Preferences

gpapilion
Joined 1,073 karma

  1. Realistically groq is a great solution but has near impossible requirements for deployment. Just look at how many adapters you need to meet the memory requirements of a small llm. SRAM is fast but small.

    I would guess their interconnect technology is what NVIDIA wants. You need something like 75 adapters for an 8b parameter model they had some really interesting tech to make the accelerator to accelerator communication work and scale. They were able to do that well before nvl 72 and they scale to hundreds of adapters since large models require more adapters still.

    We will know in a few months.

  2. I would think this is for rental fleets or bike share. The weight and design would seem to make sense for that. Though the single speed seems like and odd choice for that.
  3. This is not true. Almost all firmware is signed by every vendor, and there are standards from Intel and amd on implementation of code signing.

    Look up Intel pfr.

  4. The one vendor mentioned in the comments, AMI, is switching this code base to openbmc. Also it should be noted that often this software is system specific.
  5. The issues were durability, fire rate, and well power.

    I don’t know that the first two have changed significantly.

  6. I think that the private carriers are more likely to be helped by this, since they will manage the paperwork.

    It’s more likely a set of products that were shipping directly from factories disappears from the market. For example, the direct from factory Halloween costume.

    It could end up being a step backwards in living standards and access to daily luxuries.

  7. Gradual damage is consistent with over heating. I've seen racks of servers do the same thing.

    Overall, there is a continued challenge with CPU temperatures that requires much tighter tolerances both in the thermal solution. The torque specs need to be followed and verified that they were met correctly in manufacturing.

  8. It makes sense in any environment you have two workloads sharing compute from two parties, public clouds.

    The protection here is to ensure the vms are isolated. Without doing this there is the potential you can leak data via speculative execution across guests.

  9. I think it’s more pragmatic. We can eliminate hyperthreading to solve this, or increase memory safety at the cost of performance. One is a 50% hit in terms of vcpus, the other is now sub 50%.
  10. More generally beats better. That’s the continual lesson from data intensive workloads. More compute, more data, more bandwidth.

    The part that I’ve been scratching my head at is whether we see a retreat from aspects of this due to the high costs associated with it. For cpu based workloads this was a workable solution, since the price has been reducing. gpus have generally scaled pricing as a constant of available flops, and the current hardware approach equates to pouring in power to achieve better results.

  11. Scope, it’s all about scope of your team. Em to director requires opportunity as well as performance.

    For you that means focusing on a growing area of the company, and finding new areas to grow your team in. You also need to have a team of managers, who are growing their scope as well.

  12. I think this will eventually morph into apples server fleet. This in conjunction with the ai server factory they are opening makes a lot of sense.
  13. I don’t know this is significantly different than modern engines. They require special tools and software too.

    The bigger issue I think is most of the cars are teslas, which didn’t behave like a normal automaker for better or worse. For example the work done during the pandemic to avoid supply chain crunches may result in a maintenance headache a few years from now.

  14. The headline discussion on the podcast covers whether chatgpt is actually successful. They point to relatively few use cases emerging, and the continual or press around agi. They cover how there is now pressure to build an ads into the platform to build revenue.
  15. ...

    Fab + design... its apples to oranges.

    TSMC for example has 77k employees and looking to add 23k more.

    Packaging and testing are labor intensive, and require folks to be added in different geographies.

  16. They weren’t interested in creating an open solution. Both intel and AMD have been somewhat short sighted and looked to recreate their own cuda, and the mistrust of each other has prevented them from a solution for both of them.
  17. This is sort of pointless, because no matter how awesome it is it appears I have to pay....

    I'd also like to know more about what its doing.

  18. Consumers want faster processing the instructions are just the method to get there. And they aren’t the best since the area dedicated to the instruction could be used for something else.

    It is insane especially if you think emulation is performant enough to allow for a switch.

  19. Mishandling aside, the issue I've seen is there really isn't consumer demand for this. Prior to AMD having AVX512, most of the comments were around wasting the silicon on SIMD, rather than improving other aspects of the CPU. I'm pretty sure there was good reason to think it was largely a dark area of the chip.

    From what I've seen, but haven't heard discussed much, the naive implementation vs AVX512 is a huge gain, but AVX2 vs AVX512 was not very impressive for the application I was looking at. The complexity this code added, and the cases where we needed it to run on AMD (for other reasons), basically made taking advantage of the feature undesirable for a single digit gain.

    Things like VNNI or AMX are better wins, but they are only needed in very specific cases. VNNI in particular looked to be a 30% improvement in a BERT workload.

  20. I think the main problem is the investment in gpus delayed the adoption of ddr5 based platforms for servers.
  21. https://archive.ph/JCfqU

    To deal with the refresh loop.

  22. At the core of this article is that these platforms are setup for you to make a living off them alone. You see more successful influencers shilling vpns, and setting up newsletters on YouTube to stabilize revenue. TikTok has the same problem vine did, the content is short form so it a bit difficult to put a pitch in without making a mini infomercial.
  23. What’s the significance of UIPath?
  24. > The intercity rail part is the easy part. Getting people to actually use it requires a pretty decent city transit network on each end.

    I don’t think this is true. People take airplanes and rent a cars all the time. The same could be true for train travel. All of the ways someone would leave a train station generally exist.

    People don’t take trains for two reasons: 1. They take too long. 2. They are too expensive.

    For example sf to la takes at least ~9-13h and costs between 50-80 dollars. Versus a southwest flight for ~140 that takes an hour. For most people that extra 60 dollars for 7-11 hours is worth it.

  25. This is someone who was on disability or at the very least was entitled to protection under federal or state law. They are trying to limit liability, hence the very straight forward language.

    I would be surprised if everyone got these emails, and I would also be surprised to hear they sent lots identical to this.

  26. I think there have been two major drivers to inflation in fast food: - wages - delivery apps

    Wages is pretty obvious, and the article cited the changes in California. It is important to keep in mind that we’ve seen wages jump in the golden state 66% for fast food workers since 2019. That’s huge, especially since labor it’s the largest cost for these restaurants.

    Delivery apps added a new fee that showed consumers weren’t as price sensitive as they thought. So I think we’re seeing some increases related to that.

  27. This all has to do with how commercial real estate is valued; rent determines the valuation. If companies are unwilling to pay the rent the space sits empty waiting for a tenant willing to.

    The reason for this is if you take lower rent than previously your building value changes and your lenders make you cough up the difference to cover additional principal.

    Since the loans come due every 5ish years we’re now seeing buildings unable to justify their previous value, and having to walk away or get valued at their current rent.

    It’s going to be ugly for a while.

  28. I’d be happier with 93% of PyTorch but works on multiple gpu manufacturers.

This user hasn’t submitted anything.

Keyboard Shortcuts

Story Lists

j
Next story
k
Previous story
Shift+j
Last story
Shift+k
First story
o Enter
Go to story URL
c
Go to comments
u
Go to author

Navigation

Shift+t
Go to top stories
Shift+n
Go to new stories
Shift+b
Go to best stories
Shift+a
Go to Ask HN
Shift+s
Go to Show HN

Miscellaneous

?
Show this modal