Soldato
This feels like the longest wait for a product launch ever!
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
- Status
Imagine my wait from early 28nm cards to actual sub-28nm cards :O
As stated in my previous post, if there are any issues it is something other than their capability to mass manufacture it, my guess is that it is something else related to how it HBM2 works. You can either mass produce something or you can't. At work (admittedly a different industry) we have something called off tool samples, where we manufacture components using the exact same method that we plan on using for volume production (the only difference being we produce 1 or 2 rather than hundreds). If there is any problem with manufacturing process it would be picked when trying to produce the off tool sample.
My guess is that they got off tool samples of HBM2 2.0 Gbps, found their manufacture process works (Hence put it in the catalog). However during long term testing they found a problem that only existed in 2.0 Gbps HBM2. Maybe its to do with how the electrons flow through the material at the current/voltage/amp/etc that is needed to sustain a transfer of 2.0 Gbps, maybe its a heat issue, maybe it degrades at an accelerated rate we don't know but it has hindered them being able to sell it. Either way it not a mass production problem, it is a problem with the actual design of the component. Maybe that is what you meant but that's not how i read your post or the people before you.
No actually. Not at all. Just because they showed a demo doesn't mean they have finished all the necessary work/testing on the card. Go look at car website where they have spy shots of cars months before they are released.
Soldato
- Joined
- 30 Nov 2011
- Posts
- 11,375
With microelectronics you run in to yields. You design the chip to be made on a particular process and you design it to have particular capabilities but you know not every chip will be perfect so one chip design can fulfill a place in a range of products. A single chip design gets tested and "binned" based on what it fails on. Good yields mean more of the higher spec chips can be sold, poor yields mean you struggle to make enough of the high spec which means the cost of those high spec modules is higher or not available in volume.
Soldato
- Joined
- 9 Jun 2011
- Posts
- 3,610
can;t wait for it
I stand corrected, I was not fully aware of how yields work. But this raises more questions. Would they not be able to calculate expected yields after a few test runs?
If they can surely they would have been aware that they are unable to produce enough high spec chips? And therefore it shouldn't have even made it into the product catalog?
Also there is a difference between mass.prixtion and early sampling. A new product like memory will start off with one machine setup to produced it, and often a smaller machine that has far lower capacity but more flexibility and ability to essentially debug and tweak.
In time the producer can increase production capacity be re-tooling more machines and optimizing the process so instead of producing x-parts per week they can produced 20x or more. A product may have difficulties going into full production.
In time the producer can increase production capacity be re-tooling more machines and optimizing the process so instead of producing x-parts per week they can produced 20x or more. A product may have difficulties going into full production.
Aren't you describing prototyping, rather than off tool samples.
That's certainly a good argument.
Look at even NVIDIA with Volta. Jensen claims they can get about 1 good V100 GPU from a wafer, hence them taking preorders now. Essentially built to order.
Production issues can certainly still happen considering the size and complexity, which would delay production of those GPUs very easily.
The same applies to 3D-Stacked DRAM aka HBM2. It's a very complex thing to produce, as it's many layered levels, and they can easily have run into issues pushing it to the 2.0Gbps specification for any reasonable amount of yields.
It certainly looks like they vastly overestimated their yields considering around 1.76-1.88Gbps is the best we're seeing a year after they entered "mass production" with HBM2.
We'll certainly find out more once the Vega Frontier Edition, and the Radeon RX gaming Vega is out. Although seeing the Radeon Pro Vegas Apple are getting seem to be running at 1.6Gbps, it doesn't look good for 2.0Gbps for consumer or workstation/enterprise use any time soon.
Single good/full specs meeting GPU per wafer basically means it costing thousands.
So no wonder if those are built to order instead of made in quantities to stock.
Even disregarding cost manufacturing is so slow.
Soldato
- Joined
- 30 Nov 2011
- Posts
- 11,375
You could just google chip production and yields and read up on the subject.
“Designing microprocessors is like playing Russian roulette. You put a gun to your head, pull the trigger, and find out four years later if you blew your brains out.”
Because as interesting as the topic would be, I simply don't have a lot of time to spare shifting through useless and or outdated bits of information on google to find the answer i'm looking for with such ambiguous search terms. You could have at least named the process which D.P. is referring.
Nonsense, the super efficient Netburst architecture powering the Pentium IV will hit 4GHz before 2003 and scale to 10GHz by the end of the decade! AMD will be finished!
(For anyone wondering yes that was Intel's planned roadmap in 2001).
Permabanned
AMD need to block cryptocurrency mining on Vega.
If Vega turns out to be brilliant for mining then apply will not exist for gamers. This will be bad in the long run.
Maybe only allow it on professional cards.
The 570/580 Polaris cards are not in the hands of gamers and so AMD has bo presence currently in the gaming market.
If Vega turns out to be brilliant for mining then apply will not exist for gamers. This will be bad in the long run.
Maybe only allow it on professional cards.
The 570/580 Polaris cards are not in the hands of gamers and so AMD has bo presence currently in the gaming market.
This isn't remotely feasible. AMD nor anyone else can or will block how you use their products or what companies or individuals can make software that interacts with their hardware.
Chances are it will be. It's basically just a (much?) higher perf/w evolution of AMDs current hardware, and they've all been good at mining (so I hear, never tried it myself).
Soldato
Not really no. It's suggesting that AMD and nVidia are essentially going to be able to stop any future software that they don't like, from working on their card.Just like Nvidia technically supporting freesync but not allowing it you mean?
For example, if they were to block existing mining software, it'd just take the software devs releasing an update to get around that.
They can't block compute ability either.
If they want to do anything about it, they can mandate that retailers restrict customers to, say, 2 cards per delivery address. Or something along those lines. That alone would make it difficult for people to order cards en masse for mining.
- Status