AMD CPUs, SoC Rumors and Speculations Temp. thread 2

[8350rocks]

all im saying is amd HAS delivered in the past 5 years to what they said IPC gains would be and what power improvement will be. Carrizo uses 40% of the power kaveri used at idle. Carrizo is 10% faster than piledriver at the same clocks even without an L3 cashe.

I understand why you think they wont deliver, but look at the last 5 years and tell me they wont deliver and I don't buy it.

I think they can hit the 40% target they claim over Excavator. My only issue is that it won't be available when the 16nm GPUs launch. Next year we will be seeing a lot of DX12 games and I've been holding off for a CPU upgrade.

I agree...40% is a big jump, but a new uarch is a completely different design. I would not even be surprised if they came out slightly ahead of that to be honest.

The clock speed regression possibility is what slightly concerns me; however, I am hearing that there is pretty good confidence in the node/process...so there is that.

 

[juanrga]

all im saying is amd HAS delivered in the past 5 years to what they said IPC gains would be and what power improvement will be. Carrizo uses 40% of the power kaveri used at idle. Carrizo is 10% faster than piledriver at the same clocks even without an L3 cashe.

I understand why you think they wont deliver, but look at the last 5 years and tell me they wont deliver and I don't buy it.

I think they can hit the 40% target they claim over Excavator. My only issue is that it won't be available when the 16nm GPUs launch. Next year we will be seeing a lot of DX12 games and I've been holding off for a CPU upgrade.

I agree...40% is a big jump, but a new uarch is a completely different design. I would not even be surprised if they came out slightly ahead of that to be honest.

The clock speed regression possibility is what slightly concerns me; however, I am hearing that there is pretty good confidence in the node/process...so there is that.

I guess you mean that source that claims a base clock of 3.8 GHz, with a "Richland-esque refresh" to 4.0--4.1 GHz, right?

 

[ComputerSecurityGuy]

DX12 will let people hold off the CPU upgrade, results are showing that lower end CPUs see big improvements with DX12. A new GPU is probably a must next year though.

 

[Gaidax]

If Zen end up being 85% performance at 60% price against similar grade solution from Intel, that would be already good in my book.

Zen matching Intel offerings at high end new non-server space? Not happening. Even if by some miracle it does, then Intel will simply unleash all their hexacores for current i7 or even i5 prices and be done with it.

 

[esrever]

If Zen end up being 85% performance at 60% price against similar grade solution from Intel, that would be already good in my book.

Current AMD cpus already do this most of the time.

 

[jdwii]

Made a good point if Intel even for a second thought Amd was a threat they would just make i3 unlocked quad cores, tbh if they just made a unlocked i3 for 130$ that would pretty much get rid of amd for gamers. When they made the unlocked Pentium lots of people picked that instead even though I personally thought the 860k made more sense in most cases.

Either way what will probably happen is Amd is forced to position themselves to Intel prices the one who holds top performance dictates the price of their competition.

 

[NubUser]

when are zen cpu's expected to release?

 

[jimmysmitty]

Made a good point if Intel even for a second thought Amd was a threat they would just make i3 unlocked quad cores, tbh if they just made a unlocked i3 for 130$ that would pretty much get rid of amd for gamers. When they made the unlocked Pentium lots of people picked that instead even though I personally thought the 860k made more sense in most cases.

Either way what will probably happen is Amd is forced to position themselves to Intel prices the one who holds top performance dictates the price of their competition.

I don't think they would. In fact I doubt Intel would. They wont do anything tho be the ones that pushed AMD into deeper trouble. If anything they would make it a price war on the higher end. They need AMD to survive so they don't get called a monopoly again and be forced to break up.

 

[Vogner16]

Made a good point if Intel even for a second thought Amd was a threat they would just make i3 unlocked quad cores, tbh if they just made a unlocked i3 for 130$ that would pretty much get rid of amd for gamers. When they made the unlocked Pentium lots of people picked that instead even though I personally thought the 860k made more sense in most cases.

Either way what will probably happen is Amd is forced to position themselves to Intel prices the one who holds top performance dictates the price of their competition.

I don't think they would. In fact I doubt Intel would. They wont do anything tho be the ones that pushed AMD into deeper trouble. If anything they would make it a price war on the higher end. They need AMD to survive so they don't get called a monopoly again and be forced to break up.

they could argue that qualcom and apple and Samsung prevent them from having a monolopy :/

 

[prtskg]

when are zen cpu's expected to release?

Expect it in Q1 2017.

 

[juanrga]

likely 4-6 cores and 8-12 CUs on the top end APU...assuming at most a combination of something like 4c/12CU-6c/10CU

does anyone have a link to something to shed some light to this? 4 core 12 CU would be amazing, but maybe too good to be true.

I know juan can do some math to determine based on die size 😛

The new nodes give AMD enough space for 16CUs. They keys are memory bandwidth and dGPU business. Will AMD include expensive HBM on mainstream APUs or only top APUs (like the HPC APU will get it)? And will AMD eat its own dGPU market with powerful consumer APUs or will prefer people to purchase APU+dGPU instead?

 

[Vogner16]

The new nodes give AMD enough space for 16CUs. They keys are memory bandwidth and dGPU business. Will AMD include expensive HBM on mainstream APUs or only top APUs (like the HPC APU will get it)? And will AMD eat its own dGPU market with powerful consumer APUs or will prefer people to purchase APU+dGPU instead?

wouldn't amd introducing high end APU's to eat mid level gpu market come off as a major design win? this would take away from products such as R7 series gpu's but also take away from all of nvidias sales in that sector as an APU as fast as a 950 would be cheaper than an i5 and a 950... and it would also drive away intels sales in low end cpu's for gamers.

That strat would actually be a major design win. IF they had the HBM bandwith for Igpu to have good power.

EDIT: if amd has 20% devices on market now, and majority of PC gamers are in the R7 type category and below, then it would eat its own 16% and also another 50% from other vendors. essentially taking 16% money away from Radeon group and massively boosting money into APU group by 66%. total net gain for amd is 50% boost and then APU group can just cut a check to high end graphics from APU profit to maintain that amd has excellent high end gpu for marketing reasons, despite its lack of making money.

 

[NubUser]

The new nodes give AMD enough space for 16CUs. They keys are memory bandwidth and dGPU business. Will AMD include expensive HBM on mainstream APUs or only top APUs (like the HPC APU will get it)? And will AMD eat its own dGPU market with powerful consumer APUs or will prefer people to purchase APU+dGPU instead?

wouldn't amd introducing high end APU's to eat mid level gpu market come off as a major design win? this would take away from products such as R7 series gpu's but also take away from all of nvidias sales in that sector as an APU as fast as a 950 would be cheaper than an i5 and a 950... and it would also drive away intels sales in low end cpu's for gamers.

That strat would actually be a major design win. IF they had the HBM bandwith for Igpu to have good power.

EDIT: if amd has 20% devices on market now, and majority of PC gamers are in the R7 type category and below, then it would eat its own 16% and also another 50% from other vendors. essentially taking 16% money away from Radeon group and massively boosting money into APU group by 66%. total net gain for amd is 50% boost and then APU group can just cut a check to high end graphics from APU profit to maintain that amd has excellent high end gpu for marketing reasons, despite its lack of making money.

The broadwell chip with iris pro is also a great APU if priced well

 

[NubUser]

when are zen cpu's expected to release?

Expect it in Q1 2017.

i dont know if i can hold of my cpu purchase that long -_- i though it was somewhere in 2016

 

[salgado18]

wouldn't amd introducing high end APU's to eat mid level gpu market come off as a major design win? this would take away from products such as R7 series gpu's but also take away from all of nvidias sales in that sector as an APU as fast as a 950 would be cheaper than an i5 and a 950... and it would also drive away intels sales in low end cpu's for gamers.

That strat would actually be a major design win. IF they had the HBM bandwith for Igpu to have good power.

EDIT: if amd has 20% devices on market now, and majority of PC gamers are in the R7 type category and below, then it would eat its own 16% and also another 50% from other vendors. essentially taking 16% money away from Radeon group and massively boosting money into APU group by 66%. total net gain for amd is 50% boost and then APU group can just cut a check to high end graphics from APU profit to maintain that amd has excellent high end gpu for marketing reasons, despite its lack of making money.

If the new APU is as fast as a current 950, remember that next year (next week? ) there will be new GPUs in the market, so the 950 will become the X40 series. Current APU is as fast as an R7 250, but while it will be faster in the next iteration, all GPUs will be faster too.

That said, even if they remain in the same relative place performance-wise, an APU that runs every game on High at 1080p is a great deal.

Add to that Zen cores that perform almost as well as Intel with much better graphics, you may have an amazing product. I still think it wouldn't be that big a market for them, but still a great product.

 

[jimmysmitty]

The new nodes give AMD enough space for 16CUs. They keys are memory bandwidth and dGPU business. Will AMD include expensive HBM on mainstream APUs or only top APUs (like the HPC APU will get it)? And will AMD eat its own dGPU market with powerful consumer APUs or will prefer people to purchase APU+dGPU instead?

wouldn't amd introducing high end APU's to eat mid level gpu market come off as a major design win? this would take away from products such as R7 series gpu's but also take away from all of nvidias sales in that sector as an APU as fast as a 950 would be cheaper than an i5 and a 950... and it would also drive away intels sales in low end cpu's for gamers.

That strat would actually be a major design win. IF they had the HBM bandwith for Igpu to have good power.

EDIT: if amd has 20% devices on market now, and majority of PC gamers are in the R7 type category and below, then it would eat its own 16% and also another 50% from other vendors. essentially taking 16% money away from Radeon group and massively boosting money into APU group by 66%. total net gain for amd is 50% boost and then APU group can just cut a check to high end graphics from APU profit to maintain that amd has excellent high end gpu for marketing reasons, despite its lack of making money.

But you are assuming that it would perform well enough and that Intels next APU wont have the same level of performance. If it performed better than Intels equivalent APU it would also be priced higher.

It is much like some of the older APUS. While some from AMD were good in some cases it was almost the same cost to buy and i3 and a better discrete GPU that outperformed the top end AMD APU by 50% in some cases.

In order for AMD to grab that 50% it would require their next APU to be vastly better than Intels and Intel has been putting up a very good fight, even when people laughed at their on die GPUs due to Intels history of havng very basic iGPUs on their chipsets.

 

[Vogner16]

The new nodes give AMD enough space for 16CUs. They keys are memory bandwidth and dGPU business. Will AMD include expensive HBM on mainstream APUs or only top APUs (like the HPC APU will get it)? And will AMD eat its own dGPU market with powerful consumer APUs or will prefer people to purchase APU+dGPU instead?

wouldn't amd introducing high end APU's to eat mid level gpu market come off as a major design win? this would take away from products such as R7 series gpu's but also take away from all of nvidias sales in that sector as an APU as fast as a 950 would be cheaper than an i5 and a 950... and it would also drive away intels sales in low end cpu's for gamers.

That strat would actually be a major design win. IF they had the HBM bandwith for Igpu to have good power.

EDIT: if amd has 20% devices on market now, and majority of PC gamers are in the R7 type category and below, then it would eat its own 16% and also another 50% from other vendors. essentially taking 16% money away from Radeon group and massively boosting money into APU group by 66%. total net gain for amd is 50% boost and then APU group can just cut a check to high end graphics from APU profit to maintain that amd has excellent high end gpu for marketing reasons, despite its lack of making money.

But you are assuming that it would perform well enough and that Intels next APU wont have the same level of performance. If it performed better than Intels equivalent APU it would also be priced higher.

It is much like some of the older APUS. While some from AMD were good in some cases it was almost the same cost to buy and i3 and a better discrete GPU that outperformed the top end AMD APU by 50% in some cases.

In order for AMD to grab that 50% it would require their next APU to be vastly better than Intels and Intel has been putting up a very good fight, even when people laughed at their on die GPUs due to Intels history of havng very basic iGPUs on their chipsets.

that's the point of juans message. 16 CU's means 4 CU's for zen cpu cores, and 12 CU's for graphics cores running on GCN 2.0 uarch that's (64 x 12 = 768SPU's running on ddr4) giving you ~~ xbone performance with better cpu.

guess it depends on price as always, but I would take a bet that an apu from amd without BD style cores would sell quite well. Iris pro broadwell with that L4 cashe is just barely just as fast as amd's kaveri apu that came out last year.

add to the fact that it costs almost $200 more than a kaveri chip and ya.

My personal opinion to why kaveri sold poor? the cpu was weak and people couldn't use it as an upgrade path. marketing is great and people think amd apu's for low end gaming (as I see from the thousands of posts on this forum), but they often stray away from them when we advise that they are not a good purchase if you plan to use a discrete gpu.

if you have the promice of zen cpu performance with the above theorized gpu space, that leaves us for some serious potential power to upgrade later and have good gaming even on integrated.

 

[salgado18]

that's the point of juans message. 16 CU's means 4 CU's for zen cpu cores, and 12 CU's for graphics cores running on GCN 2.0 uarch that's (64 x 12 = 768SPU's running on ddr4) giving you ~~ xbone performance with better cpu.

guess it depends on price as always, but I would take a bet that an apu from amd without BD style cores would sell quite well. Iris pro broadwell with that L4 cashe is just barely just as fast as amd's kaveri apu that came out last year.

add to the fact that it costs almost $200 more than a kaveri chip and ya.

My personal opinion to why kaveri sold poor? the cpu was weak and people couldn't use it as an upgrade path. marketing is great and people think amd apu's for low end gaming (as I see from the thousands of posts on this forum), but they often stray away from them when we advise that they are not a good purchase if you plan to use a discrete gpu.

if you have the promice of zen cpu performance with the above theorized gpu space, that leaves us for some serious potential power to upgrade later and have good gaming even on integrated.

That potential is good only in smaller form factors, I believe. PCs will continue to have discrete CPU and GPU, as there is no constraint on power, space or heat, and upgrades are expected. Now take these APUs to notebooks or Steam Machines or the like, and suddenly there is a lot of potential for gaming and work. Think an Intel NUC with a top APU.

 

[jdwii]

that's the point of juans message. 16 CU's means 4 CU's for zen cpu cores, and 12 CU's for graphics cores running on GCN 2.0 uarch that's (64 x 12 = 768SPU's running on ddr4) giving you ~~ xbone performance with better cpu.

guess it depends on price as always, but I would take a bet that an apu from amd without BD style cores would sell quite well. Iris pro broadwell with that L4 cashe is just barely just as fast as amd's kaveri apu that came out last year.

add to the fact that it costs almost $200 more than a kaveri chip and ya.

My personal opinion to why kaveri sold poor? the cpu was weak and people couldn't use it as an upgrade path. marketing is great and people think amd apu's for low end gaming (as I see from the thousands of posts on this forum), but they often stray away from them when we advise that they are not a good purchase if you plan to use a discrete gpu.

if you have the promice of zen cpu performance with the above theorized gpu space, that leaves us for some serious potential power to upgrade later and have good gaming even on integrated.

That potential is good only in smaller form factors, I believe. PCs will continue to have discrete CPU and GPU, as there is no constraint on power, space or heat, and upgrades are expected. Now take these APUs to notebooks or Steam Machines or the like, and suddenly there is a lot of potential for gaming and work. Think an Intel NUC with a top APU.

Gaming PC's will for now until enough software and games is written to take advantage of HSA,

 

[salgado18]

That potential is good only in smaller form factors, I believe. PCs will continue to have discrete CPU and GPU, as there is no constraint on power, space or heat, and upgrades are expected. Now take these APUs to notebooks or Steam Machines or the like, and suddenly there is a lot of potential for gaming and work. Think an Intel NUC with a top APU.

Gaming PC's will for now until enough software and games is written to take advantage of HSA,

No APU can ever have the 4096 streaming processors that Fury X has. So, anyone wanting to go above mainstream will need a dedicated GPU. HSA will bring many improvements, but in graphics front, nothing beats discrete.

 

[Cazalan]

No APU can ever have the 4096 streaming processors that Fury X has. So, anyone wanting to go above mainstream will need a dedicated GPU. HSA will bring many improvements, but in graphics front, nothing beats discrete.

Sure they can. It's just a matter of when. People forget that Bulldozer was only 1.2B transistors. The Fury-X is 8.9B transistors. Add the south bridge and you're still under 10.5B transistors. There are 20nm parts shipping with that many transistors today (Oracle). They're just expensive. It would probably take the 10nm node for an APU of that size to be for mainstream consumers.

Remember the XBone APU is 5B transistors and it taped out back in 2012. The whole console now sells for $299 with a game. It's not unreasonable to think that 5 years later a part with 2x the transistors could be made for a reasonable price. ($300)

 

[cdrkf]

No APU can ever have the 4096 streaming processors that Fury X has. So, anyone wanting to go above mainstream will need a dedicated GPU. HSA will bring many improvements, but in graphics front, nothing beats discrete.

Sure they can. It's just a matter of when. People forget that Bulldozer was only 1.2B transistors. The Fury-X is 8.9B transistors. Add the south bridge and you're still under 10.5B transistors. There are 20nm parts shipping with that many transistors today (Oracle). They're just expensive. It would probably take the 10nm node for an APU of that size to be for mainstream consumers.

Remember the XBone APU is 5B transistors and it taped out back in 2012. The whole console now sells for $299 with a game. It's not unreasonable to think that 5 years later a part with 2x the transistors could be made for a reasonable price. ($300)

Of course the counter argument is always "but a discreet gpu will have *all the transistors* and therefore be more powerful*.

The thing is though, if that thinking held up then maths-co-processors would still be a thing (instead of an integrated fpu), north and southbridge chips would all be independent and so on.

Eventually, these sub components become so inconsequentially small (or the power consumption advantage / reduction in latency becomes so attractive), that there becomes no logical reason *not to integrate them*.

Currently cpu's are much smaller than gpus, but they are still large enough to be noticeable. However, I think in a few more process generations a potent cpu capable of everything an end user could need will be small enough compared to the overall transistor budget to make APU designs the only logical option. We're still a few years off that point of course.

 

[Vogner16]

Of course the counter argument is always "but a discreet gpu will have *all the transistors* and therefore be more powerful*.

The thing is though, if that thinking held up then maths-co-processors would still be a thing (instead of an integrated fpu), north and southbridge chips would all be independent and so on.

Eventually, these sub components become so inconsequentially small (or the power consumption advantage / reduction in latency becomes so attractive), that there becomes no logical reason *not to integrate them*.

Currently cpu's are much smaller than gpus, but they are still large enough to be noticeable. However, I think in a few more process generations a potent cpu capable of everything an end user could need will be small enough compared to the overall transistor budget to make APU designs the only logical option. We're still a few years off that point of course.

I believe that by 2025 almost all pc's even high end gaming rigs will run on APU's. there simply is no reason to not move to an SoC design. lower power to performance ratio and on smaller nodes allow for more transisiters per cubic inch and will be capable of fitting more and more CU's on the same die as well as SB and NB wireless BT etc... tell somebody in 2003 that in 10 years amd would have a CPU with graphics and northbridge on the same chip and be faster than your fastest components while still using a quarter of the power and they would laugh. APU's are a massive leap into the future, they just need to keep cramming more onto the die or make the die larger.

EDIT: if amd can survive to 2025 I see a potential problem for NVidia. notice that they have started spending a lot of time on tegra? notice intel has been doing 5% gains to cpu's and 10% gains to gpu's? notice the first integraged L4 cashe? notice intel moved to quad channel ram when dual channel made no difference to cpu performance? these are all steps toward a apu future. all signs point to APU.

 

[jimmysmitty]

Of course the counter argument is always "but a discreet gpu will have *all the transistors* and therefore be more powerful*.

The thing is though, if that thinking held up then maths-co-processors would still be a thing (instead of an integrated fpu), north and southbridge chips would all be independent and so on.

Eventually, these sub components become so inconsequentially small (or the power consumption advantage / reduction in latency becomes so attractive), that there becomes no logical reason *not to integrate them*.

Currently cpu's are much smaller than gpus, but they are still large enough to be noticeable. However, I think in a few more process generations a potent cpu capable of everything an end user could need will be small enough compared to the overall transistor budget to make APU designs the only logical option. We're still a few years off that point of course.

I believe that by 2025 almost all pc's even high end gaming rigs will run on APU's. there simply is no reason to not move to an SoC design. lower power to performance ratio and on smaller nodes allow for more transisiters per cubic inch and will be capable of fitting more and more CU's on the same die as well as SB and NB wireless BT etc... tell somebody in 2003 that in 10 years amd would have a CPU with graphics and northbridge on the same chip and be faster than your fastest components while still using a quarter of the power and they would laugh. APU's are a massive leap into the future, they just need to keep cramming more onto the die or make the die larger.

EDIT: if amd can survive to 2025 I see a potential problem for NVidia. notice that they have started spending a lot of time on tegra? notice intel has been doing 5% gains to cpu's and 10% gains to gpu's? notice the first integraged L4 cashe? notice intel moved to quad channel ram when dual channel made no difference to cpu performance? these are all steps toward a apu future. all signs point to APU.

There is plenty of reason to not have a SOC only. While it will work for the majority, enthusiasts and businesses will need non SoC based options. There is no APU that will be able to do what a dGPU can do in gaming or CAD.

 

[jdwii]

That potential is good only in smaller form factors, I believe. PCs will continue to have discrete CPU and GPU, as there is no constraint on power, space or heat, and upgrades are expected. Now take these APUs to notebooks or Steam Machines or the like, and suddenly there is a lot of potential for gaming and work. Think an Intel NUC with a top APU.

Gaming PC's will for now until enough software and games is written to take advantage of HSA,

No APU can ever have the 4096 streaming processors that Fury X has. So, anyone wanting to go above mainstream will need a dedicated GPU. HSA will bring many improvements, but in graphics front, nothing beats discrete.

We already settled this argument way back in the other thread with juan, one day the CPU will be nothing more then 10% of the die of the APU no one is going to fab a GPU just for the extra 10% of space. Not to mention with the GPU and CPU in the same socket you will no longer have to send information all the way to the PCI-E slot so again when software and the hardware is ready i can easily see dGPU's going away but not for some time.