bit_user usertests said: Intel's poor timing continues to impress years after the fact. Nah, the key detail is that they were purchased used . When new, I think the GB/$ wasn't that much better than DRAM. However, DRAM is now much more expensive, which makes alternative solutions like Optane DIMMs much more attractive. You really can't use this example to reason about Optane's market viability. Reply
JTWrenn bit_user said: Nah, the key detail is that they were purchased used . When new, I think the GB/$ wasn't that much better than DRAM. However, DRAM is now much more expensive, which makes alternative solutions like Optane DIMMs much more attractive. You really can't use this example to reason about Optane's market viability. Agreed, the main reason being that fab usage would be about the same for both so it's not like we would get cheaper anything out of it. Optane just like SSD's would be more expensive anyway. Optane was cool tech that was just not cheap enough to make sense. Reply
usertests bit_user said: Nah, the key detail is that they were purchased used . When new, I think the GB/$ wasn't that much better than DRAM. However, DRAM is now much more expensive, which makes alternative solutions like Optane DIMMs much more attractive. You really can't use this example to reason about Optane's market viability. The $/GB wasn't dramatically less than DRAM, but that could have improved as production scaled up. But they were losing money and didn't stay the course. Since it's off the table, we see solutions like High Bandwidth Flash stepping into the empty tier instead. I guess you can disregard write endurance if the model isn't changing rapidly. Reply
bit_user usertests said: The $/GB wasn't dramatically less than DRAM, but that could have improved as production scaled up. Not faster than DRAM, unless they managed to solve the issues that kept it from scaling in the Z-dimension like it was meant to. usertests said: Since it's off the table, we see solutions like High Bandwidth Flash stepping into the empty tier instead. HBF has yet to be deployed in any solution, so it has yet to be a proven alternative. HBF should have a structural price advantage over Optane, because it's true 3D NAND. Reply
hwertz i looked into pdimms before. Interesting, because the PERSISTENT part of them worked on very few chipsets. they're so cheap because that part won't work on most ssystems so most potential buyers take that to mean they won't work at all. But set to run without persistence, they just show up as somewhat slow DIMMS and work in far more machines, which is what it sounds like is being done here. Reply
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://www.tomshardware.com/tech-industry/artificial-intelligence/SPONSORED_LINK_URL
- https://www.tomshardware.com/tech-industry/artificial-intelligence/enthusiast-runs-1-trillion-parameter-llm-from-768gb-of-intel-optane-dimm-memory-sticks-local-kimi-k2-5-install-achieved-roughly-4-tokens-per-second#main
- https://www.tomshardware.com/subscription
- License to Stream: ‘007 First Light’ Coming to GeForce NOW With an Ultimate Bundle
- First ever Cray T3D Supercomputer goes up for auction with $81,000 reserve — Europe’s fastest supercomputer in June 1996 goes on the block
- Arctic Freezer 36-S Review: Small size, effective performance, low price
- Linked and Loaded: Gaijin Single Sign-On Now Available on GeForce NOW
- License to Stream: ‘007 First Light’ Coming to GeForce NOW With an Ultimate Bundle
Informational only. No financial advice. Do your own research.