Announcement

Collapse
No announcement yet.

AMD Working To Release MES Documentation & Source Code

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #41
    Originally posted by coder View Post
    Call it what you want, but it's meant to be similar in capability and function to far more expensive AI servers.
    .​
    Indeed it is, because AMD offers these capabilities on these cards, The scale is no where close however.​

    Originally posted by kattjaevel View Post

    You complain about the Phoronix community but your response is hardly does much to raise the bar.

    I'm aware about Tinygrad and that George Hotz has manage to raise funds for it. That's great and all. I do think it's arrogant to go to Twitter and start making demands. Starting of with a public dispute does not exactly inspire confidence in your not yet released product (Tinybox). They have shipped zero units. With Hotz background and their bafflingly simplistic website I would think it sound to skeptical. Their whole idea seems to be to take cheaper consumer grade hardware and offer it up as a cost-efficient alternative for small to medium businesses and labs. Which is quite similar to what companies have done in crypto mining. Why use expensive enterprise parts? Well maybe because they are always better quality and better supported than consumer parts? It's a tale as old as the market itself. They want quality but they don't want to pay for it. The big difference being that now it's AI that's hype train instead of crypto mining. we have all seen where this train leads should it take off. Chip shortages and higher prices for consumers.
    I'm not sure why hotz' background would make anyone skeptical. Comma.ai is a company having success, a little early to call it successful, but their open pilot works well, and they have devices shipping. Tinygrad is a pretty succsessful product too, I'm not sure if you have played with any of the demos, but it works pretty well. I don't know why people keep bringing up the website, Yes it's simplistic, but who cares?

    If AMD didn't want people running compute, they shouldn't have given their cards compute capabilities. Keep in mind that yes, the tinybox is a more professional device, but AMD is known for lackluster quality for a reason. These issues do not just effect tinycorp/tinygrad. This is why this has gotten so much attention in the firstplace. AMD is terrible for compute. AMD's high quality equipment is not exactly free of these issues either. This is why Nvidia is still so dominant. Because no one in their right minds will go AMD. Tinycorp sure as shit isn't in their right mind with how far they are willing to go to support AMD cards. Going so far as to pretty much implement their stuff at a level equal to or below rocm. The reason he can make these demands is because so many people have high hopes for the work is going to do for pushing AMD forwards.

    Comment


    • #42
      Originally posted by coder View Post
      Call it what you want, but it's meant to be similar in capability and function to far more expensive AI servers.
      Just like during the mining boom, eh?
      RTX 4090.
      The 7900 GRE, 7900 XT, and 7900 XTX all use the same dies and PCB.
      Also, what we saw with mining is that even when an adjacent card was popular with miners, it would push up prices on cards around it, as demand for the popular card got displaced towards others.​
      do you really believe that gamers who want cheap cards really could buy such vollumes of hardware of all the 30-40 billion dollar High NA Extreme ultraviolet lithography factories build in the last 3-4 years ?

      without cryptoming and without compute/HPC and without AI deep learning no one would invest so much money in 30-40 billion High NA Extreme ultraviolet lithography factories....

      so its absolute nonsense to blame these technologies for high prices.

      people always claimed AMD increased the price and it was always wrong just compare vega64 with a 6600XT same price the 6600XT had lower power consumtion and higher performance and also AV1 decode. then again 7600XT now this card is even cheaper than the vega64 was in 2017 keep in mind the vega64 was 665€-740€ at release in 2017 now see the 7600XT/7800xT/7900GRE ... the 7900GRE is the same price as vega64 but much faster and again less power consumtion and also new technolgies like AI accelerators and also AV1 decode and so one.

      this nonsense amd did increase the price only because a 7900XTX is more expensive than a vega64 is nonsense.

      all these cards 7600XT/7800xT/7900GRE beat the vega64 easily at the same price or lower price with higher performance and lower power consumtion.

      now tell me who did ever increase gpu prices ? it was always only short period price fluctuations and always only the newest generation who is heavily scalped by fraudsters.
      Phantom circuit Sequence Reducer Dyslexia

      Comment


      • #43
        Originally posted by Slartifartblast View Post
        Sigh, AMD needs to learn from Nvidia CUDA as it will run on a lowly GT1030, not spectacularly mind but it runs and a breeze to install. Most users only care if something is easy to install and it does what it says it does. Frankly ROCm is a mess and the way AMD are currently trying to position it for high end and latest cards only is doomed to failure and suicide, not to mention being a pain in the arse to install.
        PS Mike, the website is really slow today, almost DDos like slow.
        you are right it was wrong from amd to officially drop the support for vega64 ... polaris on the other hand many 4GB vram cards they are really not usefull anymore for GPU compute or AI or whatever reason is Ai and also blender want a lot of vram 8GB vram is really the absolut minimum.

        its maybe not nessesary to support old cards like the GT1030 but anything usefull who has more than 8gb vram and anything newer than vega really should be supported.

        but also keep in mind the Polaris/480/580 even with 8gb vram is not so usefull for AI and the reason is no support for mini-float formats polaris really has 32bit floading point hardware only... with vega64 the 16bit mini float hardware was introduced. and even this is not want many AI wants

        modern AI hardware has support for 4bit minifloat and 6bit minifloat and 8bit minifloat...

        so lets say anything newer than vega64 really should be supported. but of course for amd vega is a technical legacy burden...
        Phantom circuit Sequence Reducer Dyslexia

        Comment


        • #44
          Originally posted by duby229 View Post
          That's not true at all... AMD hired every single open source developer that was making significant contributions.... They are already on AMD's payroll...
          its pretty sure that these people are not interested in the truth... of course AMD does hire anyoe who make significant contributions.

          they really are not interested in the truth.
          Phantom circuit Sequence Reducer Dyslexia

          Comment


          • #45
            Originally posted by MadCatX View Post
            Can we take a moment to appreciate how odd this is? Tiny Corp wants to make business with AMD so they literally offer to fix their dumpsterfire of a kernel driver and firmware code *for free*, just so they can purchase their products? Thank you Tiny Corp and George Hotz, hopefully you can help both yourselves and AMD.
            Well George is a hacker, a real one, not a poser. Watching his occasional YouTube streams he's been doing you can see how passionate he is.
            If this doesn't work out at least he's tried everything.

            Comment

            Working...
            X