AMD Introduces New APUs For System Builder And DIY Market

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
AMD today introduced the new AMD A10-7800 Accelerated Processing Unit (APU) to the component channel. This 4th generation A-Series APU with 12 Compute Cores (4CPU + 8 GPU) unlocks the APU potential with Heterogeneous System Architecture (HSA) features, and boasts powerful AMD Radeon™ R7 Series graphics for outstanding performance across applications. Combined with AMD’s acclaimed Mantle API, the AMD A10-7800 APU can enable accelerated performance across select AMD Gaming Evolved partner titles.

Revolutionary Architecture
  • Up to 12 Compute Cores (4 CPU + 8 GPU) deliver amazing performance and responsiveness, that leverages the capability of the APU on various workloads, applications and top gaming titles
  • The breakthrough new Heterogeneous System Architecture (HSA) allows the CPU and GPU to work together in harmony by quickly dividing and directing tasks to the appropriate cores for new levels of performance and efficiency for desktop and notebook PCs, enabling a new era of compute capabilities with compute cores
  • Award-winning Graphics Core Next Architecture with AMD Radeon™ R7 series graphics and compute capability makes it a top choice for gamers who expect the best
  • Support for UltraHD (4K) resolutions and new video post-processing enhancements that make 1080p videos look amazing when upscaled on an UltraHD-enabled monitor or TV
 
HSA is the future as far as I'm concerned. Having an APU with shared memory between CPU and GPU is such an awesome idea. Now if only programmers world wide embrace it. This could be huge performance-wise and I eagerly await AMD's next gen tech as I have read news articles that AMD will in the future embrace a more modern (and Intel like) CPU architecture. Way to go AMD! :D
 
So what will this device be mainly used for? Bitcoin mining I'm guessing? What else could it be used for?
 
HSA is the future as far as I'm concerned. Having an APU with shared memory between CPU and GPU is such an awesome idea. Now if only programmers world wide embrace it. This could be huge performance-wise and I eagerly await AMD's next gen tech as I have read news articles that AMD will in the future embrace a more modern (and Intel like) CPU architecture. Way to go AMD! :D

Um, this is nothing new; the part about shared system memory between the CPU and GPU.
Years before APUs existed, low-end motherboards would have the CPU and IGP share system memory.

Perhaps the latency has gotten better with APUs, but the concept as a whole is very old.
AMD is just pushing a positive spin on a cheap solution.

This is why I was left scratching my head when everyone was so in awe at the XBone and PS4 with their shared memory architecture.
The only big thing about those was the PS4 using GDDR5 partly for system memory, woo.


So, yes, I suppose this is the "way of the future", as it has been for about the last 15 years. ;)
 
Um, this is nothing new; the part about shared system memory between the CPU and GPU.
Years before APUs existed, low-end motherboards would have the CPU and IGP share system memory.

Perhaps the latency has gotten better with APUs, but the concept as a whole is very old.
AMD is just pushing a positive spin on a cheap solution.

This is why I was left scratching my head when everyone was so in awe at the XBone and PS4 with their shared memory architecture.
The only big thing about those was the PS4 using GDDR5 partly for system memory, woo.


So, yes, I suppose this is the "way of the future", as it has been for about the last 15 years. ;)

Direct access is the important part, and why this is new and exciting. In old configurations (or even current intel) if the GPU wants to access somthing in System memory, the data has to be copied into GPU Memory / RAM before the GPU could access it. Now, its just direct access.

I love my trinity to use as an internet workstation / multi client Eve online box. I'm pretty pumped for the successor to Kaveri for an upgrade to my existing Trinity box, as I use that system way more than my dedicated Core i5 Gaming machine.
 
Oh nice, I could see how direct access could be an improvement.
 
So what will this device be mainly used for? Bitcoin mining I'm guessing? What else could it be used for?

Low power low profile htpc and light duty gaming for non enthusiast types. I'm going to set one up for a coworker once these hit shelves.
 
Direct access is the important part, and why this is new and exciting. In old configurations (or even current intel) if the GPU wants to access somthing in System memory, the data has to be copied into GPU Memory / RAM before the GPU could access it. Now, its just direct access.

Yes, copying information from system ram to vram is probably the biggest hurdle to solve for harnessing the power of the GPU.

This opens up a lot of possibilities!
 
Um, this is nothing new; the part about shared system memory between the CPU and GPU.
Years before APUs existed, low-end motherboards would have the CPU and IGP share system memory.

Perhaps the latency has gotten better with APUs, but the concept as a whole is very old.
AMD is just pushing a positive spin on a cheap solution.

This is why I was left scratching my head when everyone was so in awe at the XBone and PS4 with their shared memory architecture.
The only big thing about those was the PS4 using GDDR5 partly for system memory, woo.


So, yes, I suppose this is the "way of the future", as it has been for about the last 15 years. ;)

I think there's a misunderstanding (and please correct me if I'm wrong). Yes, currently the CPU and IGP "share" system memory, but they do not share a common memory address space. The IGP carves out a portion of the shared memory pool for itself, which the CPU cannot access at all. This is why, for example, if you have 8GB of memory Windows can only see 7.5GB of it (and the IGP only sees 512MB). The rest of it is eaten by the IGP. Data that is shared between the two also needs to be transferred between the 2 address spaces, which can be expensive.

What AMD is doing, and what is implemented in the PS4 for example, and making it so the CPU and IGP share a common address space. So the CPU and IGP both can see and access each other's data. This means both the CPU and IGP see all 8GB of that memory you have installed. This also means that data does not need to be transferred between the CPU and IGP, they all share the same data to begin with.
 
I think there's a misunderstanding (and please correct me if I'm wrong). Yes, currently the CPU and IGP "share" system memory, but they do not share a common memory address space. The IGP carves out a portion of the shared memory pool for itself, which the CPU cannot access at all. This is why, for example, if you have 8GB of memory Windows can only see 7.5GB of it (and the IGP only sees 512MB). The rest of it is eaten by the IGP. Data that is shared between the two also needs to be transferred between the 2 address spaces, which can be expensive.

What AMD is doing, and what is implemented in the PS4 for example, and making it so the CPU and IGP share a common address space. So the CPU and IGP both can see and access each other's data. This means both the CPU and IGP see all 8GB of that memory you have installed. This also means that data does not need to be transferred between the CPU and IGP, they all share the same data to begin with.

This makes perfect sense, thank you for the explanation. :cool:
 
I think there's a misunderstanding (and please correct me if I'm wrong). Yes, currently the CPU and IGP "share" system memory, but they do not share a common memory address space. The IGP carves out a portion of the shared memory pool for itself, which the CPU cannot access at all. This is why, for example, if you have 8GB of memory Windows can only see 7.5GB of it (and the IGP only sees 512MB). The rest of it is eaten by the IGP. Data that is shared between the two also needs to be transferred between the 2 address spaces, which can be expensive.

What AMD is doing, and what is implemented in the PS4 for example, and making it so the CPU and IGP share a common address space. So the CPU and IGP both can see and access each other's data. This means both the CPU and IGP see all 8GB of that memory you have installed. This also means that data does not need to be transferred between the CPU and IGP, they all share the same data to begin with.

Agreed, plus i think AMD is creating a "console like" platform for game developers, and with Mantle + HSA, this opens up the gaming market to low-cost PCs instead of just high-end Desktops
 
I love what AMD is doing with their APUs. Especially Kaveri with their new unified memory space, and heterogeneous queuing. The way Kaveri is built is going to be the future of all tech one day. AMD is an amazing company with an amazing vision. <3 it!
 
Actually this is pretty cool and I look forward to seeing where this goes next.
 
HSA is exciting but AMD still has quite a few hurdles to pass up before they can make their A10's shine.

They mentioned a possible business class for the A10, I want to see that.
 
When AMD actually delivers what they promise on any product, that will be front page news.
 
These apus are still slower than a hd7750 :(

Yeah, but for the price point and the overall performance they deliver, it's an outstanding accomplishment for a CPU containing an IGP. It wasn't that long ago that the best IGP was a MoBo with an nVidia 6150...right?
 
These apus are still slower than a hd7750 :(

You need to understand, these will never be faster their dedicated video cards, except ultra low end which are getting phased out.

Reason? They don't want to cannibalize their current money maker. As it stands, a cheap CPU+Mobo+Discrete GPU will give you better performance for similar money. Where APU's stand is they get you decent performance and simple builds.

For APU's to get better then just a Mobile platform favorite, they need to either increase the CPU's capability, lower price or increase GPU performance (which will eat into their GPU business). Increasing the GPU performance is still going to be tough as they are still stuck with a long pipe to memory, higher latency memory and still sharing it with the rest of the system.
 
Back
Top