The AMD Mantle Thread

[Noctifer616]

Nvidia cards have the feature set that Mantle requires yet AMD's 6000 series cards don't?

Someone mentioned bindless textures being a feature that the 6000 series doesn't have but Mantle requires.

 

[psoomah]

Actually, NV is rumored to be sticking an ARM core on future graphics cards to help with batching https://forums.robertsspaceindustries.com/discussion/78571/

'Rumored' doesn't do that justice.

The post was also, strangely ... "Posted: 12:00am Jan 1, 1970".

Not enough there to even speculate from.

 

[blastingcap]

'Rumored' doesn't do that justice.

The post was also, strangely ... "Posted: 12:00am Jan 1, 1970".

Not enough there to even speculate from.

This is the origin of the rumor if you don't like the forum post, and in fact the forum post even linked to it:

http://wccftech.com/alleged-nvidia-...ions-unveiled-project-denver-maxwell-refresh/

"As for the much awaited Project Denver which is supposed to launch with Maxwell will supposedly be debuted with the Maxwell refresh graphic cards. The user reports that NVIDIA will wait for the new FinFET process that is going to be used to prepare Project Denver and only then fuse them on their Maxwell chips. Their’s some truth regarding this information since NVIDIA showcased their Tegra roadmap at GTC 2013 which showcased their Parker SOC featuring NVIDIA Maxwell and Project Denver CPU inside a die which will be manufactured using the new FinFET process."

 

[3DVagabond]

That wall at ~15K batches...that is with AMD's none-multithreaded DX11 driver...right?

They weren't talking vendor specific. He was using a hardware example during the demo talking about a 4core/8thread processor. They are talking what they are capable of getting with DX in absolute terms.

It's really worth the time to listen to the demo.

 

[3DVagabond]

So you don't really know?

It scales just fine, when the GPU vendor can provide a multithreaded DX 11 driver:
http://forums.anandtech.com/showpost.php?p=31520674

This tidbit is interresting:


Double the perfomance...just from making a driver that is multithreaded...why dosn't AMD do this first?
Or would that eat into the PR promises of Mantles performance...and thus make Mantle look less appealing?

How long has this been around and how many games use it? Just adding it to drivers does no good, the game has to support it. I've also read that MTR has issues and results were quite variable. This was from the devs who are behind Mantle so you can chose to take it any way you like.

 

[bystander36]

You have to wonder if the reason it is rarely used is due to it only being supported by 1 vender.

 

[psoomah]

This is the origin of the rumor if you don't like the forum post, and in fact the forum post even linked to it:

http://wccftech.com/alleged-nvidia-...ions-unveiled-project-denver-maxwell-refresh/

"As for the much awaited Project Denver which is supposed to launch with Maxwell will supposedly be debuted with the Maxwell refresh graphic cards. The user reports that NVIDIA will wait for the new FinFET process that is going to be used to prepare Project Denver and only then fuse them on their Maxwell chips. Their’s some truth regarding this information since NVIDIA showcased their Tegra roadmap at GTC 2013 which showcased their Parker SOC featuring NVIDIA Maxwell and Project Denver CPU inside a die which will be manufactured using the new FinFET process."

I followed that link and read the wccftech article but there was no mention of adding an ARM processor to a Maxwell dGPU, that bit apparently came from some other source. I did enjoy reading the rest of the thread however.

Your bolded bit refers to Tegra, which is a mobile SOC, not a dGPU.

 

[3DVagabond]

You prove nothing.

Have you read the entirety of my post, to understand the context of the first sentence.


There is no driver of this kind, pointless for microsoft to test since the driver, since will always work for month on the sames OS with same hardware while games will be ones that would potentially cause problems individually, there is no "updates" to the driver that would fix anything games do with Mantle, there's nothing in there except basic display stuff so it works with windows desktop, there's nothing to update what applications do by them selfs.

Hence the validation layer. But the driver isn't even about directX so AMD doesn't have to tell MS anything, but maybe there will be some kind of sticker "Ready for Windows 7" or something, I don't know, I just know this won't be an issue since the driver is so simple there's no room for it to be buggy.

Yes. I'd like to add to that, that with DX the driver is optimized to the game, or in reality, games. Dozens of them (100's?). I've heard many times people question and complain about the size of the driver D/L. With Mantle the driver is transparent so the devs can see exactly what's going on and the games are optimized to the driver. People need to think about just that one aspect for a minute and how much more sense that actually makes.

 

[blastingcap]

I followed that link and read the wccftech article but there was no mention of adding an ARM processor to a Maxwell dGPU, that bit apparently came from some other source. I did enjoy reading the rest of the thread however.

Your bolded bit refers to Tegra, which is a mobile SOC, not a dGPU.

Actually, it's proof of concept and in fact Tesla is slated to get ARM: http://www.tomshardware.com/news/nvidia-armv8-soc-gpu,18838.html Tesla is the same thing as GeForce but with different drivers and without DP being artificially crippled. It's not a leap to imagine that GeForce will also have ARM cores, if NV perceives a Mantle threat.. so that top to bottom, all NV dGPUs have ARM cores to deal with some or all of the issues that Mantle is supposed to address.

From the Tom's article:

"In an article posted by InfoWorld, the company's chief technology officer for the Tesla product line, Steve Scott, was quoted saying, "Tegra is going to become GPU computing capable in the not-so-distant future. Sometime this decade we are also going to start bringing integrated CPUs and GPUs together in the Tesla line". 'Sometime this decade' is not exactly clear, but we would take a guess that Nvidia is shooting for a release prior to 2015.
Scott was referring to ARMv8 processors, recently announced as Cortex-A53 and A-57 models and a 2014 released date. However, Nvidia was not mentioned by ARM in a row of current ARMv8 licensees, which include AMD, Broadcom, Calxeda, HiSilicon, Samsung and STMicroelectronics. However, the emerging microserver market is an opportunity for a natural evolution of Nvidia's business. Combining Tesla with ARMv8 cores will allow the company to compete in a segment that will be crowded with industry heavyweights such as Samsung, Qualcomm, Texas Instruments - and Intel on the x86 side. Nvidia will also compete with its arch rival AMD in a new market.
The InfoWorld article pointed to Nvidia's Project Denver, which will be based on using the ARMv8 architecture in Nvidia products. According to ARM, the Cortex-A57 processor will be able to provide up to three times the performance of its current flagship, the 32-bit Cortex-A15 design."

 

[bystander36]

Yes. I'd like to add to that, that with DX the driver is optimized to the game, or in reality, games. Dozens of them (100's?). I've heard many times people question and complain about the size of the driver D/L. With Mantle the driver is transparent so the devs can see exactly what's going on and the games are optimized to the driver. People need to think about just that one aspect for a minute and how much more sense that actually makes.

I'm not sure that makes any sense at all. What happens when Mantle drivers get updated, it can throw off all the games designed for a particular version. The drivers would still have to be optimized for each game when Mantle is updated. Then you are still left with what we have now, games designed while using a particular driver, and as new drivers get released, AMD attempts to improve things.

As new GPU's are released, they have to be retroactively made to work with older games.

The only new part is the lower level access.

 

[3DVagabond]

AMD would need NVidia's help to enable Mantle on Kepler cards I think, but who knows if they would actually do that.

Really interesting way of looking at it, "AMD would need nVidia's help"? I don't see AMD doing anything to enable Mantle on nVidia, except hopefully not blocking it. Which is a practice I believe should be outlawed.

 

[3DVagabond]

Yep, good catch. AMD is hiding that even if Mantle makes its way to nv, the benefits will be lesser than with AMD cards.
But then, why would they talk about it, if enabling mantle for nv may never happen?

Why will the benefits be lesser on nVidia cards? Where did you get this from?

 

[Crap Daddy]

Where is mantle?

 

[3DVagabond]

That's what they want according to the presentations.

That's what the devs said they hoped for,"That it would be adopted by other vendors."

 

[(sic)Klown12]

Why will the benefits be lesser on nVidia cards? Where did you get this from?

Most likely due to the fact that Nvidia cards do things differently than GCN, Mantle's target platform, so they would have to have their driver intercept calls and make it into something that works on their architecture. That would add additonal overhead that, depending on specific interactions, would result in lower performance. The only way around that would be for them to create a GPU that exactly matches GCN, something they would never do in their right mind.

 

[3DVagabond]

Can someone answer this question:
What happens when AMD comes up with a new architecture? Will they come up with a newer version of Mantle? And will that make games using older Mantle version unplayable on the new cards?

No. Just like now, with DX compliance, the cards would need to be Mantle compliant.

 

[3DVagabond]

Nvidia cards have the feature set that Mantle requires yet AMD's 6000 series cards don't?

Nobody said that Kepler, or any other arch besides GCN, is currently capable of Mantle support. Simply that nVidia support is possible. Mantle is not purely locked to GCN's arch.

 

[3DVagabond]

You have to wonder if the reason it is rarely used is due to it only being supported by 1 vender.

If that's the case then Mantle will be still born. So far it's looking like Mantle is being adopted at a much faster rate.

 

[3DVagabond]

I'm not sure that makes any sense at all. What happens when Mantle drivers get updated, it can throw off all the games designed for a particular version. The drivers would still have to be optimized for each game when Mantle is updated. Then you are still left with what we have now, games designed while using a particular driver, and as new drivers get released, AMD attempts to improve things.

As new GPU's are released, they have to be retroactively made to work with older games.

The only new part is the lower level access.

From what I understand you are wrong. We'll have to wait and see for sure.

No point in making stuff up now unless you have a source for your position. Someone stating that the Mantle driver needs to be optimized per game or future versions will break old games?

 

[3DVagabond]

Most likely due to the fact that Nvidia cards do things differently than GCN, Mantle's target platform, so they would have to have their driver intercept calls and make it into something that works on their architecture. That would add additonal overhead that, depending on specific interactions, would result in lower performance. The only way around that would be for them to create a GPU that exactly matches GCN, something they would never do in their right mind.

That's not the way I understand it. It doesn't have to be "GCN" at all, simply support the minimum Mantle feature set. It has been reported that it is not GCN specific.

 

[psoomah]

Actually, it's proof of concept and in fact Tesla is slated to get ARM: http://www.tomshardware.com/news/nvidia-armv8-soc-gpu,18838.html Tesla is the same thing as GeForce but with different drivers and without DP being artificially crippled. It's not a leap to imagine that GeForce will also have ARM cores, if NV perceives a Mantle threat.. so that top to bottom, all NV dGPUs have ARM cores to deal with some or all of the issues that Mantle is supposed to address.

I was pointing out your link did not reference Maxwell GPUs containing ARM CPUs. Nor does the TH article you reference, the Nvidia CTO said 'sometime this decade we arealso going to start bringing integrated CPUs and GPUs together in the Tesla line' - which may or may not include Maxwell.

Yes, possibly it will happen with Maxwell and if so probably it will be included in the GeForce line, but there is nothing definitive from the links you've provided so far.

 

[Noctifer616]

That's what the devs said they hoped for,"That it would be adopted by other vendors."

http://www.youtube.com/watch?v=tDPgJB2x7dQ

Around the 2 min mark. He says that Mantle is designed for feature generation of AMD graphics architectures and even competitors cards.

 

[psoomah]

Why will the benefits be lesser on nVidia cards? Where did you get this from?

Mantle exists to leverage console code to PCs. Console code will access the full GCN feature set so GCN GPUs can fully utilize that code whereas Nvidia GPUs cannot as it doesn't contain the full GCN feature set.

 

[3DVagabond]

Mantle exists to leverage console code to PCs. Console code will access the full GCN feature set so GCN GPUs can fully utilize that code whereas Nvidia GPUs cannot as it doesn't contain the full GCN feature set.

You aren't looking at this correctly. nVidia will need to support Mantle's feature set, just like they now need to support DX and OpenGL feature sets.


What features are missing from nVidia GPU's? Is this known?

 

[psoomah]

If that's the case then Mantle will be still born. So far it's looking like Mantle is being adopted at a much faster rate.

That may have been a reference to Nvidia's multi-threading capability/drivers.