How many ACEs does Maxwell 2 have?

[Headfoot]

I don't get the impression NV is being quiet on VR or DX12. The marketing material might be a little less specific but its all still there

 

[VR Enthusiast]

Nvidia is being quiet on VR by their standards but it's because they've got nothing much to talk about yet. As Zlatan says, they don't have the hardware (no "real" ACE's) and don't have the software.

They should have the hardware with Pascal but matching Mantle is another thing entirely. AMD's lead in VR is assured for the next 2 years I would say.

It's also important to realise that AMD hardware does have more market share than Nvidia and the game developers have likely switched to GCN.

 

[sontin]

Seriously, what?
They are talking about VR:
Siggraph: http://www.nvidia.com/object/siggraph2015-best-gtc.html
Gamescom: https://www.youtube.com/watch?v=sN-qxmb57uU
Computex: http://www.anandtech.com/show/9305/nvidia-announces-gameworks-vr-branding-adds-multires-shading

 

[3DVagabond]

Nvidia is being quiet on VR by their standards but it's because they've got nothing much to talk about yet. As Zlatan says, they don't have the hardware (no "real" ACE's) and don't have the software.

They should have the hardware with Pascal but matching Mantle is another thing entirely. AMD's lead in VR is assured for the next 2 years I would say.

It's also important to realise that AMD hardware does have more market share than Nvidia and the game developers have likely switched to GCN.

nVidia doesn't operate on the leading edge. They wait until the market is profitable and then jump in. If possible with something they can make proprietary and charge an added premium to their customers for. It works well for them and their customers seem to be OK with it.

 

[VR Enthusiast]

Seriously, what?
They are talking about VR:
Siggraph: http://www.nvidia.com/object/siggraph2015-best-gtc.html
Gamescom: https://www.youtube.com/watch?v=sN-qxmb57uU
Computex: http://www.anandtech.com/show/9305/nvidia-announces-gameworks-vr-branding-adds-multires-shading

That's not even a 10th of what is available on LiquidVR. What is more telling are the companies outside of AMD that are talking about LiquidVR, ie Crytek and GE.

https://www.youtube.com/watch?v=fLlzb9KEbDY

https://www.youtube.com/watch?v=ylCIijJDTlU

nVidia doesn't operate on the leading edge. They wait until the market is profitable and then jump in. If possible with something they can make proprietary and charge an added premium to their customers for. It works well for them and their customers seem to be OK with it.

They've missed a trick with VR and now everybody who is anybody is making games on LiquidVR instead. I think this is the big difference this time around - we're not just looking at short-term AMD gains here, it's over a period of years at least. I think even the most avid Nvidia fans will tire of being second best over a long duration.

The biggest thing about VR is the latency and the late-latching that AMD has with Mantle is a huge lead. It's hard to understand the difference until you see both side-by-side, but it will be seen in benchmarks. It will be interesting to see how Nvidia counters.

 

[3DVagabond]

They've missed a trick with VR and now everybody who is anybody is making games on LiquidVR instead. I think this is the big difference this time around - we're not just looking at short-term AMD gains here, it's over a period of years at least. I think even the most avid Nvidia fans will tire of being second best over a long duration.

The biggest thing about VR is the latency and the late-latching that AMD has with Mantle is a huge lead. It's hard to understand the difference until you see both side-by-side, but it will be seen in benchmarks. It will be interesting to see how Nvidia counters.

Right now... today, there is no money to be made from it. It's purely an expense. I will be shocked if AMD doesn't do like they always do and hand everything they've done over for the greater industry to use. nVidia will get involved if they need to manipulate anything for their benefit. Like adding one or two features that their hardware supports and AMD's doesn't and claiming a coupe. Other than that, they'll just join in when there's money to be made. When they do join in they will announce that what they have is better, the masses will nod in agreement, and it's business as usual.

 

[VR Enthusiast]

Right now... today, there is no money to be made from it. It's purely an expense. I will be shocked if AMD doesn't do like they always do and hand everything they've done over for the greater industry to use. nVidia will get involved if they need to manipulate anything for their benefit. Like adding one or two features that their hardware supports and AMD's doesn't and claiming a coupe. Other than that, they'll just join in when there's money to be made. When they do join in they will announce that what they have is better, the masses will nod in agreement, and it's business as usual.

This could conceivably happen in a couple of years but AMD's advantages are real. I think VR will really have taken off within a year and it's probably too early for Nvidia.

Don't get me wrong though, I still expect Nvidia will sell more cards for VR than AMD but the gap will really close. I think even if AMD gets close to 50% they'll be doing well but I can't see how the better informed enthusiasts won't shift over to them, assuming they actually care about their VR experience.

 

[3DVagabond]

This could conceivably happen in a couple of years but AMD's advantages are real. I think VR will really have taken off within a year and it's probably too early for Nvidia.

Don't get me wrong though, I still expect Nvidia will sell more cards for VR than AMD but the gap will really close. I think even if AMD gets close to 50% they'll be doing well but I can't see how the better informed enthusiasts won't shift over to them, assuming they actually care about their VR experience.

This is the key, I think. For example, when DX11 first came along nVidia was nowhere to be seen. They simply marketed it as it's not important. There's not enough DX11 games to matter. Then, when they released their DX11 card and it was faster (especially with tessellation), they were all over it as the most important metric and their customers were all of a sudden on board. Even if it was just one or two games that it mattered with.

When VR comes along and IF AMD has an advantage nVidia will simply down play it as not important. It's early days and you (the consumer) shouldn't worry about it. Just keep playing your older games. nVidia will dig into the nuts and bolts and find some advantage their hardware has. They will then sponsor a game to leverage that and every site will do a deep revealing tech article showing nVidia's advantage and how from this day forward you should be really concerned. Because we're only seeing games just starting to use this feature (think tessellation again) and with every new release AMD will fall further behind. You'll be really sorry if you don't base your buying decision on this one aspect of performance. Every site will add it to their benchmark suite. Every site will talk at length about it.

Look at sites like TPU, for example. They used to put "doesn't support PhysX" (and still have the check boxes for it and CUDA in GPUz) as a negative on AMD reviews. When AMD had DX11 before nVidia they used to put "lack of DX11 support in games" (or words to that effect) as a negative for AMD cards. Like being DX11 capable was a bad thing.

Not only does AMD have to compete at the hardware level with nVidia they also have to fight the marketing and they tend to not do too well there.

 

[Silverforce11]

VR isn't a big deal because nothing good is in the immediate horizon. Also I worry devs may have overestimated the enthusiasm for it, I mean as long term gamers here, do many of us enjoy playing games with a helmet on? It's such a major shift in how we game, the change factor may be off-putting.

What I noticed is that the hype for DX12/VR comes from gamedev chatter on twitter, reddit, tech forums, conferences etc. It's not like DX11 where the was little excitement coming from those who make games for a living. Because of this, I know it will be a game changer.

 

[Red Hawk]

As others have pointed out, Nvidia seems to have a fundamentally different approach to asynchronous compute, at least for now. They don't seem to have multiple fixed function "asynchronous compute engines" like AMD does. It's impossible to say for sure which one will have a practical advantage until more true DX12 games come out; I'm a little inclined to bet on AMD, though, because consoles.

One of the talking points AMD has brought up about asynchronous compute is that it will be easy for developers to get a handle on because they're already using it in console games -- console games that run on AMD GCN architecture. I would love to find out just how many ACEs the console graphics chips have. The AMD chips closest to the custom APUs in the PS4 and Xbox One are Pitcairn and Bonaire, which each have two ACEs. Since they're custom chips though, AMD might have snuck in extra ACEs, particularly on the PS4. All GCN 1.0 chips have just 2 ACEs, but the Hawaii, which is GCN 1.1, has 8. The console chips are both most similar to GCN 1.1, from what I've heard, so a GCN 1.1 equivalent to Pitcairn might be designed for more ACEs. Just a guess though.

Who knows? Maybe asynchronous compute will be to AMD what tessellation was/is to Nvidia, a key advantage thanks to a superior hardware approach. Except while tessellation just came down to an easily disabled visual effect, asynchronous compute seems like it will have broader effects on general performance of graphics cards in games.

 

[Silverforce11]

By the time more games heavily use async compute, it's timeframe is what, around late 2016?

We have a few on the way but I'm not inclined to think they will really push DX12.

So the current hardware differences won't matter so much (it matters some, for sure) as Artic Island vs Pascal.

 

[Red Hawk]

By the time more games heavily use async compute, it's timeframe is what, around late 2016?

We have a few on the way but I'm not inclined to think they will really push DX12.

So the current hardware differences won't matter so much (it matters some, for sure) as Artic Island vs Pascal.

The architecture differences could still carry forward. Going with the tessellation example, AMD progressively improved its tessellation performance with innovations like multiple geometry engines, off-chip buffering, etc., but Nvidia's approach is so fundamentally better that it still holds the lead over AMD now around 5 years after tessellation's debut.

 

[raghu78]

One of the talking points AMD has brought up about asynchronous compute is that it will be easy for developers to get a handle on because they're already using it in console games -- console games that run on AMD GCN architecture. I would love to find out just how many ACEs the console graphics chips have.

Its well known that PS4 has 8 ACE with 8 queues each for a total of 64 queues , similar to R9 290X (Hawaii), while Xbox One is limited to 2 ACE with 8 queues each similar to HD 7790(Bonaire).

https://en.wikipedia.org/wiki/PlayStation_4_technical_specifications#APU
http://forums.anandtech.com/showthread.php?t=2316133

Thirdly, said Cerny, "The original AMD GCN architecture allowed for one source of graphics commands, and two sources of compute commands. For PS4, we’ve worked with AMD to increase the limit to 64 sources of compute commands — the idea is if you have some asynchronous compute you want to perform, you put commands in one of these 64 queues, and then there are multiple levels of arbitration in the hardware to determine what runs, how it runs, and when it runs, alongside the graphics that’s in the system."

http://www.tomshardware.com/reviews/microsoft-xbox-one-console-review,3681-4.html

http://www.dailytech.com/Under+the+...+PS4+1+TB+Ultimate+Player+Ed/article37409.htm

"The Xbox One has only 2 asynchronous compute engines (ACEs), where as the PS4 has 8 ACEs that can work in parallel with the GPU."

http://www.anandtech.com/show/6837/...w-feat-sapphire-the-first-desktop-sea-islands

http://www.anandtech.com/show/7457/the-radeon-r9-290x-review/2

 

[Headfoot]

The AMD chips closest to the custom APUs in the PS4 and Xbox One are Pitcairn and Bonaire, which each have two ACEs. Since they're custom chips though, AMD might have snuck in extra ACEs, particularly on the PS4. All GCN 1.0 chips have just 2 ACEs, but the Hawaii, which is GCN 1.1, has 8. The console chips are both most similar to GCN 1.1, from what I've heard, so a GCN 1.1 equivalent to Pitcairn might be designed for more ACEs. Just a guess though.

Sony specifically asked for more ACEs and got 8 in the PS4. I think AMD, as they were building that and GCN 1.1 revisions, realized it was a good move and now we've got more in all new AMD chips as well as a result. The Xbox1 has fewer though, 2 aces with 8 queues each if I recall correctly?

edit rahgu has got the links.

 

[Red Hawk]

Its well known that PS4 has 8 ACE with 8 queues each for a total of 64 queues , similar to R9 290X (Hawaii), while Xbox One is limited to 2 ACE with 8 queues each similar to HD 7790(Bonaire).

https://en.wikipedia.org/wiki/PlayStation_4_technical_specifications#APU
http://forums.anandtech.com/showthread.php?t=2316133

Thirdly, said Cerny, "The original AMD GCN architecture allowed for one source of graphics commands, and two sources of compute commands. For PS4, we’ve worked with AMD to increase the limit to 64 sources of compute commands — the idea is if you have some asynchronous compute you want to perform, you put commands in one of these 64 queues, and then there are multiple levels of arbitration in the hardware to determine what runs, how it runs, and when it runs, alongside the graphics that’s in the system."

http://www.tomshardware.com/reviews/microsoft-xbox-one-console-review,3681-4.html

http://www.dailytech.com/Under+the+...+PS4+1+TB+Ultimate+Player+Ed/article37409.htm

"The Xbox One has only 2 asynchronous compute engines (ACEs), where as the PS4 has 8 ACEs that can work in parallel with the GPU."

http://www.anandtech.com/show/6837/...w-feat-sapphire-the-first-desktop-sea-islands

http://www.anandtech.com/show/7457/the-radeon-r9-290x-review/2

Thanks! So the PS4 has 8 ACEs, very interesting. While it still has less raw power than a Pitcairn or Tahiti based GCN chip, seems like it could still hold that advantage over them.

 

[dogen1]

By the time more games heavily use async compute, it's timeframe is what, around late 2016?

I don't see why early DX12 games wouldn't use it. Everything that supports DX12 supports multi-engine, and most games already use a lot of compute shaders. Maybe I'm overlooking something, but it seems like a no-brainer.

 

[railven]

VR isn't a big deal because nothing good is in the immediate horizon. Also I worry devs may have overestimated the enthusiasm for it, I mean as long term gamers here, do many of us enjoy playing games with a helmet on? It's such a major shift in how we game, the change factor may be off-putting.

What I noticed is that the hype for DX12/VR comes from gamedev chatter on twitter, reddit, tech forums, conferences etc. It's not like DX11 where the was little excitement coming from those who make games for a living. Because of this, I know it will be a game changer.

Gonna side with this. Throw in the cost of entrance and you've isolated a good portion of the PC user base.

The console side worries me, they've put up with inferior attempts of techniques for years. Some how even I don't see them putting up with the rubbish they'll get which will lead to a huge downfall in the audience for these new techs. Think back to Wii-sensor and Kinect. Huge initial interest, so huge both companies bet the farm on it only for both techs to fail as the core audience of gamers didn't latch on.

EDIT: Raghu's post above ties into what I mean. With the two leading platforms so lopsided, a dev commits sales suicide if they show favoritism to one side unless they go exclusive. And I doubt MSFT would permit/allow for devs to show the PS4 in a better light on multi-plat games (we've seen MSFT use it's weight/influence to push some devs for parity on some games.)

 

[Carfax83]

Lots of good information in this thread, thanks to all who contributed :thumbsup:

Here's some good information about Hyper-Q in an article that Ryan wrote almost 3 years ago about GK110 based Tesla:

Sometimes the simplest things can be the most powerful things, and this is very much the case for Hyper-Q. Simply put, Hyper-Q expands the number of hardware work queues from 1 on GF100 to 32 on GK110. The significance of this being that having 1 work queue meant that GF100 could be under occupied at times (that is, hardware units were left without work to do) if there wasn’t enough work in that queue to fill every SM or if there were dependency issues, even with parallel kernels in play. By having 32 work queues to select from, GK110 can in many circumstances achieve higher utilization by being able to put different program streams on what would otherwise be an idle SMX.

And Dynamic Parallelism:

Dynamic Parallelism is NVIDIA’s name for the ability for kernels to be able to dispatch other kernels. With Fermi only the CPU could dispatch a new kernel, which incurs a certain amount of overhead by having to communicate back and forth with the CPU. By giving kernels the ability to dispatch their own child kernels, GK110 can both save time by not having to go back to the GPU, and in the process free up the CPU to work on other tasks.

 

[Red Hawk]

I remember reading in another Anandtech article that pre-Maxwell 2 HyperQ was not compatible with DirectX 12 Asynchronous Compute, though.

 

[Sabrewings]

VR isn't a big deal because nothing good is in the immediate horizon. Also I worry devs may have overestimated the enthusiasm for it, I mean as long term gamers here, do many of us enjoy playing games with a helmet on? It's such a major shift in how we game, the change factor may be off-putting.

Everyone I know that is into Star Citizen, Elite, or any other space sim as well as racing enthusiasts are all waiting for VR headsets to hit the commercial market, myself included. I know several (again, myself included) who plan to own the Oculus, SteamVR, and Hololens. I add to that with plans for a StarVR when they release.

Ever since I had my first actual VR experience with DK2, I've known it'll be a hit. It was that impressive, memorable, and lasting of an experience. Anyone who has not had one on their head simply can't understand. I didn't before.

 

[Silverforce11]

Everyone I know that is into Star Citizen, Elite, or any other space sim as well as racing enthusiasts are all waiting for VR headsets to hit the commercial market, myself included. I know several (again, myself included) who plan to own the Oculus, SteamVR, and Hololens. I add to that with plans for a StarVR when they release.

Ever since I had my first actual VR experience with DK2, I've known it'll be a hit. It was that impressive, memorable, and lasting of an experience. Anyone who has not had one on their head simply can't understand. I didn't before.

I reserve judgement until its out and I can play with it. It's a concern I have since its a revolutionary way to play games.

 

[railven]

I reserve judgement until its out and I can play with it. It's a concern I have since its a revolutionary way to play games.

What, the Wii-mote didn't sell you on the experience of a new way to play games? Or Kinect?

 

[Silverforce11]

What, the Wii-mote didn't sell you on the experience of a new way to play games? Or Kinect?

It did. I REGRETTED IT. Wii sat collecting dust real fast.

 

[casiofx]

There is the graphics market and the gaming market

Most of those intel igp's are not part of the gaming market.

For instance the devs of Witcher 3 really didn't care about intl igps. No igp can play witcher 3 at acceptable framerates.

You are sure that no iGPU can play The Witcher 3 at acceptable framerates?:sneaky:

https://www.youtube.com/watch?v=WvUu-MHZxMg

 

[Sabrewings]

What, the Wii-mote didn't sell you on the experience of a new way to play games? Or Kinect?

The problem with those examples is that they offered new ways (though extremely imprecise) of interacting and offering input, but relied on the same old same old for output to the user. The point of VR is to encompass the user's field of view and in effect transport them somewhere else.

VR is not reliant in any way on the 3D control mechanisms to achieve its goals. Sitting back in a chair with a VR headset and a HOTAS for a flight sim is enough to get your heart racing as you zip past asteroids with only a meter to spare. All VR has to do is provide enough of an environment to let the brain's natural abilities to fill the gaps take over and your perspective suddenly changes drastically.

I'd be perfectly happy playing a shooter with a mouse and keyboard. Same controls we've always used, except now the mouse is simply used to control the torso's direction, not the head. The head can be controlled by the VR headset and now you're there.

Every demo I've tried has achieved this sense of presence, and most didn't use anything but controls people already use (if it offered any controls at all).