No they can't. That's the issue. Even Kaby-G is probably an Apple semi-exclusive.
Not the cpu, it would be 3D and more expensive that way. 3x better performance, but nevermind. It just occurred to me just a moment ago because the announcements never made much sense otherwise. In 2016, Samsung made an equal measure of promoting "mobile" 2 high stack HBM proposals. I think this is why it matters. Otherwise how do you market such a thing?Are you talking about Samsung's "low cost HBM" initiative? I never got the sense that they were skipping the interposer, just that they were working to lower costs. Are you proposing that they stack the HBM on top of the CPU? That would be a horrible idea, at least.
So you expect to see Intel Kabylake G with HBM2 but not Raven Ridge with HBM2. We will see. AMD is not going to let themselves be beaten by Intel in gaming. Thats my belief.
If AMD ship HBM2 controller integrated Vega GPUs to Intel for manufacturing Kaby G and allow Intel to overcome AMD's greatest advantage which is graphics and gaming performance then the management at AMD needs to be fired for incompetency.
That's even worse.
From a cost perspective HBM has a premium over DDR , the more you add , the worse it gets (then there is the interposer, yileds and other costs).
More importantly you add way too much heat to the package for an ultrabook , drastically decreasing the TDP available for the CPUGPU so end up with a much much slower solution at a higher cost. You don't need more BW to begin with at such a low TDP so the idea is ridiculous from the start.
Intel's Iris is an entirely different solution , not even close to being comparable in costs and TDP - not that it's a good solution or one that is competitive.Intel did it because there was no competition
A large discrete GPU with HBP ,maybe even paired with a SoC in a single package ( the SoC not on the interposer) does make sense from a form factor perspective but at much higher TDP and price. Achieves a similar goal to Nvidia's downclocked garbage that was just announced but does it better. The problem with that is the low volumes and they would need to use already existing products and just pack them together.
A much cheaper HBM with no interposer or a much cheaper interposer would enable for HBM to be used in more budget friendly products - starting with discrete GPUs that are not 500$+ but those kind of solution are just emerging.
Or to other OEM's. The plus side of making something on demand that has multiple uses is once they collect the development money from Apple any further uses is pretty much pure profit. All it takes is one OEM liking the idea and asking Intel for a supply and it works out great for Intel.Apple generally gets what they want though. And they really don't want to go back to nVidia because of the lawsuit but they certainly could....
I suppose if it's competitive enough Intel could try selling Kaby-G (plus a board since it's BGA) to the retail market.
Or to other OEM's. The plus side of making something on demand that has multiple uses is once they collect the development money from Apple any further uses is pretty much pure profit. All it takes is one OEM liking the idea and asking Intel for a supply and it works out great for Intel.
To be fair you did.
Very surprised (and somewhat crestfallen) AMD aren't doing at least one at the high end with HBM2. I'd have thought it the ideal weapon for accelerated compute in scenarios where customers couldn't justify or utilise racks of GPUs, but being able to pack into a 2 socket blade one socket having a conventional Naples and the other having a high end Zen+Vega APU utilising HBM.
Not to mention the obvious high end ultrabook etc.
What if the the use of HBM removes the need for system memory? This would be a big plus for Ultrabooks/Surface like products. A $200 increase in price for 8GB of HBM might be worth not having include RAM since that would be sourced by AMD. There is a reason products such as the Iris lineup exist.
What about 2 GB of HBM2 on the package?8GB of HBM would cost AMD more than the actual APU costs them to manufacture. And it means that AMD eats the cost of the RAM pre-packaging and carries the overhead costs until sale to the OEM. And OEM can get 8GB of DRAM for <$30 in contract pricing. Iris's edram costs a small fraction of the cost of a HBM stack and works with a bog standard package substrate.
Wait! Doesn't Infinity Band support up to 512GB/s? "512GB/s" = 2 stacks of HBM2. They could design it with a single HBM stack and 3 dies on the EPYC socket and 512GB/s total for I/O, HBM as L4. If edram bandwidth is counted double for r/w this applies I suppose.What about 2 GB of HBM2 on the package?
My point was specifically for packaging reasons on premium products. There is room even at the increased power usage for a 4 core CPU that also has a powerfulish igpu. Iris is an example of that. Something like a HBM2 setup would allow full Discrete GPU performance and alleviate need for ram. I am not saying we see a dozen of these show up. But a small Alienware, an option for Apple with their notebooks and Imacs (specially the Imacs since they have space for the dispersed cooling and putting the Mem options on the CPU means that they don't have differentiate the board construction). A future Surface or Surface book. List goes on. This is something that can get them design wins assuming they have the CPU in the pocket. None of these guys is going to wait for AMD to come up with a semi-custom design for them. Otherwise they can just continue working with Intel who will give them a deal just for not talking to AMD.
That said is RR the right release to do this. Probably not. But there is a packaging advantage that a HBM APU from AMD would fill and it would be in there best interest to have it ready for the opportunities as they open.
No way. You know that stacked memory leads to cooling issues for tablets, right? With ~2.5-4W SOCs? Stacking memory on top of a 150+W GPU would be a recipe for disaster. It would probably destroy itself rather quickly, even if it had some sort of thermal vias embedded into the HBM.That's what I am predicting AMD is doing with Navi.
You mean "3x better performance for the fraction of a second it takes it to overheat and throttle". The announcements made sense in regard to further developing an up-and-coming memory technology, making it suitable for more use cases in the future.Not the cpu, it would be 3D and more expensive that way. 3x better performance, but nevermind. It just occurred to me just a moment ago because the announcements never made much sense otherwise. In 2016, Samsung made an equal measure of promoting "mobile" 2 high stack HBM proposals. I think this is why it matters. Otherwise how do you market such a thing?
I tend to agree, but I also think that depends on how powerful it turns out to be. If it's comparable to a low-end dGPU, I'd see some takers. The problem with Iris is that while it's far faster than HD 620 and the like, it's not fast enough that that actually matters. 620 does all the acceleration and media playback you want, but Iris can't really game even at 720p. If KBL-G could replace a RX 550 or GT 1030, I would want one.Other OEMs probably aren't interested, if Iris Pro is any indication.
Obviously, your version of the mobile use case does not suit even your own judgement. It is more likely to come to EPYC sockets for the same reason.You mean "3x better performance for the fraction of a second it takes it to overheat and throttle". The announcements made sense in regard to further developing an up-and-coming memory technology, making it suitable for more use cases in the future.
If Samsung is talking about "mobile use case", that's smartphones, not PCs. That's my main point. But seriously, you can't stack anything on top of a power-consuming chip with a TDP above a couple of watts. As i said above, even tablets have thermal issues due to stacked memory (that's the main reason why iPads have so much better sustained performance than anything else in the tablet space, as they have the RAM separate from the SoC).Obviously, your version of the mobile use case does not suit even your own judgement. It is more likely to come to EPYC sockets for the same reason.
Does this look to you like standard cache coherency protocol of intercore traffic? Because it does to me. You couldn't have been more wrong. They conserve from even the 32-bit lanes of DDR4 memory for mobile and you are saying they will put a 512-bit interface? Wrong...If Samsung is talking about "mobile use case", that's smartphones, not PCs. That's my main point. But seriously, you can't stack anything on top of a power-consuming chip with a TDP above a couple of watts. As i said above, even tablets have thermal issues due to stacked memory (that's the main reason why iPads have so much better sustained performance than anything else in the tablet space, as they have the RAM separate from the SoC).
*snip*
It's worth noting that both of those images are labeled in ways that designate them as specifically not (at least initially) intended for desktop/laptop PCs. "Accelerators" exist in the server/datacenter space, definitely not in home PCs. "Server & Network" is a pretty clear exclusion of end-user systems. But it's interesting to see that those slides show them moving away from stacked chips. Good call, Samsung.*snip*
EPYC multi-die sockets aren't even for HEDT, they are for servers. Match!It's worth noting that both of those images are labeled in ways that designate them as specifically not (at least initially) intended for desktop/laptop PCs. "Accelerators" exist in the server/datacenter space, definitely not in home PCs. "Server & Network" is a pretty clear exclusion of end-user systems. But it's interesting to see that those slides show them moving away from stacked chips. Good call, Samsung.
It's worth noting that both of those images are labeled in ways that designate them as specifically not (at least initially) intended for desktop/laptop PCs. "Accelerators" exist in the server/datacenter space, definitely not in home PCs. "Server & Network" is a pretty clear exclusion of end-user systems. But it's interesting to see that those slides show them moving away from stacked chips. Good call, Samsung.
Both Intel & AMD put a sack of nails into this coffin of a silly rumor that doesn't make any business sense.Kaby-G is probably an Apple semi-exclusive.
Lisa Su said:We’re not looking at enabling a competitor to compete against our products.
Are you proposing that they stack the HBM on top of the CPU?
What?! Did I miss the news that talks about the technology going into this direction, HBM on top of CPUs? Can I have a source please?That's what I am predicting AMD is doing with Navi.
Those aren't true 3D, still. Package on package is for saving mounted space. TSVs don't pass between the dies, just the molds around them.Both Intel & AMD put a sack of nails into this coffin of a silly rumor that doesn't make any business sense.
What?! Did I miss the news that talks about the technology going into this direction? Can I have a source please?
Both Intel & AMD put a sack of nails into this coffin of a silly rumor that doesn't make any business sense.
LE: So many people seem to think this is real. Get on the news bandwagon folks.