KAIST startup Panmnesia (the name means “the power to remember absolutely everything one thinks, feels, encounters, and experiences”) claims to have developed a new approach to boosting GPU memory.
The company’s breakthrough allows for the addition of terabyte-scale memory using cost-effective storage media such as NAND-based SSDs while maintaining reasonable performance levels.
Originally Posted by chovy
to be honest I don’t know much about hardware. I just want like a 32 core server with fast cpu 128gb ram and whatever GPU they have in the $100/month range
Training is often more demanding than inference, all else being equal. AI training with “whatever gpu” is gonna end in tears and frustration. Have you tried your workload against some per-second billing instance first before making a longer term commitment?
Choking on lobster could be deadly.
‘);
var sidebar_align = ‘right’;
var content_container_margin = parseInt(‘350px’);
var sidebar_width = parseInt(‘330px’);
//–>
Without an efficient way to squeeze additional computing power from existing infrastructure, organizations are often forced to purchase additional hardware or delay projects. This can lead to longer wait times for results and potentially losing out to competitors. This problem is compounded by the rise of AI workloads which require a high GPU compute load.
ClearML has come up with what it thinks is the perfect solution to this problem – fractional GPU capability for open source users, making it possible to “split” a single GPU so it can run multiple AI tasks simultaneously.
‘);
var sidebar_align = ‘right’;
var content_container_margin = parseInt(‘350px’);
var sidebar_width = parseInt(‘330px’);
//–>