Well, that didn't take long, PC gets GDDR5 system RAM support

[QUOTE="savagetwinkie"][QUOTE="04dcarraher"] being more efficient vs sheer processing power. Your efficiency arguments are flawed since all API overhead is all done on cpu's as long the cpu are fast enough you have no overhead to worry about. Now if you where comparing similar low clock cpu's and or much older based dual or quad core based cpu's then yes you would have a point. but the fact is that even with current console based multiplat games only require same era(2005/2006) based dual core cpu's around 3 ghz range to run the games as well as consoles, even with the massive overhead you keep on spouting about.ronvalencia

With similar GFLOPs, Crysis 2 on Radeon X1950 Pro rivals Xbox 360 's Crysis 2. Note that this PC is not equiped with Intel Atom class CPU design (in-order dual instruction issue per cycle type).

For large computation, the overhead issue is blown out of proportion. The problem is with the small amount computation on the GPU i.e. using the GPU like a CPU.

48 pixel shader processors 8 vertex shader processors

256-bit 8-channel GDDR4 memory interface

360 has 128bit memory interface and only has 48 shared vertex/pixel shaders, access only to gddr3 memory

[QUOTE="ronvalencia"]

[QUOTE="savagetwinkie"] its not all CPU based inefficiency, the GPU isn't being used as well as it could be because the drivers are more generic, and regardless of how fast CPU's are, the extra overhead creates latency.savagetwinkie

With similar GFLOPs, Crysis 2 on Radeon X1950 Pro rivals Xbox 360 's Crysis 2. Note that this PC is not equiped with Intel Atom class CPU design (in-order dual instruction issue per cycle type).

For large computation, the overhead issue is blown out of proportion. The problem is with the small amount computation on the GPU i.e. using the GPU like a CPU.

48 pixel shader processors 8 vertex shader processors

256-bit 8-channel GDDR4 memory interface

360 has 128bit memory interface and only has 48 shared vertex/pixel shaders, access only to gddr3 memory

ROPs+MCH advantage would be negated if the limitation was with compute e.g. Radeon X1950 Pro was running 2006 era games at higher resolution than Xbox 360.

From http://www.ati.com/products/radeonx1950/x1950pro_specs.html

Radeon x1950 Pro has

- 36 pixel shader processors and 8 vertex shaders i.e. a total of 48 shader processors.

- 256bit GDDR3 memory.

[QUOTE="ronvalencia"]

[QUOTE="savagetwinkie"] its not all CPU based inefficiency, the GPU isn't being used as well as it could be because the drivers are more generic, and regardless of how fast CPU's are, the extra overhead creates latency.savagetwinkie

With similar GFLOPs, Crysis 2 on Radeon X1950 Pro rivals Xbox 360 's Crysis 2. Note that this PC is not equiped with Intel Atom class CPU design (in-order dual instruction issue per cycle type).

For large computation, the overhead issue is blown out of proportion. The problem is with the small amount computation on the GPU i.e. using the GPU like a CPU.

actually not that much better...
the pixel and vertex shader processors are fixed on the x1950pro, meaning that the 360's Xenos has the advantage of allocating different ratio of pixel and vertex processor workloads. So if we look at the gpu specs from both (the model that ran Crysis 2 (pro))
x1950 pro
384 million transistors

36 pixel & 8 vertex processors
256mb GDDR3 at 44GB/s
able to do 248 GFLOPS

Xbox 360 Xenos
337 million transistors
48 unified shader processors less then 512mb of GDDR3 at 22GB/s
able to do 240 GFLOPS
Their pretty close with processing abilities only looking at a 8 GFLOPS difference thats only a 4% difference.

All those saying API overhead on Pc is a big monster :lol:

[QUOTE="ronvalencia"]

[QUOTE="savagetwinkie"]

dude that's lines of code, not instruction count, didn't you read the graph you posted? Also your completely wrong, this is probably done in a model environment, . Even if we are counting assembly lines that's not instruction count, a simple for loop could easily expand to a few thousand instructions but take 2 lines up. The graph is very ambiguous and the only real thing it shows is that HSA is likely a lot easier to get performance out of. But its all a relative performance comparison on a PC operating system. Secondly there is more overhead than just CPU instructions with games. The overhead comes from tons of other things. IO being used has it's overhead, the driver has its overhead, the memory has it's overhead, windows has it's overhead. Your comparing a relative performance running an algorithm with different methods. Like calling a driver in windows pushes it into an event loop so it might not even get processed immediately. This is why batch calls are important because the kernel call overhead gets too costly for calling it repeatedly. And API's make it so games can using generic calls to push data down to hardware but in the end the instructions still likely have to be processed and scaled for the hardware. Consoles can use compile time optimizations making it so the CPU doesn't have to do as much work to utilize the hardware. And that's why consoles are considered more efficient and have less overhead. Your missing the key aspect of what consoles are doing better, and that's better GPU utilization and a higher average throughput with similar hardware. None of your posts have addressed this key issue.

savagetwinkie

You missed compile(dark blue) section which is associated with the current driver stack i.e. JIT recomplier.

"Launch"(green) section would be associated with the data transfers (includes I/O and setup).

On the memory issue, the bar graph shows "copy" (yellow) and "copy back"(red) inefficiencies and it affects both C++ AMP and OpenCL.

From http://developer.amd.com/resources/heterogeneous-computing/what-is-heterogeneous-system-architecture-hsa/

A program running on the CPU queues work for the GPU using system calls through a device driver stack managed by a completely separate scheduler. This introduces significant dispatch latency, with overhead that makes the process worthwhile only when the application requires a very large amount of parallel computation.

...

The HSA team at AMD analyzed the performance of Haar Face Detect, a commonly used multi-stage video analysis algorithm used to identify faces in a video stream. The team compared a CPU/GPU implementation in OpenCL against an HSA implementation. The HSA version seamlessly shares data between CPU and GPU, without memory copies or cache flushes because it assigns each part of the workload to the most appropriate processor with minimal dispatch overhead.

You need to read more...

HSA software stack runs on discrete AMD GCNs (e.g. 7970), AMD APUs (e.g. Trinity) and other non-AMD HSA devices e.g. Qualcomm Adreno.

http://www.amd.com/jp/Documents/GCN_Architecture_whitepaper.pdf

The new GCN ISA was designed to facilitate industry standards. AMDs Heterogeneous System Architecture (HSA) is envisioned as a model for heterogeneous computing. It defines how CPUs and GPUs communicate and includes a virtual ISA (the HSA Intermediate Language), which is hardware agnostic. HSAIL code is dynamically compiled for the underlying hardware, and thus compatible with CPUs and GPUs from any vendor. GCN fully supports HSAIL, in part because of the shift to a more flexible instruction set. - page 11

Refer to http://hsafoundation.com/ for HSA partner list e.g. AMD(Radeon), ARM (Mali), Imagination (PowerVR), Qualcomm (Adreno).

Sony is part of HSA's contributors. Sony PS4 could be one example of AMD HSA box.

AMD plans to move OpenCL layer to the HSA stack. This version runs with MS Windows.

Like AMD64, AMD gains Microsoft's support....

http://hsafoundation.com/f-a-q/

Q: Will HSA support Microsoft C++ AMP?

A: HSA will support Microsoft C++ AMP

Current AMD Radeon HD driver stack i.e. notice the current OpenCL running on the same stack as DirectX and OpenGL.

AMD's Fusion partners...

[QUOTE="savagetwinkie"][QUOTE="ronvalencia"]

You missed compile(dark blue) section which is associated with the current driver stack i.e. JIT recomplier.

"Launch"(green) section would be associated with the data transfers (includes I/O and setup).

On the memory issue, the bar graph shows "copy" (yellow) and "copy back"(red) inefficiencies and it affects both C++ AMP and OpenCL.

From http://developer.amd.com/resources/heterogeneous-computing/what-is-heterogeneous-system-architecture-hsa/

A program running on the CPU queues work for the GPU using system calls through a device driver stack managed by a completely separate scheduler. This introduces significant dispatch latency, with overhead that makes the process worthwhile only when the application requires a very large amount of parallel computation.

...

The HSA team at AMD analyzed the performance of Haar Face Detect, a commonly used multi-stage video analysis algorithm used to identify faces in a video stream. The team compared a CPU/GPU implementation in OpenCL against an HSA implementation. The HSA version seamlessly shares data between CPU and GPU, without memory copies or cache flushes because it assigns each part of the workload to the most appropriate processor with minimal dispatch overhead.

ronvalencia

You need to read more... HSA software stack runs on discrete AMD GCNs (e.g. 7970), AMD APUs (e.g. Trinity) and other non-AMD HSA devices e.g. Qualcomm Adreno.

http://www.amd.com/jp/Documents/GCN_Architecture_whitepaper.pdf

The new GCN ISA was designed to facilitate industry standards. AMDs Heterogeneous System Architecture (HSA) is envisioned as a model for heterogeneous computing. It defines how CPUs and GPUs communicate and includes a virtual ISA (the HSA Intermediate Language), which is hardware agnostic. HSAIL code is dynamically compiled for the underlying hardware, and thus compatible with CPUs and GPUs from any vendor. GCN fully supports HSAIL, in part because of the shift to a more flexible instruction set. - page 11

Refer to http://hsafoundation.com/ for HSA partner list e.g. AMD(Radeon), ARM (Mali), Imagination (PowerVR), Qualcomm (Adreno).

Sony is part of HSA's contributors.

Well that copy paste didn't work, but either way you mixing up the software stack with hardware specification, the hardware specification is what's important to get the full benefits of HSA.

http://en.wikipedia.org/wiki/AMD_Accelerated_Processing_Unit#Heterogeneous_System_Architecture

there is a nice little graph for the system architecture in there.

[QUOTE="savagetwinkie"][QUOTE="ronvalencia"]

With similar GFLOPs, Crysis 2 on Radeon X1950 Pro rivals Xbox 360 's Crysis 2. Note that this PC is not equiped with Intel Atom class CPU design (in-order dual instruction issue per cycle type).

For large computation, the overhead issue is blown out of proportion. The problem is with the small amount computation on the GPU i.e. using the GPU like a CPU.

04dcarraher

actually not that much better...
the pixel and vertex shader processors are fixed on the x1950pro, meaning that the 360's Xenos has the advantage of allocating different ratio of pixel and vertex processor workloads. So if we look at the gpu specs from both (the model that ran Crysis 2 (pro))
x1950 pro
384 million transistors

36 pixel & 8 vertex processors
256mb GDDR3 at 44GB/s
able to do 248 GFLOPS

Xbox 360 Xenos
337 million transistors
48 unified shader processors less then 512mb of GDDR3 at 22GB/s
able to do 240 GFLOPS
Their pretty close with processing abilities only looking at a 8 GFLOPS difference thats only a 4% difference.

All those saying API overhead on Pc is a big monster :lol:

The API overhead is a monster for multiple little compute loads i.e. using GPU as a CPU. ;)

The current APIs are fine with large compute loads.

i mean

HSA MMU GPU can access the entire system memory through the translation services and page fault management of HSA MMU. [27]

This is clearly important for avoiding IO and memory copies, but its definitly not a software feature.

[QUOTE="ronvalencia"]

[QUOTE="savagetwinkie"] So a little more reading on HSA, this is a completely different hardware standard targeted towards SoC's and apu's, its eve less relevant than I originally thought since gaming isn't going to detach itself from directx anytime soon. The entire point of DirectX is to abstract specific hardware compatibility out to make games deployable across all PC'ssavagetwinkie

You need to read more... HSA software stack runs on discrete AMD GCNs (e.g. 7970), AMD APUs (e.g. Trinity) and other non-AMD HSA devices e.g. Qualcomm Adreno.

http://www.amd.com/jp/Documents/GCN_Architecture_whitepaper.pdf

The new GCN ISA was designed to facilitate industry standards. AMDs Heterogeneous System Architecture (HSA) is envisioned as a model for heterogeneous computing. It defines how CPUs and GPUs communicate and includes a virtual ISA (the HSA Intermediate Language), which is hardware agnostic. HSAIL code is dynamically compiled for the underlying hardware, and thus compatible with CPUs and GPUs from any vendor. GCN fully supports HSAIL, in part because of the shift to a more flexible instruction set. - page 11

Refer to http://hsafoundation.com/ for HSA partner list e.g. AMD(Radeon), ARM (Mali), Imagination (PowerVR), Qualcomm (Adreno).

Sony is part of HSA's contributors.

"Can be used both on discrete GPUs or on the APU."

Also notice "HSA Algorithms in Games" i.e. could be applied for Xbox Next, PS4 and PC.

From http://www.slideshare.net/zlatan4177/gpgpu-algorithms-in-games

The "HSA Algorithms in Games" slides includes reference to Sony's PS3 and hard-core gaming PC owners (with multi-GPU setups).

HSA solves the main problems with DirectX.

http://www.brightsideofnews.com/news/2011/11/30/radeon-hd-7000-revealed-amd-to-mix-gcn-with-vliw4--vliw5-architectures.aspx

i mean

HSA MMU GPU can access the entire system memory through the translation services and page fault management of HSA MMU. [27]

This is clearly important for avoiding IO and memory copies, but its definitly not a software feature.

savagetwinkie

So if you care enough about this GDDR5 thing, you will be able to have it on a PC.

Long story short: when DDR4 arrives, GDDR5 will be rendered irrelevant. Quote:

"Once the DDR4 memory comes along the advantages of GDDR5 start to diminish again. DDR4 will enable higher densities at comparable speeds as well as upgradeable modules which might be more desirable for customers."

Also

GDDR6 Memory Coming in 2014

IgGy621985

You don't know what RAID is ?

Also, SSD have goten cheaper... 160$ is more like a 240GB SSD. Not only does it load faster, but the access time is instant compare to HDD.

Bebi_vegeta

I see a new thing here, the instruction set which can run on GCN's appears to be a different thing then the HSA hardware system architecture. The instruction set will just allow programs that are compiled now be able to run on future iterations of the hardware.

Well that copy paste didn't work, but either way you mixing up the software stack with hardware specification, the hardware specification is what's important to get the full benefits of HSA.

http://en.wikipedia.org/wiki/AMD_Accelerated_Processing_Unit#Heterogeneous_System_Architecture

there is a nice little graph for the system architecture in there.

savagetwinkie

[QUOTE="savagetwinkie"][QUOTE="ronvalencia"]

You need to read more... HSA software stack runs on discrete AMD GCNs (e.g. 7970), AMD APUs (e.g. Trinity) and other non-AMD HSA devices e.g. Qualcomm Adreno.

http://www.amd.com/jp/Documents/GCN_Architecture_whitepaper.pdf

The new GCN ISA was designed to facilitate industry standards. AMDs Heterogeneous System Architecture (HSA) is envisioned as a model for heterogeneous computing. It defines how CPUs and GPUs communicate and includes a virtual ISA (the HSA Intermediate Language), which is hardware agnostic. HSAIL code is dynamically compiled for the underlying hardware, and thus compatible with CPUs and GPUs from any vendor. GCN fully supports HSAIL, in part because of the shift to a more flexible instruction set. - page 11

Refer to http://hsafoundation.com/ for HSA partner list e.g. AMD(Radeon), ARM (Mali), Imagination (PowerVR), Qualcomm (Adreno).

Sony is part of HSA's contributors.

ronvalencia

"Can be used both on discrete GPUs or on the APU."

Also notice "HSA Algorithms in Games" i.e. could be applied for Xbox Next, PS4 and PC.

From http://www.slideshare.net/zlatan4177/gpgpu-algorithms-in-games

The "HSA Algorithms in Games" slides includes reference to Sony's PS3 and hard-core gaming PC owners (with multi-GPU setups).

HSA solves the main problems with DirectX.

http://www.brightsideofnews.com/news/2011/11/30/radeon-hd-7000-revealed-amd-to-mix-gcn-with-vliw4--vliw5-architectures.aspx

[QUOTE="MK-Professor"]

Tormentos is still mad that the ps4 is getting a low-end hardware in comparison with the xbox back in 2005.

ronvalencia

A GCN with 18 CUs is not low end hardware i.e. it's not Bonaire XT (aka Radeon HD 7790) .

http://wccftech.com/amd-reported-preparing-radeon-hd-7790-based-bonaire-xt-gpu-launches-april/

yes it is, the HD7850 introduced last year as a mid-range GPU, when ps4 will be released the HD7850 is going to be almost 2 years old mid-range GPU = low-end:|

[QUOTE="ronvalencia"]

[QUOTE="MK-Professor"]

Tormentos is still mad that the ps4 is getting a low-end hardware in comparison with the xbox back in 2005.

MK-Professor

A GCN with 18 CUs is not low end hardware i.e. it's not Bonaire XT (aka Radeon HD 7790) .

http://wccftech.com/amd-reported-preparing-radeon-hd-7790-based-bonaire-xt-gpu-launches-april/

yes it is, the HD7850 introduced last year as a mid-range GPU, when ps4 will be released the HD7850 is going to be almost 2 years old mid-range GPU = low-end:|

[QUOTE="MK-Professor"]

[QUOTE="ronvalencia"]

A GCN with 18 CUs is not low end hardware i.e. it's not Bonaire XT (aka Radeon HD 7790) .

http://wccftech.com/amd-reported-preparing-radeon-hd-7790-based-bonaire-xt-gpu-launches-april/

 

tormentos

yes it is, the HD7850 introduced last year as a mid-range GPU, when ps4 will be released the HD7850 is going to be almost 2 years old mid-range GPU = low-end:|

it means that the ps4 is getting a low-end hardware in comparison with the xbox back in 2005. still mad?

[QUOTE="savagetwinkie"] Under a unified address space... and if there is a second GPU dedicate the apu to compute. Do you even read the stuff you post? This is still all talking about using APU's where the memory address is completely shared.tormentos

edit: so basically in a few years if HSA completely catches on, and nvidia starts making compliant hardware, you could see the new directx be built on that instaed. But it wouldn't be a simple gpu upgrade for the new directx.

[QUOTE="tormentos"][QUOTE="MK-Professor"]

yes it is, the HD7850 introduced last year as a mid-range GPU, when ps4 will be released the HD7850 is going to be almost 2 years old mid-range GPU = low-end:|

MK-Professor

it means that the ps4 is getting a low-end hardware in comparison with the xbox back in 2005. still mad?

In terms for consoles a 7850 type gpu is great its better then what most have expected. however others wanted a stronger gpu rivaling top ended pc gpu's some have come to terms that its not going to happen and others are still clinging on to that false sense of hope.  With todays series of gpu's a 7850 is a mid ranged gpu no matter how you slice it. Being GCN 2 and having HSA means nothing since GCN 2 vs GCN = no real architectural  changes made to actually improve gpu porition of performance. And HSA is available with the 7000 series and any newer series...

Let alone  a 7850 isnt much faster them my mid ranged gpu from 2011.

[QUOTE="Bebi_vegeta"]

 

You don't know what RAID is ?

Also, SSD have goten cheaper... 160$ is more like a 240GB SSD. Not only does it load faster, but the access time is instant compare to HDD.

tormentos

I don't care if you don't need that **** or not, I can use it on my PC you can't use it on your console... it's a simple matter that I have the option and you don't. I'm glad you love a close and simple system... I have nothing agaisnt it.

PC argument as never been about price.... since it can be expensive or cheap, it's up to the user because again WE HAVE THE OPTION.

Look I'll beleive when I see it, last time Sony BS us with 120fps X 2 screens full HD and 4th dimension stuff... which never happened.

Also, you seriously need to stops it with yout crystal ball prediction as to what the PS4 GPU can do, will do or should do. You don't know **** about it with the crappy numbers, they barely gave any info of it.

I never said SSD was cheap, I said it was better then the price you suggested... which thank you for proving my point with the link.

We all know SSD for game is only about loading stuff, since all the game important files are loaded on the ram anyway.

I have raided SSD, you say it's trowing money away... just let me enjoy what I bought, cause I like it. If I could afford a Ferrari I would buy one... but you'd probably say I'd be wasting money.

[QUOTE="ronvalencia"]

[QUOTE="savagetwinkie"] I see a new thing here, the instruction set which can run on GCN's appears to be a different thing then the HSA hardware system architecture. The instruction set will just allow programs that are compiled now be able to run on future iterations of the hardware. AMD HSA Implementation Type HSA Feature First Implemented Notes Optimized Platform GPU Compute C++ Support 2012 Trinity APU Support OpenCL C++ directions and Microsofts upcoming C++ AMP language. This eases programming of both CPU and GPU working together to process support parallel workloads. [27] HSA MMU GPU can access the entire system memory through the translation services and page fault management of HSA MMU. [27] Shared Power Management Power budget is now shared by CPU and GPU, while priority of having more power budget depends on which is more suited to the current tasks.[27] Architectural Integration Unified Address Space for CPU and GPU[1][19] 2013 Kaveri APU CPU and GPU now access the memory with the same address space. Pointers can now be freely passed between CPU and GPU. [27] Fully coherent memory between CPU & GPU GPU can now access and cache data from coherent memory regions in the system memory, and also reference the data from CPU's cache. Cache coherency is maintained. [27] GPU uses pageable system memory via CPU pointers GPU can take advantage of the shared virtual memory between CPU and GPU, and pageable system memory can now be referenced directly by the GPU, instead of being copied or pinned before accessing. [27] System Integration GPU compute context switch 2014 Compute tasks on GPU can be context switched, allowing a multi-tasking environment and also faster interpretation between applications, compute and graphics. [27] GPU graphics pre-emption Long-running graphics tasks can be pre-empted in order to allow processes having low latency access to the GPU.[27] Quality of Service[1] In addition to context switch and pre-emption, hardware resources can be either equalized or prioritized among multiple users and applications.[27] savagetwinkie

"Can be used both on discrete GPUs or on the APU."

Also notice "HSA Algorithms in Games" i.e. could be applied for Xbox Next, PS4 and PC.

From http://www.slideshare.net/zlatan4177/gpgpu-algorithms-in-games

The "HSA Algorithms in Games" slides includes reference to Sony's PS3 and hard-core gaming PC owners (with multi-GPU setups).

HSA solves the main problems with DirectX.

http://www.brightsideofnews.com/news/2011/11/30/radeon-hd-7000-revealed-amd-to-mix-gcn-with-vliw4--vliw5-architectures.aspx

LOL. Are you claiming discrete GCNs doesn't support HSA driver stack?

The current discrete GCN already has X86-64 pointer support i.e. X86 IOMMU(1). Do you even read the stuff I posted?

1. http://www.amd.com/jp/Documents/GCN_Architecture_whitepaper.pdf Page 11.

GCN incorporates an I/O Memory Management Unit (IOMMU), which can transparently map x86 addresses for the GPU. This means that the DMA engines in GCN can easily access pageable CPU memory, without the overhead of address translation, to move data.

From http://www.slideshare.net/zlatan4177/gpgpu-algorithms-in-games last slide

"APU can use to off load the CPU" i.e. it's an option.

Remember, a two socket CPUs are connected by an external bus which has an "unified address space" and each socketed CPU has it's own memory controllers.

Unified logically doesn't have to mean unified physically. Think of NUMA with heterogenous nodes.

It would be a stupid move for AMD if HSA can't scale like a NUMA setup i.e. hint: the competition can already do this (guess who).

:lol: Just now? PS4 announced it last month. PC is always a step behind it seems. *shakes head*Sooshy

:lol: Just now? PS4 announced it last month. PC is always a step behind it seems. *shakes head*Sooshy

it means that the ps4 is getting a low-end hardware in comparison with the xbox back in 2005. still mad?MK-Professor

[QUOTE="MK-Professor"]it means that the ps4 is getting a low-end hardware in comparison with the xbox back in 2005. still mad?tormentos

I don't care if you don't need that **** or not, I can use it on my PC you can't use it on your console... it's a simple matter that I have the option and you don't. I'm glad you love a close and simple system... I have nothing agaisnt it.

PC argument as never been about price.... since it can be expensive or cheap, it's up to the user because again WE HAVE THE OPTION.

Look I'll beleive when I see it, last time Sony BS us with 120fps X 2 screens full HD and 4th dimension stuff... which never happened.

Also, you seriously need to stops it with yout crystal ball prediction as to what the PS4 GPU can do, will do or should do. You don't know **** about it with the crappy numbers, they barely gave any info of it.

I never said SSD was cheap, I said it was better then the price you suggested... which thank you for proving my point with the link.

We all know SSD for game is only about loading stuff, since all the game important files are loaded on the ram anyway.

I have raided SSD, you say it's trowing money away... just let me enjoy what I bought, cause I like it. If I could afford a Ferrari I would buy one... but you'd probably say I'd be wasting money.

 

 

Bebi_vegeta

[QUOTE="Sooshy"]:lol: Just now? PS4 announced it last month. PC is always a step behind it seems. *shakes head*mitu123

[QUOTE="clyde46"] Smoke and mirrors mostly.tormentos

[QUOTE="Bebi_vegeta"]I don't care if you don't need that **** or not, I can use it on my PC you can't use it on your console... it's a simple matter that I have the option and you don't. I'm glad you love a close and simple system... I have nothing agaisnt it.

PC argument as never been about price.... since it can be expensive or cheap, it's up to the user because again WE HAVE THE OPTION.

Look I'll beleive when I see it, last time Sony BS us with 120fps X 2 screens full HD and 4th dimension stuff... which never happened.

Also, you seriously need to stops it with yout crystal ball prediction as to what the PS4 GPU can do, will do or should do. You don't know **** about it with the crappy numbers, they barely gave any info of it.

I never said SSD was cheap, I said it was better then the price you suggested... which thank you for proving my point with the link.

We all know SSD for game is only about loading stuff, since all the game important files are loaded on the ram anyway.

I have raided SSD, you say it's trowing money away... just let me enjoy what I bought, cause I like it. If I could afford a Ferrari I would buy one... but you'd probably say I'd be wasting money.

 

 

tormentos

I don't care if "U" don't need it, it's a feature PC has. Just like SSD, raid HDD will help load the games faster or whatever else that needs data transfering, it's not just for game smart guy.

Again, I never said PC is cheaper for F sake. I don't expect a system that can do everything to be cheaper then a machine that is limited.

Yet, you would still buy a Ferrari if you could and not a God damn Porsche.... and wth is a 2.5 HDD ?

[QUOTE="clyde46"] Smoke and mirrors mostly.tormentos

Hell we'll never know because that game never came out on X360.

[QUOTE="ronvalencia"] Again, "Can be used both on discrete GPUs or on the APU." tormentos

It's a no brainer that HSA is a "design" i.e. it's a combined software and hardware architecture.

The current discrete GCNs has support for HSA driver.

From http://www.amd.com/jp/Documents/GCN_Architecture_whitepaper.pdf

Equally important, the cache hierarchy was designed to integrate with x86 microprocessors. The GCN virtual memory system can support 4KB pages, which is the natural mapping granularity for the x86 address space - paving the way for a shared address space in the future. In fact, the IOMMU used for DMA transfers can already translate requests into the x86 address space to help move data to the GPU and this functionality will grow over time. Additionally, the caches in GCN use 64B lines, which is the same size as x86 processors. This sets the stage for heterogeneous systems to transparently share data between the GPU and CPU through the traditional caching system, without explicit programmer control. - page 10

....

The new GCN ISA was designed to facilitate industry standards. AMDs Heterogeneous System Architecture (HSA) is envisioned as a model for heterogeneous computing.

It defines how CPUs and GPUs communicate and includes a virtual ISA (the HSA Intermediate Language), which is hardware agnostic. HSAIL code is dynamically compiled for the underlying hardware, and thus compatible with CPUs and GPUs from any vendor. GCN fully supports HSAIL - page 11.

...

The DMA engines are one area where AMD's experience with x86 microprocessors has paid off. GCN incorporates an I/O Memory Management Unit (IOMMU), which can transparently map x86 addresses for the GPU. This means that the DMA engines in GCN can easily access pageable CPU memory, without the overhead of address translation, to move data. The IOMMU is a step towards tighter heterogeneous integration and will evolve over time.

-------------

For NVIDIA camp. AMD's HSA driver is similar to NVIDIA's TCC driver. TCC driver model is not an open standard.

With NVIDIA's TCC (Tesla Compute Cluster) driver, it enables Fermi (GF100)'s X86-64 unified address features.

For example, NVIDIA also has unified pointer between the host CPU (X86-64) and Fermi (only for device capability 2.0 and TCC driver model). http://developer.download.nvidia.com/compute/cuda/4_2/rel/toolkit/docs/online/group__CUDA__UNIFIED.html

TCC driver = Tesla Compute Cluster driver(1). Device capability 2.0 = GF100(2)(same ASIC as Geforce GTX480) and above.

1. http://http.developer.nvidia.com/ParallelNsight/2.1/Documentation/UserGuide/HTML/Content/Tesla_Compute_Cluster.htm Also, Nvidia made the similar comment as AMD on how crap is MS's WDDM i.e. TCC driver reduces the CUDA kernel launch (dispatch) overhead.

2. http://en.wikipedia.org/wiki/CUDA

Nvidia needs to make TCC driver available for normal Geforce Fermi end-users i.e. it has some issues to be solved e.g. interop with GPU's graphics units and support for legacy Direct3D. TCC driver is locked to "Tesla" hardware i.e. needs to "soft mod" the normal Geforce Fermi/Kepler.

Like OpenCL, HSA standard is open and therefore NVIDIA can support HSA games feature.

AMD's HSA is a response to NVIDIA's TCC. i.e. With TCC driver, GF100 says Hi.

From https://devtalk.nvidia.com/default/topic/529622/gtx-690-cuda-shared-memory-space-tcc-uva-gpudirect-/

"I've used this idea very effectively to give GPUs easy runtime access to a database over 14GB in size."

Notice "zero copy" feature with NVIDIA GPUs..

[QUOTE="savagetwinkie"]I see a new thing here, the instruction set which can run on GCN's appears to be a different thing then the HSA hardware system architecture. The instruction set will just allow programs that are compiled now be able to run on future iterations of the hardware.

Well that copy paste didn't work, but either way you mixing up the software stack with hardware specification, the hardware specification is what's important to get the full benefits of HSA.

http://en.wikipedia.org/wiki/AMD_Accelerated_Processing_Unit#Heterogeneous_System_Architecture

there is a nice little graph for the system architecture in there.

tormentos

From http://www.amd.com/jp/Documents/GCN_Architecture_whitepaper.pdf

Equally important, the cache hierarchy was designed to integrate with x86 microprocessors. The GCN virtual memory system can support 4KB pages, which is the natural mapping granularity for the x86 address space - paving the way for a shared address space in the future. In fact, the IOMMU used for DMA transfers can already translate requests into the x86 address space to help move data to the GPU and this functionality will grow over time. Additionally, the caches in GCN use 64B lines, which is the same size as x86 processors. This sets the stage for heterogeneous systems to transparently share data between the GPU and CPU through the traditional caching system, without explicit programmer control. - page 10

....

The new GCN ISA was designed to facilitate industry standards. AMDs Heterogeneous System Architecture (HSA) is envisioned as a model for heterogeneous computing.

It defines how CPUs and GPUs communicate and includes a virtual ISA (the HSA Intermediate Language), which is hardware agnostic. HSAIL code is dynamically compiled for the underlying hardware, and thus compatible with CPUs and GPUs from any vendor. GCN fully supports HSAIL - page 11.

...

[QUOTE="tormentos"][QUOTE="savagetwinkie"] Under a unified address space... and if there is a second GPU dedicate the apu to compute. Do you even read the stuff you post? This is still all talking about using APU's where the memory address is completely shared.savagetwinkie

edit: so basically in a few years if HSA completely catches on, and nvidia starts making compliant hardware, you could see the new directx be built on that instaed. But it wouldn't be a simple gpu upgrade for the new directx.

NVIDIA's Fermi GF100 (April 2011) architecture can also take X86-64bit pointers (with NVIDIA's TCC driver).

I don't care if "U" don't need it, it's a feature PC has. Just like SSD, raid HDD will help load the games faster or whatever else that needs data transfering, it's not just for game smart guy.

Again, I never said PC is cheaper for F sake. I don't expect a system that can do everything to be cheaper then a machine that is limited.

Yet, you would still buy a Ferrari if you could and not a God damn Porsche.... and wth is a 2.5 HDD ?

Bebi_vegeta

Hell we'll never know because that game never came out on X360.Bebi_vegeta

[QUOTE="tormentos"][QUOTE="ronvalencia"] Again, "Can be used both on discrete GPUs or on the APU." ronvalencia

It's a no brainer that HSA is a "design" i.e. it's a combined software and hardware architecture.

The current discrete GCNs has support for HSA driver.

From http://www.amd.com/jp/Documents/GCN_Architecture_whitepaper.pdf

Equally important, the cache hierarchy was designed to integrate with x86 microprocessors. The GCN virtual memory system can support 4KB pages, which is the natural mapping granularity for the x86 address space - paving the way for a shared address space in the future. In fact, the IOMMU used for DMA transfers can already translate requests into the x86 address space to help move data to the GPU and this functionality will grow over time. Additionally, the caches in GCN use 64B lines, which is the same size as x86 processors. This sets the stage for heterogeneous systems to transparently share data between the GPU and CPU through the traditional caching system, without explicit programmer control. - page 10

....

The new GCN ISA was designed to facilitate industry standards. AMDs Heterogeneous System Architecture (HSA) is envisioned as a model for heterogeneous computing.

It defines how CPUs and GPUs communicate and includes a virtual ISA (the HSA Intermediate Language), which is hardware agnostic. HSAIL code is dynamically compiled for the underlying hardware, and thus compatible with CPUs and GPUs from any vendor. GCN fully supports HSAIL - page 11.

...

The DMA engines are one area where AMD's experience with x86 microprocessors has paid off. GCN incorporates an I/O Memory Management Unit (IOMMU), which can transparently map x86 addresses for the GPU. This means that the DMA engines in GCN can easily access pageable CPU memory, without the overhead of address translation, to move data. The IOMMU is a step towards tighter heterogeneous integration and will evolve over time.

-------------

For NVIDIA camp. AMD's HSA driver is similar to NVIDIA's TCC driver. TCC driver model is not an open standard.

With NVIDIA's TCC (Tesla Compute Cluster) driver, it enables Fermi (GF100)'s X86-64 unified address features.

For example, NVIDIA also has unified pointer between the host CPU (X86-64) and Fermi (only for device capability 2.0 and TCC driver model). http://developer.download.nvidia.com/compute/cuda/4_2/rel/toolkit/docs/online/group__CUDA__UNIFIED.html

TCC driver = Tesla Compute Cluster driver(1). Device capability 2.0 = GF100(2)(same ASIC as Geforce GTX480) and above.

1. http://http.developer.nvidia.com/ParallelNsight/2.1/Documentation/UserGuide/HTML/Content/Tesla_Compute_Cluster.htm Also, Nvidia made the similar comment as AMD on how crap is MS's WDDM i.e. TCC driver reduces the CUDA kernel launch (dispatch) overhead.

2. http://en.wikipedia.org/wiki/CUDA

Nvidia needs to make TCC driver available for normal Geforce Fermi end-users i.e. it has some issues to be solved e.g. interop with GPU's graphics units and support for legacy Direct3D. TCC driver is locked to "Tesla" hardware i.e. needs to "soft mod" the normal Geforce Fermi/Kepler.

Like OpenCL, HSA standard is open and therefore NVIDIA can support HSA games feature.

AMD's HSA is a response to NVIDIA's TCC. i.e. With TCC driver, GF100 says Hi.

From https://devtalk.nvidia.com/default/topic/529622/gtx-690-cuda-shared-memory-space-tcc-uva-gpudirect-/

"I've used this idea very effectively to give GPUs easy runtime access to a database over 14GB in size."

Notice "zero copy" feature with NVIDIA GPUs..

[QUOTE="tormentos"][QUOTE="savagetwinkie"]I see a new thing here, the instruction set which can run on GCN's appears to be a different thing then the HSA hardware system architecture. The instruction set will just allow programs that are compiled now be able to run on future iterations of the hardware.

Well that copy paste didn't work, but either way you mixing up the software stack with hardware specification, the hardware specification is what's important to get the full benefits of HSA.

http://en.wikipedia.org/wiki/AMD_Accelerated_Processing_Unit#Heterogeneous_System_Architecture

there is a nice little graph for the system architecture in there.

ronvalencia

From http://www.amd.com/jp/Documents/GCN_Architecture_whitepaper.pdf

Equally important, the cache hierarchy was designed to integrate with x86 microprocessors. The GCN virtual memory system can support 4KB pages, which is the natural mapping granularity for the x86 address space - paving the way for a shared address space in the future. In fact, the IOMMU used for DMA transfers can already translate requests into the x86 address space to help move data to the GPU and this functionality will grow over time. Additionally, the caches in GCN use 64B lines, which is the same size as x86 processors. This sets the stage for heterogeneous systems to transparently share data between the GPU and CPU through the traditional caching system, without explicit programmer control. - page 10

....

The new GCN ISA was designed to facilitate industry standards. AMDs Heterogeneous System Architecture (HSA) is envisioned as a model for heterogeneous computing.

It defines how CPUs and GPUs communicate and includes a virtual ISA (the HSA Intermediate Language), which is hardware agnostic. HSAIL code is dynamically compiled for the underlying hardware, and thus compatible with CPUs and GPUs from any vendor. GCN fully supports HSAIL - page 11.

...

[QUOTE="Bebi_vegeta"]Hell we'll never know because that game never came out on X360.tormentos

[QUOTE="tormentos"][QUOTE="Bebi_vegeta"]Hell we'll never know because that game never came out on X360.faizan_faizan

[QUOTE="ronvalencia"]

[QUOTE="tormentos"] You are loosing your time i already show Ron that HSA is not something you active by software,HSA requires a design change which most PC lack,APU systems are HSA designs,i even quote from AMD own page.. https://www.youtube.com/watch?v=Me0g_FsgU0U

tormentos

From http://www.amd.com/jp/Documents/GCN_Architecture_whitepaper.pdf

Equally important, the cache hierarchy was designed to integrate with x86 microprocessors. The GCN virtual memory system can support 4KB pages, which is the natural mapping granularity for the x86 address space - paving the way for a shared address space in the future. In fact, the IOMMU used for DMA transfers can already translate requests into the x86 address space to help move data to the GPU and this functionality will grow over time. Additionally, the caches in GCN use 64B lines, which is the same size as x86 processors. This sets the stage for heterogeneous systems to transparently share data between the GPU and CPU through the traditional caching system, without explicit programmer control. - page 10

....

The new GCN ISA was designed to facilitate industry standards. AMDs Heterogeneous System Architecture (HSA) is envisioned as a model for heterogeneous computing.

It defines how CPUs and GPUs communicate and includes a virtual ISA (the HSA Intermediate Language), which is hardware agnostic. HSAIL code is dynamically compiled for the underlying hardware, and thus compatible with CPUs and GPUs from any vendor. GCN fully supports HSAIL - page 11.

...

Did you forget NUMA when you have two socket CPUs with each having it's own memory controllers? AMD Opterons already has support for X86-64 NUMA.

You can have mix memory speed with unified address mode i.e. this was already been done with nVidia Tesla Fermi (with TCC driver), AMD Opterons and Intel Xeons.

Replace one of the Opteron node with a GCN node. Logically, all four Opteron nodes has unified memory address space and each can have different memory module speed.

It would be stupid for AMD if HSA can't scale via NUMA type model i.e. NVIDIA Fermi/Kepler already scales via NUMA type model (with X86-64bit point pointers, thus unified memory address) with TCC drivers.

Notice FSA is enabled on discrete GPUs as well as AMD Fusion.

PS; FSA is the old name for HSA before the trademark claim.

Stop thinking about the consoles i.e. AMD Tahiti GCN ASIC doubles as a server product e.g. AMD Firepro S10000 (similar to Radeon HD 7990). http://www.amd.com/us/products/workstation/graphics/firepro-remote-graphics/S10000/Pages/S10000.aspx

AMD HSA is just AMD's 1st party Torrenza initiative. http://en.wikipedia.org/wiki/Torrenza For old school AMD fans, HSA is a reminder for Torrenza.

Intel Genesseo (2007) http://news.mynavi.jp/articles/2007/10/05/idf/002.html

Notice "shared memory support via I/O MMU".

AMD saying similar statements as Intel...

Intel, AMD and NVIDIA are heading towards the same direction.

HSA creates an improved processor design that exposes the benefits and capabilities of mainstream programmable compute elements, working together seamlessly. With HSA, applications can create data structures in a single unified address space and can initiate work items on the hardware most appropriate for a given task. Sharing data between compute elements is as simple as sending a pointer. Multiple compute tasks can work on the same coherent memory regions, utilizing barriers and atomic memory operations as needed to maintain data synchronization (just as multi-core CPUs do today). No more to say... HSA driver and MS support will not redesign million and million of computers out there and turn them into HSA,to do HSA the CPU and GPU most share 1 common memery address something that is not that way on PC,so unless you use an APU on a PC you are not doing HSA,unless you don't have HSA hardware is not really HSA..

tormentos

Intel has been working on extending PCI-E with shared memory via I/O MMU since 2007.

Notice Intel identifies software overheads (latencies).

I guess your just a newbie with anything X86.

AMD and Intel has been working on HSA's hardware infrastructure since 2006 i.e. soon after Intel Core 2's release.

[QUOTE="faizan_faizan"]

[QUOTE="tormentos"] Exactly never.. I was playing Uncharted 3 again yesterday and i am surprise how good and clean this game look,while also porting some incredible physics,animation and the most incredible element simulation i have ever seen on consoles,the water effect are down right incredible and not only that they even interact with drake you get close to a stream or something like that and drake cloth and his hair even get wet.. I have never see this king of detail and simulations on 360,and to think the PS3 actually had a weaker GPU.tormentos

[QUOTE="Bebi_vegeta"]I don't care if "U" don't need it, it's a feature PC has. Just like SSD, raid HDD will help load the games faster or whatever else that needs data transfering, it's not just for game smart guy.

Again, I never said PC is cheaper for F sake. I don't expect a system that can do everything to be cheaper then a machine that is limited.

Yet, you would still buy a Ferrari if you could and not a God damn Porsche.... and wth is a 2.5 HDD ?

tormentos

So what if RAID is old, it's still great solution for those who want more speed and space.

SSD is great for speed, that's it for now... nobody is buying SSD because they need to store songs or videos.

What's wrong with paying more if I can afford it and I like a PC for multiple purposes ?

There's a few rare games were SSD is actually a benefit for performance in gaming, like ARM 2 and WoW.

http://www.tomshardware.com/reviews/battlefield-rift-ssd,3062-14.html

[QUOTE="tormentos"][QUOTE="faizan_faizan"]What do you mean by Element Simulations? can you show me a vid or something please?

faizan_faizan

If the water is scripted why does it react to drake and if the sand is a texture why does it also react to drake?  It would benefit you to play the game before you randomly bash it.