What is going to happen to AMD?

[AtenRa]

And 19W kaveri throttles like mad, especially when the game is CPU heavy.

Which Kaveri APU is 512 shaders at 19W? The only mobile 512 shader Kaveri chip on the market that I can find is the FX-7600P.

A10-7300 is quad core 1.9GHz (3.2GHz turbo) with 512 shaders at 533MHz.

http://skinflint.co.uk/eu/?fs=a10-7300&in=

 

[frozentundra123456]

And 19W kaveri throttles like mad, especially when the game is CPU heavy.

Which Kaveri APU is 512 shaders at 19W? The only mobile 512 shader Kaveri chip on the market that I can find is the FX-7600P.

Yea, you are correct, at least according to AMD's own website. The only FX with 512 shaders is 35 watt TDP.

Edit: According to Notebook check, the A10-7300 that Aten mentioned is 384 shaders, not 512.

 

[AtenRa]

heh too early in the morning. Yes only the FX 35W TDP has 512 shaders my bad.

 

[Silverforce11]

https://youtu.be/9RA6t3Mx0ms?t=1h15m36s

You guys need to watch this to understand how fundamental the change with HBM is going to cause to the APU scene, for servers and for gaming.

 

[Shehriazad]

https://youtu.be/9RA6t3Mx0ms?t=1h15m36s

You guys need to watch this to understand how fundamental the change with HBM is going to cause to the APU scene, for servers and for gaming.

Seriously, though...AMD needs to go the extra mile an make socketable HBM a thing. 2 module slots are tiny enough to still fit even on an ITX board while there are also DDR4 slots. (But then again...replace DDR4 slots with moar HBM slots so you can put 32GB of HBM² on there)

That's a little off topic , though.


But on topic...we don't even know when the first HBM APUs will hit the market, right? Zen APUs not before 2017/18...Excavator APUs might not be good enough to make use of it unless they put a really strong iGPU on there...it would have to be at least twice of what it is now. (Which btw would be insane...double 250 performance? Gaming wise that would be like a 270)

 

[Silverforce11]

I think its very possible in 2017 on GF 14nm ff. Thats when that process should be mature enough to handle the complex 2.5D stacking with a large APU. Definitely a game changer for APU in all markets except for the very high end CPU + dGPU combo.

 

[Paul98]

It seems people just assume AMD will continue to be where they are at this moment product wise forever. Their current products are all old tech. With major products coming later this year and next year, along with potentially game changing things in that same time period. It's no surprise AMD is where they are at this point in time, but with some new decent products they will be fine.

 

[2is]

It seems people just assume AMD will continue to be where they are at this moment product wise forever. Their current products are all old tech. With major products coming later this year and next year, along with potentially game changing things in that same time period. It's no surprise AMD is where they are at this point in time, but with some new decent products they will be fine.

It also seems like people forget, that in the last decade plus, people in forums have pointed to AMD's next game changing new product, but the game never changed and their struggles get worse with every passing year. What is it that gives you so much hope that it will be different "this time?"

 

[PhIlLy ChEeSe]

Yeah, we need more pointless speculation threads..

Spam threads, you'll notice the site is full of them. time to move on.......

 

[Tuna-Fish]

Seriously, though...AMD needs to go the extra mile an make socketable HBM a thing.

Socketable HBM absolutely cannot be made. The whole point of HBM is reducing the pitch and capacitance of the traces between the chip and the ram, and then making a lot of them. HBM cannot live on the motherboard, they need to live on a silicon chip right next to the controller in the same package. Moving them any further away would make them not work.

 

[Silverforce11]

It also seems like people forget, that in the last decade plus, people in forums have pointed to AMD's next game changing new product, but the game never changed and their struggles get worse with every passing year. What is it that gives you so much hope that it will be different "this time?"

Bandwidth, lots of it. No longer limited by system ram.

Think about it, if you were making an APU, would you have made the iGPU portion much faster? Nope, because it would have been limited by bandwidth regardless.

You could only do that if you bypassed the normal PC setup of motherboard + DDR3 + PCIE lanes. As what happened with Iris Pro, XBone & PS4 SOCs. Dedicated edram to remove bandwidth bottlenecks, or pure GDDR5 as system ram. These can only occur in custom designs and not the mainstream PC market.

With HBM, an AMD Zen APU with 2048 GCN+ cores and 4/8GB HBM on the APU itself, will be an amazing product. That's mid-range dGPU performance right there in a small package.

 

[Yuriman]

Bandwidth, lots of it. No longer limited by system ram.

Think about it, if you were making an APU, would you have made the iGPU portion much faster? Nope, because it would have been limited by bandwidth regardless.

You could only do that if you bypassed the normal PC setup of motherboard + DDR3 + PCIE lanes. As what happened with Iris Pro, XBone & PS4 SOCs. Dedicated edram to remove bandwidth bottlenecks, or pure GDDR5 as system ram. These can only occur in custom designs and not the mainstream PC market.

With HBM, an AMD Zen APU with 2048 GCN+ cores and 4/8GB HBM on the APU itself, will be an amazing product. That's mid-range dGPU performance right there in a small package.

Unfortunately, until HSA really takes off, most of the market probably won't have any use for AMD's iGPUs. They'll still be a niche product. What percent of the population honestly wants or needs 2048+ GCN cores?

Some other challenges large-iGPU-Zen will face:
-Building a large die on a process not completely optimized for either CPU or GPU (sacrifice CPU clockspeed most likely, possibly other downsides)
-Produce it more cheaply than two smaller dies (yields are worse on larger dies)
-Reliance on HBM, which (I think?) is more expensive per capacity than GDDR5
-Requirement of replacing CPU when upgrading GPU, or vice versa

 

[2is]

Bandwidth, lots of it. No longer limited by system ram.

Think about it, if you were making an APU, would you have made the iGPU portion much faster? Nope, because it would have been limited by bandwidth regardless.

You could only do that if you bypassed the normal PC setup of motherboard + DDR3 + PCIE lanes. As what happened with Iris Pro, XBone & PS4 SOCs. Dedicated edram to remove bandwidth bottlenecks, or pure GDDR5 as system ram. These can only occur in custom designs and not the mainstream PC market.

With HBM, an AMD Zen APU with 2048 GCN+ cores and 4/8GB HBM on the APU itself, will be an amazing product. That's mid-range dGPU performance right there in a small package.

I understand the technical benefits, but how is that going to get AMD out of its hole? They came out with dual cores first, they came out with 64bit consumer CPU's first, they got PS4/XB1 design wins, Mantle, I believe they were the first to use GDDR5. They used an IMC in their processors long before Intel did. There's been plenty of milestone moments in AMD's tenure, why is HBM "the one?"

Don't get me wrong, I don't want AMD to fail, I just think people are being a bit naïve if they think this will give AMD a boost like we've never seen before. I mean I hope I'm wrong and it does so we can get some competition in both the CPU and GPU markets, but I just don't see a whole lot changing simply due to AMD being "first" to use the technology.

 

[Silverforce11]

Unfortunately, until HSA really takes off, most of the market probably won't have any use for AMD's iGPUs. They'll still be a niche product. What percent of the population honestly wants or needs 2048+ GCN cores?

High performance notebooks. It can scale down too.

mITX builds. Boutique rigs. Actual PC gaming rig if price is competitive.

My only concern is the pricing due to this rumor of HBM being expensive. If it remains cheaper to go with a CPU + dGPU combo for similar performance, it will not be a good option outside of the aforementioned niches (well, notebooks aren't exactly niche).

 

[soccerballtux]

to die, to die, to die...

 

[frozentundra123456]

High performance notebooks. It can scale down too.

mITX builds. Boutique rigs. Actual PC gaming rig if price is competitive.

My only concern is the pricing due to this rumor of HBM being expensive. If it remains cheaper to go with a CPU + dGPU combo for similar performance, it will not be a good option outside of the aforementioned niches (well, notebooks aren't exactly niche).

Most of us on these forums understand the benefit of HBM. What a lot of people forget is that PC gamers, at least in the way we think of it (not just facebook, pogo, sims, etc) are a relatively small portion of the overall PC market.

It really comes down to notebooks. In the desktop, even with HBM an APU (either intel or AMD) will always be limited by size and TDP, and the necessity to upgrade both cpu and gpu at the same time. Laptops is where an igpu with HBM could shine, since it is pretty much impossible to upgrade or add a dgpu to a laptop.

But again, the market is somewhat limited, and also depends on price. An igpu gaming laptop has to be significantly cheaper than a dgpu model.

And we also have to see how AMD executes. Just because a 40% ipc was mentioned, based on projections, not even actual silicon, AMD fans have suddenly taken that as a certainty. And even if it happens, they have to keep the clocks up as well. Zen APUs have already been delayed to 2017, so lets see when they come out and if they do in fact have HBM. One would assume they will, but one would also have assumed the first Zen cpus would have an igp too, since AMD has been touting HSA, Fusion, gpgpu compute for what, going on ten years now?

My best guess is that AMD will make a moderate comeback in both cpu and dgpu sales, but I dont really think HBM is going to be the huge gamechanger that some do. As another poster said, they really need HSA to take off to make a major comeback. Otherwise, just a guess, but I would say only perhaps 10% at most of the market is really that interested in a powerful igpu.

 

[Ferzerp]

Pretty sure based on the track record that the eventual fate of AMD should be decided based on products that actually exist instead of the endless inflated claims.

"No really, wait for the next release! We really mean it this time guys". All the while their CPU performance gets farther and farther behind. At this point, the wild claims are worth less than the individual photons sending them to my eyes.

 

[AtenRa]

Just food for thought,

Current Kaveri 19W TDP A10-7300 is a Dual Module/Quad Core 1.9GHz base/3.2GHz turbo with 384 GPU cores at 533MHz and DDR-3 1600MHz.

Carrizo will have +5% IPC at 40% less power than Kaveri. So a 2GHz Currizo(base clocks) will be 5-10% faster in single thread than A10-7300 Kaveri(base clocks) but at 15W TDP.

Now, lets take that ZEN will have 40% IPC. A 2GHz ZEN APU will be 40% faster in single thread than Currizo(base clocks) without even increasing the clocks.

Add lower consumption from 14nm FF, add a faster iGPU and perhaps even HBM and 2017 APU could be exceptional.

 

[ShintaiDK]

Bandwidth, lots of it. No longer limited by system ram.

Think about it, if you were making an APU, would you have made the iGPU portion much faster? Nope, because it would have been limited by bandwidth regardless.

You could only do that if you bypassed the normal PC setup of motherboard + DDR3 + PCIE lanes. As what happened with Iris Pro, XBone & PS4 SOCs. Dedicated edram to remove bandwidth bottlenecks, or pure GDDR5 as system ram. These can only occur in custom designs and not the mainstream PC market.

With HBM, an AMD Zen APU with 2048 GCN+ cores and 4/8GB HBM on the APU itself, will be an amazing product. That's mid-range dGPU performance right there in a small package.

The problem is for AMD its not going to do anything result wise. Desktop APUs is pretty much dead, and any fast bandwidth IGPs will just massacre the part of the GPU business that makes any money. At best AMD can only choose the lesser of the 2 evils. And thats assuming AMD got a CPU that anyone wants to buy in the first place.

 

[ShintaiDK]

Just food for thought,

Current Kaveri 19W TDP A10-7300 is a Dual Module/Quad Core 1.9GHz base/3.2GHz turbo with 384 GPU cores at 533MHz and DDR-3 1600MHz.

Carrizo will have +5% IPC at 40% less power than Kaveri. So a 2GHz Currizo(base clocks) will be 5-10% faster in single thread than A10-7300 Kaveri(base clocks) but at 15W TDP.

Now, lets take that ZEN will have 40% IPC. A 2GHz ZEN APU will be 40% faster in single thread than Currizo(base clocks) without even increasing the clocks.

Add lower consumption from 14nm FF, add a faster iGPU and perhaps even HBM and 2017 APU could be exceptional.

How much does that 19W kaveri throttle?

 

[ShintaiDK]

Pretty sure based on the track record that the eventual fate of AMD should be decided based on products that actually exist instead of the endless inflated claims.

"No really, wait for the next release! We really mean it this time guys". All the while their CPU performance gets farther and farther behind. At this point, the wild claims are worth less than the individual photons sending them to my eyes.

Yep. This is why AMD is the new VIA. VIA made the exact same claims on their way down the ladder. Always the next product would turnaround the company. While in the end they never delivered. AMD havent delivered for ~10 years now. AMD will need a minor miracle to even hit 2B$ revenue in CPU+GPU PC segment combined for the year. Desktop APUs as we can see from JPR numbers quarter after quarter is a complete flop. So there is no future in desktop APUs. And graphics wise they lost badly to nVidia and now sits in a ~20/80 case.

 

[AtenRa]

How much does that 19W kaveri throttle?

On the CPU ???

 

[ShintaiDK]

On the CPU ???

http://www.notebookcheck.net/HP-EliteBook-745-G2-Notebook-Review.125934.0.html

The EliteBook processed the stress test (Prime95 and Furmark run for at least one hour) in AC and battery modes in the same manner. The CPU was throttled to a speed of 1.1 to 1.3 GHz, and the GPU mostly clocks with 288 MHz. The graphics core was occasionally boosted to 411 to 500 MHz. The CPU's temperature settled to approximately 67 °C in AC mode.

 

[coercitiv]

How much does that 19W kaveri throttle?

http://www.notebookcheck.net/HP-EliteBook-745-G2-Notebook-Review.125934.0.html

What you forgot to mention is power usage, since people will presume throttling is power related. We noticed similar behavior with the FX 7500 in monstercameron's review thread, and the funny thing was... system power usage was fine even with boost clocks under Cinebench.

Interesting: CPU is throttling, but it's not frequency related and temps are fine.

At 17:31 thread usage goes down fast and system power usage goes from 25-28W to 18-20W, and stays that way for the remainder of the test.

 

[DrMrLordX]

Why do they use Prime95 + Furmark as a stress test? Of course it's going to throttle. They're trying to murder the poor chip. Even my desktop Kaveri tried defaulting to p4 clockspeed in horror when I pulled that stunt.

And they ran it that way for a solid hour?