Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
So let me see, when the watercooler was rumoured everyone went mental and I suggested that maybe, just maybe, there would be more than one edition of the card. Stock cooled and watercooled, then let the user decide what they want. I also said 8GB was more than a possibility. It also says up to 8GB, not just 8GB, this indicates 4 and 8GB versions which implies that there will be 4 stacks and both 1 and 2GB stack densities, again something I suggested was more than possible while others were posting up an irrelevant HBM 2.0 spec and insisting it wasn't possible for a couple of years.
It's almost like a small amount of research, some logical thinking, some comparisons to previous technologies and you can pretty accurately predict and translate what little information is available.
Watercooled card... what was my reasoning? Simple, a higher volume(compared to a 295x2 for instance) lower cost part ONLY having a AIO cooler option was exceedingly unlikely, what is the simplest alternative explanation..... two editions, stock cooler and AIO cooler. This really isn't rocket science.
Same with HBM, just read 2-3 short articles explaining the tech, would take 5 minutes and give you more than enough information to realise why 4 stacks made a large amount of sense again for a higher volume lower cost part. With every other JEDEC spec never, ever has the industry been limited to specs, they are used as a guideline to push everyone forward in the same direction, a minimum spec to achieve in the future. Again simple, never before has any memory maker been limited to ONLY Jedec standards, and 512MB sized chips are being made in GDDR5 since mid last year.
The industry is at the 512MB chips stage and there is no fundamental reason why HBM shouldn't be achieving relative parity in density on the same nodes with the same basic design. It's not GDDR5 but in general simpler due to lack of per die logic required. The ramp in density was a given and should have been quick, I think it will surpass density per chip in the not too distance future as well.
It's almost like a small amount of research, some logical thinking, some comparisons to previous technologies and you can pretty accurately predict and translate what little information is available.
Watercooled card... what was my reasoning? Simple, a higher volume(compared to a 295x2 for instance) lower cost part ONLY having a AIO cooler option was exceedingly unlikely, what is the simplest alternative explanation..... two editions, stock cooler and AIO cooler. This really isn't rocket science.
Same with HBM, just read 2-3 short articles explaining the tech, would take 5 minutes and give you more than enough information to realise why 4 stacks made a large amount of sense again for a higher volume lower cost part. With every other JEDEC spec never, ever has the industry been limited to specs, they are used as a guideline to push everyone forward in the same direction, a minimum spec to achieve in the future. Again simple, never before has any memory maker been limited to ONLY Jedec standards, and 512MB sized chips are being made in GDDR5 since mid last year.
The industry is at the 512MB chips stage and there is no fundamental reason why HBM shouldn't be achieving relative parity in density on the same nodes with the same basic design. It's not GDDR5 but in general simpler due to lack of per die logic required. The ramp in density was a given and should have been quick, I think it will surpass density per chip in the not too distance future as well.
Last edited: