News AMD Renoir APU Graphics Configurations Seemingly Discovered In Driver

[Admin]

The cat is partially out of the bag.

AMD Renoir APU Graphics Configurations Seemingly Discovered In Driver : Read more

 

[Rdslw]

anyone know place where I can preorder?
With 6/12 & 10×64 (assuming it will be identical to current, and I think they might go for 84 shaders per CU) its enough for casual usage, and with their power efficiency that laptop gonna be dope. I dont care about brand, I just want a mobile unit that can game sometimes without 2x 200W bricks.....
I've seen laptops with R7 & 32 GB of ram priced same way as i5 with 8GB of ram.... and I like what I saw, but I need 6/12 CPU..... so....
http://shutupandtakemymoney.com/wp-content/uploads/2018/06/shut-up-logo-1000.png

 

[ChaosFenix]

Isn't the bigger story here that support for AMD APUs are being found in APPLE bootcamp drivers? If this is accurate wouldn't it mean that apple is considering AMD APUs for at least some of apple products. And importantly not just as the GPU vendor that they have been for years but that apple would actually be using them for the CPU as well? At least in my book that would be an even bigger win than the Surface laptop if we are going to start seeing AMD APUs in like the Macbook Air.

 

[The_Arioch]

What is special in Navi comparing to Radeon 7, Instinct M, etc ?

It was said, AMD reduced computing power and used freed transistor budget to increase gaming power. Became more GeForcish.

Now i wonder, does Navi really suit APU well?
On gaming vs GPGPU scale desktop gaming cards and laptop MSM cards are clearly towards gaming.

Who of discrete cards customers wants more of compute - should instead buy an unrestricted Instinct M for premium price.

Now, what about pure APU w/o discrete videocard? Obviously there will be no two line-ups on the shelves, so to say APU-Instinct vs API-Radeon.
There'll be no choice.

So AMD has to maintain another gaming vs GPGPU balance there. And consider that they want to sell Navi MSM cards for gaming laptops, too.

Maybe shifting balance pro-gaming against-computing is not really needed for APU segment afterall?
Maybe keeping previous, GPGPU-friendly Radeon architecture makes sense there, actually?

 

[TJ Hooker]

anyone know place where I can preorder?
With 6/12 & 10×64 (assuming it will be identical to current, and I think they might go for 84 shaders per CU) its enough for casual usage, and with their power efficiency that laptop gonna be dope. I dont care about brand, I just want a mobile unit that can game sometimes without 2x 200W bricks.....
I've seen laptops with R7 & 32 GB of ram priced same way as i5 with 8GB of ram.... and I like what I saw, but I need 6/12 CPU..... so....
http://shutupandtakemymoney.com/wp-content/uploads/2018/06/shut-up-logo-1000.png

Where do you see that Renoir will be 6 core/12 thread?

All information I've seen points to the iGPUs being mostly the same as for Picasso, except with a clock speed bump (due to the node shrink). So nothing crazy in terms of GPU performance gains, I wouldn't expect a 10 CU renoir iGPU to perform hugely better than existing 10 CU APUs.

I don't think AMD has even officially announced these chips, let alone made them available for pre-order.

 

[ChaosFenix]

What is special in Navi comparing to Radeon 7, Instinct M, etc ?

It was said, AMD reduced computing power and used freed transistor budget to increase gaming power. Became more GeForcish.

Now i wonder, does Navi really suit APU well?
On gaming vs GPGPU scale desktop gaming cards and laptop MSM cards are clearly towards gaming.

Who of discrete cards customers wants more of compute - should instead buy an unrestricted Instinct M for premium price.

Now, what about pure APU w/o discrete videocard? Obviously there will be no two line-ups on the shelves, so to say APU-Instinct vs API-Radeon.
There'll be no choice.

So AMD has to maintain another gaming vs GPGPU balance there. And consider that they want to sell Navi MSM cards for gaming laptops, too.

Maybe shifting balance pro-gaming against-computing is not really needed for APU segment afterall?
Maybe keeping previous, GPGPU-friendly Radeon architecture makes sense there, actually?

I am not exactly sure on your questions here but Navi is more performant for gaming than Vega. The RDNA architectureuses less power and is more powerful of an architecture.This is how a 5700XT with 40 CUs is able to tie a Radeon VII with 60CUs, Both of these would be desirable in an APU as they are usually meant for mobile. RDNA is more or less backwards compatible on compute though with GCN so it is able to provide a performance boost for gaming while still being useful for GPGPU. The ideal APU here would be one with Zen2 on the CPU side and RDNA on the GPU side. We know they will be Zen2 but it is still unknown if it will use GCN or RDNA, though it is suspected it will be GCN. I honestly don't think that AMD is serious about APUs and mobile yet the margins in those spaces are lower and they have a limited fab capacity. If they can only produce 1000 chips a day it makes more sense to them to sell that silicon to enterprise with 65% margins than to consumers with 35% margins.

 

[Rdslw]

Where do you see that Renoir will be 6 core/12 thread?

All information I've seen points to the iGPUs being mostly the same as for Picasso, except with a clock speed bump (due to the node shrink). So nothing crazy in terms of GPU performance gains, I wouldn't expect a 10 CU renoir iGPU to perform hugely better than existing 10 CU APUs.

I don't think AMD has even officially announced these chips, let alone made them available for pre-order.

I don't know, but its logical thing as 3000 series don't even have 4 core part in desktop ....
and About IGPU, I dont really need sick gains, but I care for better power/heat work culture, and zen2 brought those to desktop.

 

[TJ Hooker]

I don't know, but its logical thing as 3000 series don't even have 4 core part in desktop ....

Well, neither did the 2000 series unless you count OEM-only parts.

 

[Rdslw]

Well, neither did the 2000 series unless you count OEM-only parts.

Well I want at least one with 6/12 config or more, not all. They push every little possible parameter each upgrade, let me have hope I could run away from intel mobile platform....

 

[MasterMadBones]

anyone know place where I can preorder?
With 6/12 & 10×64 (assuming it will be identical to current, and I think they might go for 84 shaders per CU) its enough for casual usage, and with their power efficiency that laptop gonna be dope. I dont care about brand, I just want a mobile unit that can game sometimes without 2x 200W bricks.....
I've seen laptops with R7 & 32 GB of ram priced same way as i5 with 8GB of ram.... and I like what I saw, but I need 6/12 CPU..... so....
http://shutupandtakemymoney.com/wp-content/uploads/2018/06/shut-up-logo-1000.png

Vega's design is limited to 64 SPs per CU. 84 SPs would require a major redesign.

Vega can't access each SP individually either. Each CU is comprised of 4 16-wide 64-way (4 cycles at 16 per cycle) SIMD units. 84 would be misaligned. Even 80 would have been more likely, not taking into account other design elements.

It was said, AMD reduced computing power and used freed transistor budget to increase gaming power. Became more GeForcish.

Navi hasn't reduced compute power compared to Vega at all. AMD has gone from the aforementioned configuration to two 32-way single-cycle SIMD units per CU, which reduces the throughput and latency penalties having to perform an instruction on a single data point, which happens more often in gaming or 3D rendering scenarios. This also improves compute performance but to a lesser degree. Especially machine learning tasks will not benefit very much, although RDNA does have less requirement for memory bandwidth, which is a big performance limiter in ML.

The more complex decoders and new cache hierarchy have increased the number of transistors required per CU, however, which is why large GCN GPUs make more sense for workloads that act on large collections of equivalent data.

Nevertheless, I don't think GCN will still be around after Arcturus. As FPGAs and ASICs are taking over the ML stage, GPUs continue to make less sense for these extremely SIMD-heavy workloads. GCN can't scale beyond 64 CUs in its current state either. Add to that the fact that the transistor benefit for GCN over RDNA becomes less of a consideration as new process nodes rise and mature.