance go up 4-8x so most won't need a $500+ GPU? I'm considering an Apple studio, once refreshed with the M2 Max, even for Linux.
656.
by sliken 4y ago
Sure. Various forums corroborate the reducing the Ryzen 7000 series at 65 TDP is often a less than 10% decrease in performance. Also keep in mind that "performance mode" on Intel can significantly increase the default 253 watt TD
657.
by sliken 4y ago
Does it happen, sure. Seems like Apple's increasingly happy to hit lower price points and disrupt the price structure for AMD and Intel based product lines. Similarly Intel seems to finally be getting it's act together and setting
658.
by sliken 4y ago
Sad, big win for the Ryzen or future Zen 4 threadripper.
659.
by sliken 4y ago
Assuming Intel didn't add code that's: if (AMD_CPU) { go_slow() } .... again
660.
by sliken 4y ago
Heh, I've been tracking desktops for quite some time, not noticed any significant reliability differences between AMD and Intel. My current desktop is a 2015 Xeon e3-1230 which was CHEAPER then the similar i7, but allowed ECC. I do

More ›