11–20 of 39 posts
Not naming the company but seems like HBM manufacturers might be going all-in to benefit from Nvidia's stock surge.
How about higher capacity GDDR6X? 48GB consumer cards (or 96GB pro cards) would sell like hotcakes if AMD/Intel dare to break the artificial VRAM segmentation status quo.
I swear I'm getting Deja Vu right now, I coulda sworn I've seen this thread before. There's gonna be a guy commenting that "you don't need it" somebody else saying "but I want it!" And a few trying to figure out the economics of it and whether or not it makes any sense. Personally I'd love to have as much VRAM as possible (and as high a bandwidth as is possible too) to mess around with simulations in- but that's defi…
How about higher capacity GDDR6X? 48GB consumer cards (or 96GB pro cards) would sell like hotcakes if AMD/Intel dare to break the artificial VRAM segmentation status quo.
Earlier quoted context omitted.
I swear I'm getting Deja Vu right now, I coulda sworn I've seen this thread before. There's gonna be a guy commenting that "you don't need it" somebody else saying "but I want it!" And a few trying to figure out the economics of it and whether or not it makes any sense. Personally I'd love to have as much VRAM as possible (and as high a bandwidth as is possible too) to mess around with simulations in- but that's defi…
A friend just got an M3 Max with 128GB (V)RAM and he's extremely happy with it (AI workloads wise). That could be an option if you can run your simulations on macOS.
Earlier quoted context omitted.
A friend just got an M3 Max with 128GB (V)RAM and he's extremely happy with it (AI workloads wise). That could be an option if you can run your simulations on macOS.
That machine with 128GB is $5000, with 48GB is still well over $3000, and has as much memory bandwidth as a $400 GPU. At the current spot price, 128GB of GDDR6 is <$400 and 48GB is <$150, implying that they could be paired with any existing <$1000 GPU to produce something significantly faster for dramatically less money. If anyone could be bothered to make one.
Looking at what’s available right now. You need 3 A100 40GB to get this amount of VRAM which will cost you way north of 20000$.
Doing it with A6000s is still about 15k$.
There’s not that many high VRAM options out there you know..
Earlier quoted context omitted.
That machine with 128GB is $5000, with 48GB is still well over $3000, and has as much memory bandwidth as a $400 GPU. At the current spot price, 128GB of GDDR6 is <$400 and 48GB is <$150, implying that they could be paired with any existing <$1000 GPU to produce something significantly faster for dramatically less money. If anyone could be bothered to make one.
I agree with you, that’s how it should be, but that’s not how it currently is. Looking at what’s available right now. You need 3 A100 40GB to get this amount of VRAM which will cost you way north of 20000$. Doing it with A6000s is still about 15k$. There’s not that many high VRAM options out there you know..
But all of this is kludges. The Radeon RX 7900 XTX has more than twice the memory bandwidth of the M3 Max with much better performance per watt than an array of P40s. What you want is that with more VRAM, not any of this misery.
How about higher capacity GDDR6X? 48GB consumer cards (or 96GB pro cards) would sell like hotcakes if AMD/Intel dare to break the artificial VRAM segmentation status quo.
AMD is doing the same thing, the only high memory cards they put out (MI300) are for data centers.
How about higher capacity GDDR6X? 48GB consumer cards (or 96GB pro cards) would sell like hotcakes if AMD/Intel dare to break the artificial VRAM segmentation status quo.
Don't you think Nvidia analyzed, sliced and diced the market to figure out how to maximize profits? AMD is doing the same thing, the only high memory cards they put out (MI300) are for data centers.
And Samsung just launched 36 GB packages of HBM3E: https://www.anandtech.com/show/21278/samsung-launches-12hi-3... Combined with this[1] interesting paper from summer 2023 on HBM combined with Xeon processors which would now allow for 144 GB on a single CPU. In theory at least. 1: https://lenovopress.lenovo.com/lp1738-implementing-intel-hig...
> launched Not true. They announced completion of development, with intention to begin mass production in H1. So a couple months out, at least. > Micron's memory roadmap for AI is further solidified with the upcoming release of a 36 GB 12-Hi HBM3E product in March 2024. So likely competitive timing with Samsung for 36/12.