Level1Techs Forums Forum

I run my R9700 at the lowest settings the firmware allows. The power cap is lowered by 90W and the clock by 500MHz, and I also undervolted the GPU by 80 mV. It doesn’t impact token generation, it is bottlenecked by memory bandwidth, not the GPU, but it does reduce prompt processing by 15-20%.

instant video switching is possible with blackmagic atem, but it is not KVM switch. maybe next step is true glitch free video switching using hardware video clock synchronization with BNC trilevel genlock with older quadro cards and some of aja products

Mora: Those 2 million dollar fines mean nothing when they just made 100x that by being dodgy. Often corpos will do something wrong knowing they will be fined eventually, but do the calculus such that if the profits outweigh the fine, then they will do the bad thing a pay the fine and come out on top. In other words, they are actually incentivized to do the bad thing, as it is more profitable to do so, they are rewarded for bad behavior. The fines should scale by revenue or something, rather than some insignificant arbitrary amount. That way it doesn’t matter if you are a million dollar corpo, or a multi-billion dollar corpo, or a zillion dollar corpo; the fine should hurt just as much regardless of size.

mattlach: …but this is mostly irrelevant these days, at least from my understanding Yes and no. It has nothing to do with “these days” but about the size of your stripes. VM that use zvols use 16k by default. So ALL the data of that VM disk will be multiple 16k writes. Here the geometry and padding issues kicks in extremly hard. Datasets on the other hand use 128k by default. 128k is so big, the padding or geometry issues becomes less relevant. But that 128k is just a max value. It can be any multiple of 4k. So If you only put 4k files into your 128k record size dataset, every files will become a 4k record and suffer extremly from these issues. Link to efficiency table and explanation 128k is missing there, so we will do it for ourselves for a 128k record size, 8 wide, RAIDZ2: RAIDZ2 with 8 drives With 8 drives, we get a stripe 8 drives wide. But not all stripes have six 4k data sectors (24k) and two 4k parity sectors. For a volblock of 128k, we need 5.33 stripes (128k/24k). Five stripes have six 4k data sectors, which offers in total 120k data. These five stripes each have two 4k sectors for parity, in total 40k. The sixth stripe is smaller, needs only two 4k data sectors for the missing 8k (128 - 120). The sixth stripe also has two 4k sectors for parity. In total, we have 32 4k data sectors and 12 4k parity sectors. So 32 + 12 = 44 sectors. 44 sectors can't be divided by 3 so there is padding needed. We add one 4k padding sector. That gets us to 45k sectors. 45 can be devid

I think I found a solution. Disabling power management and mctp broadcast on that pcie port allows the server to boot successfully. I’m guessing the drive didn’t like something about the boot process, and simplifying the pcie bus options helped it power up ok. Thanks very much for your help.

Seriously, are women okay???

Hi all, got a new laptop for work - ASUS Zenbook 14 with a Core 9 Ultra 285H. It ran Windows okay but it would shut off with large e2e testing runs. I suspected overheating because it’s a thin laptop, too thin I think. With our Java + Node workflow the 32GB of memory was just not enough for parallel e2e testiung so I convinced them to let me trial Linux instead. I wanted the potential for team adoption so steered clear of bleeding edge stuff so it was Ubuntu 26.04 and I punted Ubuntu Pro as an option that helps with our compliance which landed well. Then the laptop started stuttering quite bad, it’d freeze for a good 5-10s - then be fine for a bit. The company sent me a replacement laptop, but since I work remotely, I had both laptops with me for a bit. The new one, which is practically identical (although not technically identical as I’d find out) - worked fine. I used Claude to help me diagnose, and at the end of the rabbit hole, I narrowed it down to the combination of: Micron SSD (the unaffected laptop used a Samsung SSD) Intel VMD enabled in the BIOS (default on) Using btrfs as the filesystem (doesn’t cause it but could make it worse) Simplest way to resolve it was to disable Intel VMD in the BIOS, and now the thing is flying as expected. I’m still curious why and hoping fellow Linux neckbeards can help me understand. Claude helped me get to the answer, but while we solved it from a usage standpoint, it’s explanation is quite speculative. So hoping to learn from this com

Yeah, but that’s far too limited for what I need it to do. Over the past time, I’ve looked into what I can do with LibreOffice. And while I can replicate most of what I need, it is a terrible workflow experience compared to Acrobat So for now, I’ll accept that while possible, it isn’t a workable replacement.

I’ve heard this idea periodically but not as often as I’d expect. Is p2p distribution just falling out of favor or is there something inherent about the problem space that would make this more challenging than any other file share. Metadata is just metadata after all?

The specific error from iLO is: Uncorrectable PCI Express Error Detected. Slot 6 (Segment 0x0, Bus 0xAE, Device 0x2, Function 0x0). Uncorrectable Error Status: 0x4000

That’s a really good idea, unfortunately I taped off the relevant pins and it made no difference. I found a 1-drive riser in a cupboard (one of the low-cost ones from ebay with almost no electronics on it), and tested in multiple slots, any time the nvme drive is plugged in the machine faults before completing post with a CRC error on the slot that the nvme drive is plugged into. The nvme drives should be backwards compatible because they worked in another machine with a pcie3.0 slot. Could it be an issue from the HP side with the drive bus negotiations? The SSD is a Lexar NM990.

0xDE57: accountability. fines when companies collect without consent or use dark patterns. And not just a slap on the wrist, I mean actual consequences for once. A full refund of the purchase price direct to the end user, along with the estimated financial gain they made through selling the information, including ad revenue. Those 2 million dollar fines mean nothing when they just made 100x that by being dodgy.