AMD Confirms RX 480 At $199 USD, Other APU & Polaris Announcements

[koyoot]

I don't know what it has to save. So far Polaris is selling so well that AMD is not able to make enough GPUs. Read the quote in 4th post from yours.

And Lisa Su said that Vega is coming within 2 upcoming quarters. Which means: small availability until the end if the year, and huge availability after.

Does Apple need more than 100 000 GPUs for only one computer line, and nieche one?

80% of computers sold in a lineup with dual Polaris 10 as first two GPU tiers, and Vega 10 as third tier would account for 20%, don't you think?

 

[Zarniwoop]

As I posted on another thread today, it could be that Apple really is looking forward to Zen that will be used on any Mac with dGPU in future. Why? Because of HSA. There's no mention on Intel's' (Corrected: Core -family) roadmap about support for Unified Memory Access... except only between their iGPU and CPU it is and will be possible. But not with any external co-processor. So...

If Apple ever is going to use openCL 2.0 and later, they need OS and hardware support for Shared Virtual Memory. Only Zen can open it for Apple (with dGPUs) and let them still keep the x86 code base.

So, based on this speculation, Apple might continue with Intel for many years on low power laptops. iGPU only. But the rest is going to be AMDs territory (CPU wise). And ARM of course.

 

[koyoot]

It is possible with OmniPath Coherent Fabric. You need Coherent Fabric, at least for now, for support of HSA.

 

[Zarniwoop]

It is possible with OmniPath Coherent Fabric. You need Coherent Fabric, at least for now, for support of HSA.

Ah, but then we're not talking about Intel Core processors, but Phi instead. But true, I used "Intel" too vaguely...

So let me correct that...

What are changes that we will see Phi on a Mac in near future?

Added later:
Hmm... it seems that Xeon v4 should be compatible with this tech... so maybe this could be used in Mac Pro.
[doublepost=1470955861][/doublepost]So, all above combined, HSA is possible with:

• Any Intel iGPU solution
• Zen
• Xeon E5 v4 (on theory)

We could have Intel Macbooks and Mac Pro along with AMD Mac Mini (or new VR Center) and iMac.

 

[koyoot]

AMD will bring similar coherent fabric that can be used with any CPU and any GPU out there.

This coherent fabric can work as a switch and connection between multiple GPUs and... SSD's.

Imagine this:
8 Core Intel CPU, Coherent fabric that connects both GPUs to CPU, using dual 8 PCIe connection, so the GPU appear as single cluster for applications, and then you have dual SSD - one for system, one for data for the GPUs.
If you don't need SSD for GPUs, you can always build RAID0 from them.
All memory: RAM, HBM2, SSD would appear as one huge coherent cluster.

It is a mix of technology, that can be added to Mac Pro 7.1.

Of course I am wrong. This is only my understanding of how it Could be implemented and I can miss something from big picture.

About OpenCL 2.0. It would be ironic that Apple would use it for their pro machine, which hardware does not support it .
OpenCL 2.0 is possible only on GCN1.1 and further. Tahiti and Pitcairn are completely excluded from this.

 

[xsmi123]

I predict if No One is correct in the Mac Pro Rising thread, they will need ALL of those cards and then some.

 

[Zarniwoop]

AMD will bring similar coherent fabric that can be used with any CPU and any GPU out there.

This coherent fabric can work as a switch and connection between multiple GPUs and... SSD's.

Imagine this:
8 Core Intel CPU, Coherent fabric that connects both GPUs to CPU, using dual 8 PCIe connection, so the GPU appear as single cluster for applications, and then you have dual SSD - one for system, one for data for the GPUs.
If you don't need SSD for GPUs, you can always build RAID0 from them.
All memory: RAM, HBM2, SSD would appear as one huge coherent cluster.

It is a mix of technology, that can be added to Mac Pro 7.1.

Of course I am wrong. This is only my understanding of how it Could be implemented and I can miss something from big picture.

About OpenCL 2.0. It would be ironic that Apple would use it for their pro machine, which hardware does not support it .
OpenCL 2.0 is possible only on GCN1.1 and further. Tahiti and Pitcairn are completely excluded from this.

I think what you missed there is the Unified Memory Access... your solution doesn't answer the biggest reason for HSA to be: Shared Virtual Memory with unified memory pointers. Intel is not offering a solution for this with OmniPath Coherent Fabric, is it?

 

[koyoot]

I think what you missed there is the Unified Memory Access... this solution doesn't answer the biggest reason for HSA to be: Shared Virtual Memory with unified memory pointers. Intel is not offering a solution for this with OmniPath Coherent Fabric, is it?

I don't know.

I predict if No One is correct in the Mac Pro Rising thread, they will need ALL of those cards and then some.

He/she is not correct.

 

[Zarniwoop]

About OpenCL 2.0. It would be ironic that Apple would use it for their pro machine, which hardware does not support it .
OpenCL 2.0 is possible only on GCN1.1 and further. Tahiti and Pitcairn are completely excluded from this.

Yes, But I suppose that if Apple adds openCL 2.x to future OS, even M395 in iMac is not able to fully run it... thanks to Intel's' limitations. It can run it, but wont benefit the speed boost of Unified Memory.

So, for HSA compliance, Apple needs both software and hardware support for it.

 

[xsmi123]

Hasn't been proven not to be yet.

 

[Zarniwoop]

Even with Intel's iGPU, it can be a HSA-like solution between iGPU and CPU only. If Apple wants to add DSP (and I'm sure they want, for Siri and Logic Pro for instance), it cannot work in HSA scope. So, it would create an overhead (still DSP with an overhead is better than no DSP).

Hence, the only true solution (for consumer/prosumer market) to be fully HSA complaint is AMD APU. And thanks to AMD's willingness to make custom APU's, it can be very tempting for Apple... adding all DSP/Special IO-chips/ISP..etc on one SoC. IF Zen is any good. I hope it is... otherwise there wont be much new in Mac world for following years. Except design and thinness and edelweiss gold...

HSA is a way to make computers much more efficient that they are today. Similar to Metal, but will make an additional step toward efficiency. Could be a HSA driver under Metal and then there can be Happily ever after with Metal and openCL.

 

[AidenShaw]

I'm thinking Apple will have to use Polaris 10 in the new rMBP instead of Polaris 11, or it will be a wash.

Or Apple can realize that going all ATI was a colossal mistake, swallow their pride (and fire some "innovative" managers), and put GTX 1080m in the new laptops....]

 

[Stacc]

I don't know what it has to save. So far Polaris is selling so well that AMD is not able to make enough GPUs. Read the quote in 4th post from yours.

Nvidia had the same problem. Both AMD and Nvidia have been supply constrained. We have no reputable sources to indicate how well any new card is selling.

And Lisa Su said that Vega is coming within 2 upcoming quarters. Which means: small availability until the end if the year, and huge availability after.

AMD earlier this year said vega was not coming until 2017. Coming out within 2 quarters basically means its coming out no earlier than 6 months from now.

Or Apple can realize that going all ATI was a colossal mistake, swallow their pride (and fire some "innovative" managers), and put GTX 1080m in the new laptops....]

I agree with this. Apple could release a mac pro now with dual GP104 chips with roughly a combined 14-16 TFLOPS of computer performance. The best AMD has to offer is Polaris 10 with a combined ~10 TFLOPS. They could wait for Vega but it may not be any faster than GP104 and it still probably won't be able to catch up to GP102.

 

[AidenShaw]

I agree with this. Apple could release a mac pro now with dual GP104 chips with roughly a combined 14-16 TFLOPS of computer performance. The best AMD has to offer is Polaris 10 with a combined ~10 TFLOPS. They could wait for Vega but it may not be any faster than GP104 and it still probably won't be able to catch up to GP102.

But the cylinder has castrated the Mac Pro with a small form factor that nobody asked for (well, maybe it was good for the MacMini folks who wanted something more), a power supply with the wattage that Dell and HP put in systems with integrated GPUs, and a thermal system that can't even deal with the tiny power supply.

From line engineer to product management to VP - Apple should fire everyone involved in bringing out the MP6.1.

You too, Tim. You've helped Apple sacrifice its soul on a rose gold altar. It's time to go.

 

[koyoot]

Nvidia had the same problem. Both AMD and Nvidia have been supply constrained. We have no reputable sources to indicate how well any new card is selling.



AMD earlier this year said vega was not coming until 2017. Coming out within 2 quarters basically means its coming out no earlier than 6 months from now.

Nvidia does not have this problem. They have extremely limited capacity in TSMC Fabs, because despite that TSMC has currently 60 thousand wafers per month capacity in their fabs and in Q4 it will be 80 thousand, they have 7 clients to feed.
And this is proof for this: http://www.tweaktown.com/news/53375/nvidias-next-gen-gpus-made-14nm-samsung/index.html

They are changing TSMC to Samsung 14 LPP FinFet, because of limited capacity in TSMC fabs. And yes, this is next gen. GPUs - Volta. I will post this in Pascal thread as well.

AMD does not have this problem with capacity. They have GLoFo which has 40 thousand wafers per month, and 60 thousand in Q4. And there are only two clients: AMD and Qualcomm. To put this into perspective. AMD on launch have had 180 thousand RX 480 GPUs just from one Fab. And there were 3 fabs that were in full production for previous 6 months(before release). Realistic expectations say that AMD sold 600 thousand GPUs around the world since the start.

Expectations for Nvidia are lower, much lower. Around 200 thousand GPUs(from all 3 price tiers). Because of the lack of capacity. People who were not able to buy for example GTX 1060 just went with RX 480 instead. And at this point funniest part is that currently there are massive rebates on Nvidia hardware which is not able to sell. It is seen in UK, USA, Canada, Australia. AMD mostly you do not see in stock. And because of this demand prices went up.

And one last bit. Because of the demand AMD have had to hire Samsung to produce 14 nm GPUs. This strategy looks like worked well for AMD. They clean WSA, with those GPUs, they earn massive amounts of money for running their business, they even increased their R&D expenditures lately. And now we are facing Zen and Vega. If Zen is as good as the rumors are telling we do not have a killer, but very competitive product for Intel offerings.

And one more thing, both Zen and Vega are coming out this year. Why? AMD is preparing their own gaming/compute platform based on Vega 10 and Zen CPU: Project Quantum.

 

[ActionableMango]

They clean WSA

What is WSA?

 

[koyoot]

What is WSA?

Wafer Supply Agreement. AMD had to pay GloFo for wafers that they even didn't wanted. Polaris was pipe cleaner for this.

 

[ManuelGomes]

Come on, give it a rest already

http://wccftech.com/nvidia-volta-gv110-gv104-gv102-gpu/

Pascal full lineup isn't even out yet!!
[doublepost=1471040285][/doublepost]I'm starting to see the HSA solution coming to the nMP and the AMD choice might have made sense.
However, an all AMD nMP was not in my plans.
Never had an AMD product in any of my rigs.
I could live with the FirePros but Zen as well? It should be a good CPU but I prefer them Xeons.
And better yet if a couple of Pascals go in there as well but that seems unlikely.

 

[koyoot]

There will not be Zen based Mac Pro for simple reason. First Zen CPUs available will have 95W TDP with 8 core version(Skylake i7-6700/6700K direct competitors).

Those CPUs are not meant for Mac Pro. In 2017 there will be Server versions with up to 32 cores.

At least: for some time there will not be Zen based Mac Pro. In future, long enough that we cannot see it? Possibly. But rather - maybe.

 

[ManuelGomes]

Depends on if the nMP will really come in a short time or if it will be delayed till next year, again!
This might be what Apple is waiting for, hope not.

 

[Stacc]

Nvidia does not have this problem. They have extremely limited capacity in TSMC Fabs, because despite that TSMC has currently 60 thousand wafers per month capacity in their fabs and in Q4 it will be 80 thousand, they have 7 clients to feed.
And this is proof for this: http://www.tweaktown.com/news/53375/nvidias-next-gen-gpus-made-14nm-samsung/index.html

Rumor has it Nvidia won the gpu in nintendos next console. My bet is this is what Samsung is manufacturing since cost is the priority not efficiency.

AMD does not have this problem with capacity. They have GLoFo which has 40 thousand wafers per month, and 60 thousand in Q4. And there are only two clients: AMD and Qualcomm. To put this into perspective. AMD on launch have had 180 thousand RX 480 GPUs just from one Fab. And there were 3 fabs that were in full production for previous 6 months(before release). Realistic expectations say that AMD sold 600 thousand GPUs around the world since the start.

Expectations for Nvidia are lower, much lower. Around 200 thousand GPUs(from all 3 price tiers). Because of the lack of capacity. People who were not able to buy for example GTX 1060 just went with RX 480 instead. And at this point funniest part is that currently there are massive rebates on Nvidia hardware which is not able to sell. It is seen in UK, USA, Canada, Australia. AMD mostly you do not see in stock. And because of this demand prices went up.

And one last bit. Because of the demand AMD have had to hire Samsung to produce 14 nm GPUs. This strategy looks like worked well for AMD. They clean WSA, with those GPUs, they earn massive amounts of money for running their business, they even increased their R&D expenditures lately. And now we are facing Zen and Vega. If Zen is as good as the rumors are telling we do not have a killer, but very competitive product for Intel offerings.

And one more thing, both Zen and Vega are coming out this year. Why? AMD is preparing their own gaming/compute platform based on Vega 10 and Zen CPU: Project Quantum.

Again, you have no source on these numbers. The only thing we do know is that nvidia just had their largest revenue for a quarter ever.

Project quantum is nothing more than a custom pc to show at trade shows that took advantage of the small form factor of Fiji. It was embarrassing that they had to put an Intel processor in it to get decent performance. amd is not about to become some sort of system vendor.

Also I would bet good money that we don't see vega or zen until next year. Amd has already stated this about vega and rumors point to zen being pushed into early 2017.

 

[koyoot]

Rumor has it Nvidia won the gpu in nintendos next console. My bet is this is what Samsung is manufacturing since cost is the priority not efficiency.

Again, you have no source on these numbers. The only thing we do know is that nvidia just had their largest revenue for a quarter ever.

Project quantum is nothing more than a custom pc to show at trade shows that took advantage of the small form factor of Fiji. It was embarrassing that they had to put an Intel processor in it to get decent performance. amd is not about to become some sort of system vendor.

Also I would bet good money that we don't see vega or zen until next year. Amd has already stated this about vega and rumors point to zen being pushed into early 2017.

Yes, I do have sources. You can find them yourself. People who worked in the GloFo factories talked about this. People from retail line talk about this. When do you guys will learn that I am not pulling this information from my ass?

And yes, AMD will become system producer, because if you did not read from my previous posts about where the industry is going, it is in the device, not parts, business. Intel will do exactly the same thing, they already do with their NUC's.

Zen Server version is going to be 2017. Mainstream market however, will get limited quantity Zen CPUs this year. One reason for this is the launch of... Project Quantum. And Zen most likely will be launched/presented coincidentally with Battlefield 1 premiere. Will there be Vega 10 launched? Nobody knows. Will it be presented there? Most likely.

Design for Vega is already done, the silicon is undergoing currently qualification, so you need to add 3 months from ending the qualification to the chip being widely available. I presume very limited quantity of this GPUs on customer market, and 90% of orders going to OEMs designing systems like... Project Quantum, or Mac Pro. The design work for Vega has ended I think at the beginning of July, so where it puts the timeframe of qualification? October. And for which month information floated that it might be launched? October. And when is premiere of BF1? October.

 

[tuxon86]

Yes, I do have sources. You can find them yourself. People who worked in the GloFo factories talked about this. People from retail line talk about this. When do you guys will learn that I am not pulling this information from my ass?

And yes, AMD will become system producer, because if you did not read from my previous posts about where the industry is going, it is in the device, not parts, business. Intel will do exactly the same thing, they already do with their NUC's.

Zen Server version is going to be 2017. Mainstream market however, will get limited quantity Zen CPUs this year. One reason for this is the launch of... Project Quantum. And Zen most likely will be launched/presented coincidentally with Battlefield 1 premiere. Will there be Vega 10 launched? Nobody knows. Will it be presented there? Most likely.

Design for Vega is already done, the silicon is undergoing currently qualification, so you need to add 3 months from ending the qualification to the chip being widely available. I presume very limited quantity of this GPUs on customer market, and 90% of orders going to OEMs designing systems like... Project Quantum, or Mac Pro. The design work for Vega has ended I think at the beginning of July, so where it puts the timeframe of qualification? October. And for which month information floated that it might be launched? October. And when is premiere of BF1? October.

Except this isn't how this work. You made the claim, you've to provide the source. Most of the sources that you use are anonymous blog post on forums, not really worthy of the "source" attribute in the first place.

We get it, you love AMD and can only find fault with NVidia and Intel, but don't let your own prejudice cloud the reality of the market.

 

[koyoot]

Except this isn't how this work. You made the claim, you've to provide the source. Most of the sources that you use are anonymous blog post on forums, not really worthy of the "source" attribute in the first place.

We get it, you love AMD and can only find fault with NVidia and Intel, but don't let your own prejudice cloud the reality of the market.

Stop accusing people when they tell things which offend your preferences.

All of the sources are easy to find. And they are reliable. More reliable than for example WCCFtech. People like Ashraf Eassa. People like Richard Wargo who worked for Intel. People who sell computer hardware over the internet and have the sales numbers. People who worked in factories which produced the silicon chips. They are on number of forums over the internet, its your thing to find them.

 

[Stacc]

Stop accusing people when they tell things which offend your preferences.

All of the sources are easy to find. And they are reliable. More reliable than for example WCCFtech. People like Ashraf Eassa. People like Richard Wargo who worked for Intel. People who sell computer hardware over the internet and have the sales numbers. People who worked in factories which produced the silicon chips. They are on number of forums over the internet, its your thing to find them.

You are half way there. Now link us to the content where they make those claims. Random message board posts do not count. This is how citing your sources work.

Don't get too excited on vega. Amd themselves have said its coming in 2017. You have a history of being too overly optimistic on amd hardware launches.