[CLAUDE.AI.]
can you theoretically make it have video output somehow?
Theoretically, it might be possible to add video output capabilities to an NVIDIA HGX H100 platform, but it would require significant custom engineering and modifications that go well beyond the scope of consumer-level tinkering. Here's a hypothetical approach:
Custom PCB design: You would need to design a custom printed circuit board (PCB) that can interface with the HGX H100 platform and provide the necessary connections for video output. This would require extensive knowledge of electrical engineering and GPU architecture.
Video output controller: Integrate a custom video output controller onto the PCB that can translate the GPU's output into a standard video signal, such as HDMI or DisplayPort. This would involve designing or adapting a compatible controller and ensuring proper signal integrity.
Cooling and power: Ensure that the custom PCB and video output controller have adequate cooling and power delivery to function properly in conjunction with the HGX H100 platform.
Mechanical integration: Design a custom enclosure or modify the existing HGX H100 enclosure to accommodate the new PCB and video output connectors.
Keep in mind that this is a purely theoretical exercise and would be an extremely complex and costly endeavor. It would require a team of highly skilled engineers and access to proprietary NVIDIA hardware and software. Furthermore, even if successful, the performance of such a modified system for gaming or consumer use would likely be suboptimal compared to purpose-built consumer graphics cards.
In reality, it is not practical or advisable to attempt adding video output capabilities to an HGX H100 platform for consumer use. It is essential to use the appropriate hardware for the intended application, and in the case of gaming and general consumer use, that would be consumer-grade graphics cards designed specifically for those purposes.
cant be that hard, modern cards can draw to a memory framebuffer which you could then copy over pcie back to the host and send to a "dummy" consumer gpu as a texture stream that gets mapped to a full-screen quad
The usual, they arrive on private-run delivery services on big-ass boxes with wide and big as frick foams and the delivery guys ask for id and double check to drop it.
I've bought it stuff (palo alto, hpe, poweredge, etc)
maybe in europe. i'm not in tech but i've bought 100k+ gear that arrives via common carrier. it's always some class a tractor trailer driver that looks like he just flew in from mexico. it's fricking disturbing
Because in the US serious truck transport is independant business as we still support the idea of a man providing for himself.
That guy got paid $2,500 to drive 600 miles in 14 hours. Frick Fedex, bunch of untrustworthy pussies. >just flew in from Mexico
And people who sell $100k equipment find those people more reliable than a massive common carrier. Fascinating.
3 months ago
Anonymous
independents don't really do LTL's. they straight loads. i don't know much about trucking but no one is going to send a high value LTL item through an independent carrier.
i just meant that we don't do specialized courier shit in the states that much.
i think fedex has a critical delivery service with 28ft 2 man box trucks
3 months ago
Anonymous
Medical devices are shipped via private courier all over town. Lots of similar stuff, but usually when it’s time sensitive.
Hopper doesn't support DirectX or Vulkan, but I think you can make it work with some driver frickery. MI300x can't play vidya at all, it's missing hardware.
If the AI firms werent allowed to use other peoples IP to generate the models they would have no money to pay for the hardware
So part of the price is the baked in lawsuits that will follow as the AI firms are trying to go with "hey we already did it so its fine to steal from others"
Does it even have video output?
[CLAUDE.AI.]
can you theoretically make it have video output somehow?
Theoretically, it might be possible to add video output capabilities to an NVIDIA HGX H100 platform, but it would require significant custom engineering and modifications that go well beyond the scope of consumer-level tinkering. Here's a hypothetical approach:
Custom PCB design: You would need to design a custom printed circuit board (PCB) that can interface with the HGX H100 platform and provide the necessary connections for video output. This would require extensive knowledge of electrical engineering and GPU architecture.
Video output controller: Integrate a custom video output controller onto the PCB that can translate the GPU's output into a standard video signal, such as HDMI or DisplayPort. This would involve designing or adapting a compatible controller and ensuring proper signal integrity.
Cooling and power: Ensure that the custom PCB and video output controller have adequate cooling and power delivery to function properly in conjunction with the HGX H100 platform.
Mechanical integration: Design a custom enclosure or modify the existing HGX H100 enclosure to accommodate the new PCB and video output connectors.
Keep in mind that this is a purely theoretical exercise and would be an extremely complex and costly endeavor. It would require a team of highly skilled engineers and access to proprietary NVIDIA hardware and software. Furthermore, even if successful, the performance of such a modified system for gaming or consumer use would likely be suboptimal compared to purpose-built consumer graphics cards.
In reality, it is not practical or advisable to attempt adding video output capabilities to an HGX H100 platform for consumer use. It is essential to use the appropriate hardware for the intended application, and in the case of gaming and general consumer use, that would be consumer-grade graphics cards designed specifically for those purposes.
desu.
Do you like CLAUDE.AI more compared to ChatGPT?
100%. sadly the newer version is cucked, but once you get used to it you can convince you do to what you want.
in terms of retention, hallucinating, summarizing large texts (even books), reading text from images, it's way better than GPT.
Does it even have graphical processing units?
Yes. It's still a GPU at the end of the day. Basically a beefed up version of Nvidia's mining cards without video ports.
It only has 24 ROPs, which are equivalent to the 32 ROPs of a GTX 680, due to the faster clocks. It'd basically struggle above 1080p/60.
cant be that hard, modern cards can draw to a memory framebuffer which you could then copy over pcie back to the host and send to a "dummy" consumer gpu as a texture stream that gets mapped to a full-screen quad
ur momma
the performance doesn't seem that high tho. think I read a 4090 reached 100TFLOPs under OC. but 640GB VRAM yeah
only 2kg? damn
what does ChatGPT run on?
on my wiener
He asked what it ran on, not what it was running away from because it's so small
An array of 1060s and 1070s
So does this ship fedex or UPS?
what happens when Dequan drops it off the back of his truck because he was lighting dat ill weed homie
The usual, they arrive on private-run delivery services on big-ass boxes with wide and big as frick foams and the delivery guys ask for id and double check to drop it.
I've bought it stuff (palo alto, hpe, poweredge, etc)
maybe in europe. i'm not in tech but i've bought 100k+ gear that arrives via common carrier. it's always some class a tractor trailer driver that looks like he just flew in from mexico. it's fricking disturbing
Because in the US serious truck transport is independant business as we still support the idea of a man providing for himself.
That guy got paid $2,500 to drive 600 miles in 14 hours. Frick Fedex, bunch of untrustworthy pussies.
>just flew in from Mexico
And people who sell $100k equipment find those people more reliable than a massive common carrier. Fascinating.
independents don't really do LTL's. they straight loads. i don't know much about trucking but no one is going to send a high value LTL item through an independent carrier.
i just meant that we don't do specialized courier shit in the states that much.
i think fedex has a critical delivery service with 28ft 2 man box trucks
Medical devices are shipped via private courier all over town. Lots of similar stuff, but usually when it’s time sensitive.
Hopper doesn't support DirectX or Vulkan, but I think you can make it work with some driver frickery. MI300x can't play vidya at all, it's missing hardware.
no it doesn't have any ROPs
>weight 2kg
lolwut
My 4070 weighs almost 1 kg.
>
Shame on you, Nvidia.
Can it run Crysis?
soon consumer level gpus will be in this price range too
for the low low price of only 3999.99$ you too can own the 5090 dear consumer!
no gaming drivers so nope unless implement your own
Any small size OAM/SXM motherboard/server available?
The Lenovo ThinkSystem SR670 V2 is the closest I could find.
>Lenovo
Hi Jarrod. Not this time.
What?
>The price includes all legal fees
If the AI firms werent allowed to use other peoples IP to generate the models they would have no money to pay for the hardware
So part of the price is the baked in lawsuits that will follow as the AI firms are trying to go with "hey we already did it so its fine to steal from others"
No moron, the price includes all taxes and duties imposed by the governments involved. Yuropoors have a hard-on for that kind of thing.
You dont have this yet? What? Are you poor or something?
Moshe Rosenburg
Poster from my Iphone
I thought that this thing has all the stuff responsible for 3D rendering stripped and only uses the die space for compute
Almost. It has 24 ROPs and 456 TMUs. It'd most likely cost more in terms of R&D to remove all of them. No video output though.
Pretty sure it would support 32k at 1000fps.