AMD CPUs, SoC Rumors and Speculations Temp. thread 2

[8350rocks]

Well I doubt it's ARM. I wouldn't be surprised if the Xbox1/Xbox1S actually use the same SoC unless somebody has confirmation of something else. Just a faster clock due to more mature process and driver included 4K video support.

Xbox1S is a new 16nm SoC from AMD made at TSMC.

http://www.eurogamer.net/articles/digitalfoundry-2016-inside-xbox-one-s-tech-interview

Largely just a die shrink + 4k capabilities. Uses much less power as the CPU clocks were kept the same. GPU clocks got a slight boost (853 MHz to 914 MHz).

Strange that they went with TSMCs 16nm and not GloFlos 14nm. Could that be a sign that yields are not high enough or that GloFlo can't handle that big of an order on top of the Radeon and Zen they already have?

Who produced the original chip? If TSMC, it could just be a prior contractual agreement.

This, it was TSMC, because of the GPU requirements being what they were, they used their GPU foundry to produce the whole chip. Especially consider that TSMC had ramped production of Jaguar already.

 

[gamerk316]

Well I doubt it's ARM. I wouldn't be surprised if the Xbox1/Xbox1S actually use the same SoC unless somebody has confirmation of something else. Just a faster clock due to more mature process and driver included 4K video support.

Xbox1S is a new 16nm SoC from AMD made at TSMC.

http://www.eurogamer.net/articles/digitalfoundry-2016-inside-xbox-one-s-tech-interview

Largely just a die shrink + 4k capabilities. Uses much less power as the CPU clocks were kept the same. GPU clocks got a slight boost (853 MHz to 914 MHz).

Strange that they went with TSMCs 16nm and not GloFlos 14nm. Could that be a sign that yields are not high enough or that GloFlo can't handle that big of an order on top of the Radeon and Zen they already have?

Who produced the original chip? If TSMC, it could just be a prior contractual agreement.

Hard to say but I would assume agreements would stop at a certain process because I am sure AMD would want the best process for the new APU.

I doubt they'd want to change process after the fact though; remember, AMD can't break ANYTHING in the process.

 

[jimmysmitty]

Well I doubt it's ARM. I wouldn't be surprised if the Xbox1/Xbox1S actually use the same SoC unless somebody has confirmation of something else. Just a faster clock due to more mature process and driver included 4K video support.

Xbox1S is a new 16nm SoC from AMD made at TSMC.

http://www.eurogamer.net/articles/digitalfoundry-2016-inside-xbox-one-s-tech-interview

Largely just a die shrink + 4k capabilities. Uses much less power as the CPU clocks were kept the same. GPU clocks got a slight boost (853 MHz to 914 MHz).

Strange that they went with TSMCs 16nm and not GloFlos 14nm. Could that be a sign that yields are not high enough or that GloFlo can't handle that big of an order on top of the Radeon and Zen they already have?

Who produced the original chip? If TSMC, it could just be a prior contractual agreement.

This, it was TSMC, because of the GPU requirements being what they were, they used their GPU foundry to produce the whole chip. Especially consider that TSMC had ramped production of Jaguar already.

Then why did they go to GloFlo for Polaris? It is pretty obvious that TSMC would have vastly more and better experience with GPUs than GloFlo.

Well I doubt it's ARM. I wouldn't be surprised if the Xbox1/Xbox1S actually use the same SoC unless somebody has confirmation of something else. Just a faster clock due to more mature process and driver included 4K video support.

Xbox1S is a new 16nm SoC from AMD made at TSMC.

http://www.eurogamer.net/articles/digitalfoundry-2016-inside-xbox-one-s-tech-interview

Largely just a die shrink + 4k capabilities. Uses much less power as the CPU clocks were kept the same. GPU clocks got a slight boost (853 MHz to 914 MHz).

Strange that they went with TSMCs 16nm and not GloFlos 14nm. Could that be a sign that yields are not high enough or that GloFlo can't handle that big of an order on top of the Radeon and Zen they already have?

Who produced the original chip? If TSMC, it could just be a prior contractual agreement.

Hard to say but I would assume agreements would stop at a certain process because I am sure AMD would want the best process for the new APU.

I doubt they'd want to change process after the fact though; remember, AMD can't break ANYTHING in the process.

Maybe, maybe not. I just would assume that they would not write an agreement to include an untested process. Remember the debacle at 20nm for TSMC? Neither AMD nor NVidia went to it due to issues and thus we were stuck on 28nm, which hurt AMD more than it did NVidia since Fury was supposed to be 20nm.

I would just assume AMD would be looking for the best process for either product and not be assuming TSMC or GloFlo would have the better process.

 

[Cazalan]

Then why did they go to GloFlo for Polaris? It is pretty obvious that TSMC would have vastly more and better experience with GPUs than GloFlo.

With 32nm orders winding down they needed more SKU at GF to meet their wafer purchase agreements.

 

[jimmysmitty]

Then why did they go to GloFlo for Polaris? It is pretty obvious that TSMC would have vastly more and better experience with GPUs than GloFlo.

With 32nm orders winding down they needed more SKU at GF to meet their wafer purchase agreements.

Which might not be favorable in the long run as GFs 14nm doesn't seem as power efficient as TSMCs 16nm.

That may be due to the uArch but it makes me worry about Zen because if it is the process that is power hungry then Zen will not fare as well in the server environment.

 

[ComputerSecurityGuy]

Depends. TSMC just seems to have an exceptionally efficient 16nm process. It's more efficient than Samsung's 14nm as well- look at the iPhone. It might have other issues we don't know about though. Clockspeed would be my guess. I'm not a huge fab expert though.

 

[jimmysmitty]

Depends. TSMC just seems to have an exceptionally efficient 16nm process. It's more efficient than Samsung's 14nm as well- look at the iPhone. It might have other issues we don't know about though. Clockspeed would be my guess. I'm not a huge fab expert though.

The results in the iPhone were interesting but not on this level. Look at the RX 480. It is higher than the GTX 1070 but a bit lower than the GTX 1080 yet doesn't equal the performance of either GPU nor does it have a clock speed as high as either GPU.

This tells me that in higher power devices GFs 14nm LPP is not as efficient as it could be and makes me wonder if AMD wen with TSMCs 16nm could Polaris have been a much better GPU?

And how will Zen fare? It is a 8 core high powered CPU and if it means to compete in the server realm, power is vastly more important than on the consumer end. Servers are designed around a TDP and right now Intel has a massive advantage on TDP that AMD needs to be able to compete with.

 

[Cazalan]

Then why did they go to GloFlo for Polaris? It is pretty obvious that TSMC would have vastly more and better experience with GPUs than GloFlo.

With 32nm orders winding down they needed more SKU at GF to meet their wafer purchase agreements.

Which might not be favorable in the long run as GFs 14nm doesn't seem as power efficient as TSMCs 16nm.

That may be due to the uArch but it makes me worry about Zen because if it is the process that is power hungry then Zen will not fare as well in the server environment.

Really can't determine efficiency of the process when comparing Pascal to Polaris. Nvidia apparently had some tricks up their sleeves when it came to rendering that AMD didn't have.

http://www.realworldtech.com/tile-based-rasterization-nvidia-gpus/

 

[jimmysmitty]

Then why did they go to GloFlo for Polaris? It is pretty obvious that TSMC would have vastly more and better experience with GPUs than GloFlo.

With 32nm orders winding down they needed more SKU at GF to meet their wafer purchase agreements.

Which might not be favorable in the long run as GFs 14nm doesn't seem as power efficient as TSMCs 16nm.

That may be due to the uArch but it makes me worry about Zen because if it is the process that is power hungry then Zen will not fare as well in the server environment.

Really can't determine efficiency of the process when comparing Pascal to Polaris. Nvidia apparently had some tricks up their sleeves when it came to rendering that AMD didn't have.

http://www.realworldtech.com/tile-based-rasterization-nvidia-gpus/

Tricks are one thing but I am looking at the clock speed difference and power numbers.

Still it is just something to draw from and try to extrapolate the next CPU from AMD. I know that GPUs and CPUs are different beasts and I hope for the best but I am not as confident as I want to be in the process tech.

 

[Embra]

AMD ZEN Engineering Sample Shows Promising Perf:

http://www.guru3d.com/news-story/amd-zen-engineering-sample-shows-promising-perf.html

 

[jimmysmitty]

AMD ZEN Engineering Sample Shows Promising Perf:

http://www.guru3d.com/news-story/amd-zen-engineering-sample-shows-promising-perf.html

Look at the water mark in the lower right hand of the image. WCCF.

I wont trust it until it is on a site that does not post every single rumor that comes out and is also not linked to that site.

ES chips are hard to tell actual performance or numbers but that clock speed looks a bit lower than you would want to be competitive.

I wonder what the power numbers are.

 

[-Fran-]

Look at the water mark in the lower right hand of the image. WCCF.

I wont trust it until it is on a site that does not post every single rumor that comes out and is also not linked to that site.

ES chips are hard to tell actual performance or numbers but that clock speed looks a bit lower than you would want to be competitive.

I wonder what the power numbers are.

This is the original link; give it a read at least, haha.

http://wccftech.com/amd-zen-es-benchmarks/

And man... That comment section... Such a hungry hell pit of despair and evilness. *shrug*

Cheers!

 

[gamerk316]

I'd be willing to accept the article as truthful, but mostly because it matches my expectations (2.8 GHz clock, Haswell i7 level performance).

 

[jimmysmitty]

Look at the water mark in the lower right hand of the image. WCCF.

I wont trust it until it is on a site that does not post every single rumor that comes out and is also not linked to that site.

ES chips are hard to tell actual performance or numbers but that clock speed looks a bit lower than you would want to be competitive.

I wonder what the power numbers are.

This is the original link; give it a read at least, haha.

http://wccftech.com/amd-zen-es-benchmarks/

And man... That comment section... Such a hungry hell pit of despair and evilness. *shrug*

Cheers!

I did read it. And I am taking it with a grain of salt the size of the mountain in the picture on the site because they have proven they do not care about posting facts.

I remember the rumor mill that was pouring from them for what became Fury X. It was a new idea per day.

As I said, until it is on a reputable site like Toms, Anand etc and not linked back to that site I wont trust it for anything more than a rumor.

 

[ComputerSecurityGuy]

Plausible but not reliable. That's my call.

 

[gamerk316]

Just did some math, and it's not pretty for AMD. At all.

For this discussion, I'm disregarding SMT effects; i7 and the 8350 have eight cores, Zen has 16. I'm also assuming Turbo clocks. So these are fairly rough estimates, but good enough for ballpark IPC estimates.

Also remember IPC in this case is relative, not absolute. Only when you compare the IPC numbers do you get something worthwhile.

Performance = (IPC * Clock) * Number of Cores

i7 4790k:

65.4 = 4.0 * 8 * IPC
IPC = ~2.04

Zen:

58 = 3.2 * 16 * IPC
58 = 6.4 * IPC
IPC = ~1.13

FX-8350:

42 = 4.2 * 8 * IPC
IPC = 1.25


Relative IPC difference between FX-8350 and Zen: -10.08%
Relative IPC difference between i7 4790k and Zen: -57.41%

Ignore Intel; the math says Zen has LOWER IPC per core in AoS then Piledriver. OK, fine, some of that is due to the fact I'm ignoring SMT performance loss, but that also holds true of the i7. Some of that is likely also due to lower core workload due to scaling. What is missing here is core loading numbers; if Zen has lower core usage due to better scaling, then my math is underestimating it by potentially a significant amount. If the core loading is similar, then Zen is simply not that efficient.

My initial take: If these numbers hold, when review sites do gaming benchmarks, AMD better pray they are all DX12.

 

[ComputerSecurityGuy]

Do we have any info on what GPU(s) they used?
To note, the i7-5960X is probably the best comparison chip to 8C/16T Zen.

 

[grsychckn]

Performance = (IPC * Clock) * Number of Cores

i7 4790k:

65.4 = 4.0 * 8 * IPC
IPC = ~2.04

Zen:

58 = 3.2 * 16 * IPC
58 = 6.4 * IPC
IPC = ~1.13

FX-8350:

42 = 4.2 * 8 * IPC
IPC = 1.25


Relative IPC difference between FX-8350 and Zen: -10.08%
Relative IPC difference between i7 4790k and Zen: -57.41%

I see what you've done there but we must keep in mind the limitations of the software to leverage multithreading. From what I've seen here it appears that AotS can really only leverage up to 6 threads on the CPU side which means your math changes somewhat. Running your numbers again using a maximum of 6 threads, we see a much different picture:


Performance = (IPC * Clock) * Number of Cores

i7 4790k:
65.4 = 4.0 * 6 * IPC
IPC = 2.73

Zen:
58 = 3.2 * 6* IPC
IPC = ~3.02

FX-8350:
42 = 4.2 * 6 * IPC
IPC = ~1.67

Relative IPC difference between FX-8350 and Zen: 80.84%
Relative IPC difference between i7 4790k and Zen: 11.85%

I agree with you that these numbers are VERY theoretical and not necessarily evident of measurable CPU performance. The AotS benchmark also doesn't appear to easily reproduce results across similar hardware. That said, it is fun to play the guessing game and if the boost clock was disabled or varying across cores, the efficiency improves.

 

[ComputerSecurityGuy]

And not only that, from some quick reading DX12 seems to cap out at about 6 cores. I hope that there is a six core Zen chip (I'm pretty sure there is supposed to be) snuck in at the $300-350 price point it could decimate the i7 lineup. Of course it would have to have sane motherboards- the Z170/X99 cost difference is really the only thing keeping everybody from recommending the i7-5820K over the i7-6700K.

 

[-Fran-]

Nitpick: DX12 doesn't "cap" at X or Y amount of cores. It goes as wide as the Dev can make it's engine use available cores. Just like DX11 can, but no one really cares, haha.

And we're all expecting between Sandy and Haswell performance for Zen, so the numbers from WTFBBQTech are in line with that, so there's that.

Cheers!

 

[jdwii]

Its still to close to call but i doubt Zen would perform worse ha ha be crazy i still expect sandy-ivy IPC(with 3.2Ghz clock speeds) with 8 true cores and HT and probably a 95 watt TDP.

 

[jdwii]

And not only that, from some quick reading DX12 seems to cap out at about 6 cores. I hope that there is a six core Zen chip (I'm pretty sure there is supposed to be) snuck in at the $300-350 price point it could decimate the i7 lineup. Of course it would have to have sane motherboards- the Z170/X99 cost difference is really the only thing keeping everybody from recommending the i7-5820K over the i7-6700K.

Not with poor IPC they can make a 100 core zen but if a 4 core I5 beats it in most things its worthless. Be like repeating what we already have with the FX line up.

 

[8350rocks]

Depends. TSMC just seems to have an exceptionally efficient 16nm process. It's more efficient than Samsung's 14nm as well- look at the iPhone. It might have other issues we don't know about though. Clockspeed would be my guess. I'm not a huge fab expert though.

The results in the iPhone were interesting but not on this level. Look at the RX 480. It is higher than the GTX 1070 but a bit lower than the GTX 1080 yet doesn't equal the performance of either GPU nor does it have a clock speed as high as either GPU.

This tells me that in higher power devices GFs 14nm LPP is not as efficient as it could be and makes me wonder if AMD wen with TSMCs 16nm could Polaris have been a much better GPU?

And how will Zen fare? It is a 8 core high powered CPU and if it means to compete in the server realm, power is vastly more important than on the consumer end. Servers are designed around a TDP and right now Intel has a massive advantage on TDP that AMD needs to be able to compete with.

Just for clarification, Polaris is made @ TSMC.

 

[jimmysmitty]

Depends. TSMC just seems to have an exceptionally efficient 16nm process. It's more efficient than Samsung's 14nm as well- look at the iPhone. It might have other issues we don't know about though. Clockspeed would be my guess. I'm not a huge fab expert though.

The results in the iPhone were interesting but not on this level. Look at the RX 480. It is higher than the GTX 1070 but a bit lower than the GTX 1080 yet doesn't equal the performance of either GPU nor does it have a clock speed as high as either GPU.

This tells me that in higher power devices GFs 14nm LPP is not as efficient as it could be and makes me wonder if AMD wen with TSMCs 16nm could Polaris have been a much better GPU?

And how will Zen fare? It is a 8 core high powered CPU and if it means to compete in the server realm, power is vastly more important than on the consumer end. Servers are designed around a TDP and right now Intel has a massive advantage on TDP that AMD needs to be able to compete with.

Just for clarification, Polaris is made @ TSMC.

So TSMC now has a 14nm FinFET design but their 16nm is superior??? Because last I checked Polaris used a 14nm design and TSMC is still running 16nm.

 

[gamerk316]

I see what you've done there but we must keep in mind the limitations of the software to leverage multithreading. From what I've seen here it appears that AotS can really only leverage up to 6 threads on the CPU side which means your math changes somewhat.

And I'm not disagreeing; that's why I noted that without per-core CPU loads I'm almost certainly underestimating Zen's IPC. Regardless, still not good for AMD, since these are the EXACT types of benchmarks people will look at when deciding to get Zen, and loosing outright to a CPU that is, what, four years old now, is not going to do AMD any favors.