• 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.

AMD and Intel teaming up

4K8KW10 said:
Intel isn't simply an AMD enemy, it's an enemy because of which AMD is still close to and haven't escaped the threat of bankrupcy.

Remember all those anti-competitive practices Intel have been using during past all these years just to damage and hurt to the maximum AMD.
Why will AMD suddenly forget all these and join the dirty games? :confused:

Click to expand...

Whatever the history between Intel and AMD Intel have gone to AMD with some money and said we want to buy your products, As Steampunk put it its not a good idea to say no because that would give Intel no alternative but to go to nVidia with the same offer, that would serve to put AMD's only other competitor in a stronger position, as it is this weakens nVidia while strengthening AMD.

Also from a more pragmatic point of view you can't keep hating and dwelling on history, perhaps this is not Intel's hand of friendship to AMD but frankly i think it would be a good thing if this is a start of a bit more friendliness and respect between these two, we could all benefit from that and Intel - AMD do have a common enemy in nVidia, yes Intel take the threat from nVidia very seriously, perhaps Intel might figure its a good thing to have a friend in AMD, no?
 
4K8KW10 said:
I have already understood what happened. But I don't understand why AMD didn't just say "Oh, no, thanks, we are going to make it on our own and sell it under the AMD brand."
Click to expand...
1. AMD didn't have the know how at the point this design was signed off, probably for Apple; EMIB seems to be the key thing here.
2. At that same point in time AMD couldn't show a CPU that was good enough or Apple were wary of committing to a company that was still in a financial mess.
3. There's no reason why AMD can't offer both the CPU + GPU in the future if they can convince Apple they have the tech and can supply in the volumes required.
4. So this is a good first step for AMD and even if they never get to supply the CPU it's still a GPU design win.
 
Apple is moody, changes suppliers like women change their clothes. I wouldn't take their position seriously even despite how much cash they have.
Point 2: yeah, instead of investing in AMD, they will judge AMD, like the whole market takes the Intel (or the dark) side... :o
 
4K8KW10 said:
Point 2: yeah, instead of investing in AMD, they will judge AMD.
Click to expand...
It's not a matter of judgement, you look at the basics and if the company can't deliver the exact product you want with a high degree of certainty why take the risk?
You don't risk your own company like that unless you are desperate and Apple are hardly that and are fairly conservative.
 
Rroff said:
Wait weren't people claiming it would take Intel years and years to develop a multi-die interconnect and be years behind IF?
Click to expand...

EMIB is nothing like infinity, EMIB is akin to the silicon interposer that AMD use on Fury/Vega. Also the EMIB is ONLY used between the AMD gpu and the HBM2, the CPU is separated and connected over organic packaging.

Infinity fabric can be used over a silicon interposer, over organic packaging, over pci-e, over HT, over anything, it's the method of communication signalling, it's the processing and handling before sending it over literally any connection of any type. EMIB is a connection, it's nothing to do with signalling.

In effect, from what I understand, rather than with interposers having all the TSVs inside the chips themselves then connecting with one basic layer of copper connections in a cheap as chips interposer, it's basically taking 'normal' chips and sticking all the complex stuff at a higher level in the EMIB chip and Intel claim this is way cheaper but there is little to no evidence of this in the real world. Silicon interposers are dirt cheap because they can be made on older process nodes so can be made in cheaper fabs which don't have a huge demand on them anyway and they are just a few metal layers rather so take a very very small fraction of the time to process that a normal wafer does. If a normal 40nm wafer now costs along the lines of say $2000 to process taking 4 weeks, a interposer wafer probably costs $200 and getting 80-100 interposers off it even for chips like a Vega. Compared to what probably costs $0.5 for an organic interposer it's 'expensive' because it costs 300% more, but it's still only $2 and allows a large reduction in power.

EMIB uses smaller interposers but also makes them more complicated and requires sticking more of the die to die interconnect in a smaller area, this isn't usually a problem, most of the infinity connections in Zen are at the edges anyway, though that is partly because it uses organic packaging, if it used a full interposer maybe the connections die to die would be more direct, more numerous and offer lower latency as rather than going through one main line to each die, almost every part of every die could have a direct connection. That is the cost, EMIB is in effect, a baby interposer, but when an interposer has overall very small costs it's a pretty pointless inbetween step anyway. As shown, you can achieve pretty similar to EMIB at an organic packaging level as AMD have done with Ryzen.

EDIT:- Also to point out, EMIB has been talked about for a couple of years at least, hence when people were saying Intel don't have anything like infinity yet, they already knew about EMIB. Again in reality, a full interposer is realistically more advanced and offers the ability to make far larger number of die to die connections.
 
Last edited:
smilingcrow said:
1. AMD didn't have the know how at the point this design was signed off, probably for Apple; EMIB seems to be the key thing here.
2. At that same point in time AMD couldn't show a CPU that was good enough or Apple were wary of committing to a company that was still in a financial mess.
3. There's no reason why AMD can't offer both the CPU + GPU in the future if they can convince Apple they have the tech and can supply in the volumes required.
4. So this is a good first step for AMD and even if they never get to supply the CPU it's still a GPU design win.
Click to expand...

Yup, if this was finalised last year it would have had work started likely a year before that. Zen has had delays since then, Apple knew they could have delays, Intel already had 14nm chips by then and even if 10nm was late putting a 14nm chip with it would have been easy and it was fairly well known that even if AMD achieved Zen targets it wouldn't be particularly faster than Intel's chips core for core anyway. Intel CPU + AMD GPU was a dramatically safer bet than Zen for a company like Apple. If this launches very soon(not sure when it's supposed to) and Raven Ridge got pushed back another quarter then Apple get screwed.

I do expect though that once AMD has established Zen, established performance and gained that trust that they could be a very viable option for their own version of this chip, that being a customised Raven Ridge with a larger GPU and HBM all on one die to save even more power.

I do think there is a much higher chance that Mac pros go Threadripper or EPYC alongside Vega 20 chips at some point next year and if Apple are happy with the performance, supply and pricing it will encourage them to move to AMD for their other products on CPU as well.
 
BTW,looking at the AdoredTV video on the announcement he mentioned this:

https://www.pcper.com/category/tags/hades-canyon-vr

url.png


Looks like there will be 65W and 100W desktop parts too.
 
@drunkenmaster the problem with interposers is that these processes are designed for chips and with large chips (e.g. vega) you end up hitting the reticle limit.

EMIB doesn't need TSVs as the TSVs are to connect the chip "through" the interposer to the substrate.

With EMIBs you could construct a large "chip" that was way beyond what would be possible with a single interposer. For example, you could make a version of Vega that was 4 Vegas all linked together by EMIBs with HBM2 linked on to each. Vega is already at the reticle limit with it's interposer, so this would not be possible otherwise.
 
I find it a bit odd myself. AMD are in the market of CPU's and GPU's but are helping a competitor to take the fight to Nvidia with a CPU&GPU combination. Why can't AMD produce both the CPU and GPU for themselves? If there's a big demand in this area that Intel are trying to compete with Nvidia on.....why can't AMD take the fight to them both.
Also a bit odd that AMD are not producing the best GPU's at the moment. So Intel taking the fight to Nvidia using a competitor fairly weak in that area at the moment.

I hope they're pwned by Nvidia for this daft collaboration :D.
Showing the "weak" position AMD always seem to have unfortunately.
 
DarrenM343 said:
I find it a bit odd myself. AMD are in the market of CPU's and GPU's but are helping a competitor to take the fight to Nvidia with a CPU&GPU combination. Why can't AMD produce both the CPU and GPU for themselves? If there's a big demand in this area that Intel are trying to compete with Nvidia on.....why can't AMD take the fight to them both.
Also a bit odd that AMD are not producing the best GPU's at the moment. So Intel taking the fight to Nvidia using a competitor fairly weak in that area at the moment.

I hope they're pwned by Nvidia for this daft collaboration :D.
Showing the "weak" position AMD always seem to have unfortunately.
Click to expand...

Indeed. They cannot compete with intel or Nvidia at the moment. Maybe this can bring in revenue so they can get a decent r&d budget.
 
rtho782 said:
@drunkenmaster the problem with interposers is that these processes are designed for chips and with large chips (e.g. vega) you end up hitting the reticle limit.

EMIB doesn't need TSVs as the TSVs are to connect the chip "through" the interposer to the substrate.

With EMIBs you could construct a large "chip" that was way beyond what would be possible with a single interposer. For example, you could make a version of Vega that was 4 Vegas all linked together by EMIBs with HBM2 linked on to each. Vega is already at the reticle limit with it's interposer, so this would not be possible otherwise.
Click to expand...

In reality this isn't true simply because there is absolutely nothing stopping using more than one interposer already. EMIB is Intel trying to seem like they are totally up to speed with interposers. There is nothing stopping one chip sitting on top of the join between two interposers then having a range of chips on each interposer connected to that central chip. The only real advantage is that the chips themselves are smaller, but when costs are already very low saving 1-2 dollars doesn't mean much and then you get into increasing problems with EMIB in terms of thermal expansion which are not insignificant. Mixing organic and silicon interposers both connected to the same chips above at the same time is significantly more complex than putting everything on an interposer at one layer and everything on organic at another layer.
 
drunkenmaster said:
EMIB is nothing like infinity, EMIB is akin to the silicon interposer that AMD use on Fury/Vega. Also the EMIB is ONLY used between the AMD gpu and the HBM2, the CPU is separated and connected over organic packaging.

Infinity fabric can be used over a silicon interposer, over organic packaging, over pci-e, over HT, over anything, it's the method of communication signalling, it's the processing and handling before sending it over literally any connection of any type. EMIB is a connection, it's nothing to do with signalling.

In effect, from what I understand, rather than with interposers having all the TSVs inside the chips themselves then connecting with one basic layer of copper connections in a cheap as chips interposer, it's basically taking 'normal' chips and sticking all the complex stuff at a higher level in the EMIB chip and Intel claim this is way cheaper but there is little to no evidence of this in the real world. Silicon interposers are dirt cheap because they can be made on older process nodes so can be made in cheaper fabs which don't have a huge demand on them anyway and they are just a few metal layers rather so take a very very small fraction of the time to process that a normal wafer does. If a normal 40nm wafer now costs along the lines of say $2000 to process taking 4 weeks, a interposer wafer probably costs $200 and getting 80-100 interposers off it even for chips like a Vega. Compared to what probably costs $0.5 for an organic interposer it's 'expensive' because it costs 300% more, but it's still only $2 and allows a large reduction in power.

EMIB uses smaller interposers but also makes them more complicated and requires sticking more of the die to die interconnect in a smaller area, this isn't usually a problem, most of the infinity connections in Zen are at the edges anyway, though that is partly because it uses organic packaging, if it used a full interposer maybe the connections die to die would be more direct, more numerous and offer lower latency as rather than going through one main line to each die, almost every part of every die could have a direct connection. That is the cost, EMIB is in effect, a baby interposer, but when an interposer has overall very small costs it's a pretty pointless inbetween step anyway. As shown, you can achieve pretty similar to EMIB at an organic packaging level as AMD have done with Ryzen.

EDIT:- Also to point out, EMIB has been talked about for a couple of years at least, hence when people were saying Intel don't have anything like infinity yet, they already knew about EMIB. Again in reality, a full interposer is realistically more advanced and offers the ability to make far larger number of die to die connections.
Click to expand...

Taking my post way seriously :p was just being tongue in cheek about how some people were claiming Intel weren't even close to being able to produce any kind of multi-die design never mind something like IF.
 
Rroff said:
Taking my post way seriously :p was just being tongue in cheek about how some people were claiming Intel weren't even close to being able to produce any kind of multi-die design never mind something like IF.
Click to expand...

No one ever said that, maybe that's why he took you seriously.
 
drunkenmaster said:
In reality this isn't true simply because there is absolutely nothing stopping using more than one interposer already. EMIB is Intel trying to seem like they are totally up to speed with interposers. There is nothing stopping one chip sitting on top of the join between two interposers then having a range of chips on each interposer connected to that central chip. The only real advantage is that the chips themselves are smaller, but when costs are already very low saving 1-2 dollars doesn't mean much and then you get into increasing problems with EMIB in terms of thermal expansion which are not insignificant. Mixing organic and silicon interposers both connected to the same chips above at the same time is significantly more complex than putting everything on an interposer at one layer and everything on organic at another layer.
Click to expand...

At that point, with multiple interposers, you basically have EMIB just with larger EMIBs and therefore TSVs.
 
You must log in or register to reply here.
Back
Top Bottom