Radeon Software 18.3.3 snags Vulkan 1.1 support on AMD GCN GPUs

[raphidy]

57 minutes ago, Jito463 said:

I've yet to see what you find so laughable about that list.  At least 6 of those would qualify as AAA games right off the bat (albeit, one is only supporting it on Linux), and another 4 are still very popular games regardless of their AAA status.  The remaining 4 (which, by the way, adds up to 14, not 13) I've either not heard of or they're still ports of actual games, even if they're older ones.

At least 6, which half was probably pushed by AMD. Doom and Wolfenstein are very good implementation of Vulkan.

 

I am still hoping for DICE to implement Vulkan on BF games instead of DX12. I mean, EA pushed mantle but not vulkan, so weird. It seems like developers mostly port opengl games to vulkan to benefit modern hardware/ or compatibility with linux but that's it.

[Valentyn]

28 minutes ago, BuckGup said:

Has trouble running math computations in research which it's advertised for so yeah 

Shhhhhh... no need for that now.

[DrMacintosh]

Does this mean I’ll have another reason not to get rid of my R9 380?  

 

#stillbetterthana1050ti

[Humbug]

That is pretty great: if you bought a gcn 1.0 graphics card back in 2012 you get Vulkan 1.1 today.

 

When these graphics cards came out mantle had not even been announced! Let alone dx12 or Vulkan...

[Doobeedoo]

Nice. My R9 290 still holds well so far, so no rush to upgrade yet. 

[Humbug]

4 hours ago, AluminiumTech said:

I wonder when AMD will stop providing optimizations and new features for GCN 1st Gen considering the equivalent Nvidia cards are (basically) on life support.

It's pretty interesting also to follow the AMD devs with their Linux open source graphics driver development. Because you gain a lot of insights as to how things are being prioritized, whereas on windows everything happens behind closed doors and is suddenly released.

 

You can see by watching the Linux news  closely that it is actually a massive amount of work for AMD to keep pushing all these new features to gcn 1.0 and gcn 1.1. The drivers diverge quite a bit.

 

It's been impressive to watch on Linux they tend to push out new features first to Polaris and Vega because those are the GPUs now on sale, but then later more patches pop up to get things working on older gcn too, as long as the hardware is capable.

[Humbug]

9 minutes ago, Doobeedoo said:

Nice. My R9 290 still holds well so far, so no rush to upgrade yet. 

My r9 290 vapor x is epic too. Sapphire did a great job : it's fast, cool and quiet. Great overclock headroom too.

 

Besides not like we have much choice than to stick with it anyway LOL. The gtx 1060 and rx 580 are only marginally faster than us. And the cards that are properly fast enough right now to tempt an upgrade cost a fortune.

 

I just need my 290 to not die, cause my 3 year warranty is up, so if something happens to it I will have to pick up a gtx 1070 or vega56: but they are so expensive.

[M300843499]

3 hours ago, i_build_nanosuits said:

these are the 13 games you talking about bro?! ahhh yeah...those 13 AAA titles right...laughable, come on!

 

I don't understand the subtle anger in this post.

Vulkan is new much like DX12, DX12 in fact has a similar amount of titles / years being used.

 

We all laughed at DX11 when it first arrived and nobody cared for it, i believe one of the very first games to use some DX11 features was Bad Company 2 and many people were still buying up GTX 280 / 4890 back then thinking their card will remain relevant.

 

 

On topic, it's great to see such brilliant support for GCN overall by AMD.

[M300843499]

9 minutes ago, Humbug said:

My r9 290 vapor x is epic too. Sapphire did a great job : it's fast, cool and quiet. Great overclock headroom too.

 

Besides not like we have much choice than to stick with it anyway LOL. The gtx 1060 and rx 580 are only marginally faster than us. And the cards that are properly fast enough right now to tempt an upgrade cost a fortune.

 

I just need my 290 to not die, cause my 3 year warranty is up, so if something happens to it I will have to pick up a gtx 1070 or vega56: but they are so expensive.

R9 Fury <3

[Doobeedoo]

6 minutes ago, Humbug said:

My r9 290 vapor x is epic too. Sapphire did a great job : it's fast, cool and quiet. Great overclock headroom too.

 

Besides not like we have much choice than to stick with it anyway LOL. The gtx 1060 and rx 580 are only marginally faster than us. And the cards that are properly fast enough right now to tempt an upgrade cost a fortune.

 

I just need my 290 to not die, cause my 3 year warranty is up, so if something happens to it I will have to pick up a gtx 1070 or vega56: but they are so expensive.

Yeah that's an awesome card.

When Vega launched I was tentative for upgrade even if it was it's msrp but prices increased and no availability at all,specially custom cards. 

 

I guess Navi next to look for. Better upgrade too. 

[Humbug]

1 hour ago, Doobeedoo said:

I guess Navi next to look for. Better upgrade too. 

At least with Navi there will less of the vega64 paradigm of trying to beat the corresponding Nvidia part by using high stock clockspeeds and high stock voltages. It will now be about scalability and going wide via infinity fabric... Basically the threadripper of GPUs.

 

Navi will be on next gen 7nm. We can only hope that TSMC and Global Foundries plan their next generation fabrication plants with enough capacity to cater to both miners and gamers.

 

Cause AMD cards are normally good at compute. Add to that the whole Navi philosophy of making a powerful graphics card via infinity fabric scalability using multiple modestly clocked efficient chips to go wide sounds like a miner's wet dream on paper. We can only hope the fabs are ready and the memory supply is fixed by then.

[Doobeedoo]

9 minutes ago, Humbug said:

At least with Navi there will less of the vega64 paradigm of trying to beat the corresponding Nvidia part by using high stock clockspeeds and high stock voltages. It will now be about scalability and going wide via infinity fabric... Basically the threadripper of GPUs.

 

Navi will be on next gen 7nm. We can only hope that TSMC and Global Foundries plan their next generation fabrication plants with enough capacity to cater to both miners and gamers.

 

Cause AMD cards are normally good at compute. Add to that the whole Navi philosophy of making a powerful graphics card via infinity fabric scalability using multiple modestly clocked efficient chips to go wide sounds like a miner's wet dream on paper. We can only hope the fabs are ready and the memory supply is fixed by then.

Yeah, new architecture and with 7nm also along with that infinity fabric scalability, it will be awesome to see this. Really wonder how this will translate on GPU side like it is on CPU now. It will definitely be cheaper to produce such dies. As far as memory and specially graphics cards in general they should go down in prices by then. 

[M300843499]

16 minutes ago, Humbug said:

At least with Navi there will less of the vega64 paradigm of trying to beat the corresponding Nvidia part by using high stock clockspeeds and high stock voltages. It will now be about scalability and going wide via infinity fabric... Basically the threadripper of GPUs.

 

Navi will be on next gen 7nm. We can only hope that TSMC and Global Foundries plan their next generation fabrication plants with enough capacity to cater to both miners and gamers.

 

Cause AMD cards are normally good at compute. Add to that the whole Navi philosophy of making a powerful graphics card via infinity fabric scalability using multiple modestly clocked efficient chips to go wide sounds like a miner's wet dream on paper. We can only hope the fabs are ready and the memory supply is fixed by then.

Before GCN arrived, Fermi trampled on anything AMD had with Fermi, but that was nVidia's last hurrah in regards to real compute power.

[Jito463]

3 hours ago, Terryv said:

I didn't count X4 because it's still unreleased.

Fair enough, I counted it because it's on Steam and they have pre-release video footage there (plus, it says 2018, which is this year).

[Trixanity]

27 minutes ago, Humbug said:

Navi will be on next gen 7nm. We can only hope that TSMC and Global Foundries plan their next generation fabrication plants with enough capacity to cater to both miners and gamers.

 

Cause AMD cards are normally good at compute. Add to that the whole Navi philosophy of making a powerful graphics card via infinity fabric scalability using multiple modestly clocked efficient chips to go wide sounds like a miner's wet dream on paper. We can only hope the fabs are ready and the memory supply is fixed by then.

There is no issue producing GPUs right now. It's only memory. Of course that doesn't mean 7nm couldn't spell trouble but if it's similar to now, then it'll still be bottlenecked only by memory production.

[hobobobo]

4 hours ago, Valentyn said:

That's only if they can get scaling as well as Ryzen.

It also means Vulkan and DX12 multi GPU support needs to be there likely also.

 

Lots of if's, and possibilities. Sadly AMD's enemy is time and money; neither of which they much of at the moment. ( fingers crossed Epyc get some proper sales )

 

They also a proper process; that isn't hamstrung by Global Foundries. IBM can manage 5Ghz on their 7nm for base speeds.

I'm very concerned GloFo won't be able to match that; their process and fabs are always little behind TSMC quality.

The process is a real concern, new GloFo CEO said in the recent anandtech interview that navi will the the pipecleaner for 7nm. Good news is zen2 is going to be on TSMC 7nm.

Also, speed is not only process dependant, IBM, im pretty sure, said they designed their chips to run at 5ghz stock

 

As far as i remember, Raja or some1 else in the old interview said that navi would need no extra API support, APIs would see it as 1 GPU

[Humbug]

21 minutes ago, hobobobo said:

As far as i remember, Raja or some1 else in the old interview said that navi would need no extra API support, APIs would see it as 1 GPU

Yep. As far as the OS is concerned it is one GPU. The driver+firmware does everything via infinity fabric. Which obviously will be a faster and more mature technology by the time of Navi.

 

It's like threadripper. Windows just sees it as one multicore CPU, no visibility into the infinity fabric interconnects etc.

 

If Navi had relied on API programming for scalability that would be a disaster. Cause we have seen how bad devs are at supporting multi GPU.

 

As an enthisiast Navi is very interesting and novel. It may be awesome, it also may suck as the first architecture with this approach. We can't really predict anything cause nobody has tried this before on GPUs.

[Trixanity]

11 minutes ago, hobobobo said:

The process is a real concern, new GloFo CEO said in the recent anandtech interview that navi will the the pipecleaner for 7nm. Good news is zen2 is going to be on TSMC 7nm.

Also, speed is not only process dependant, IBM, im pretty sure, said they designed their chips to run at 5ghz stock

 

As far as i remember, Raja or some1 else in the old interview said that navi would need no extra API support, APIs would see it as 1 GPU

Vega on 7nm will be the pipe cleaner. Zen2 will not be on TSMC. Where are you getting all this? 

 

AMD has talked about splitting their orders between GloFo and TSMC based on product lines. There may be some Zen products made by TSMC but it's more likely that it'll be GPU products. No details on that though. Given the teams and experience working at either place, I'm pretty confident AMD will give GloFo full custody of their CPU business. It's after all IBM people in charge of that and they're going specifically for high frequency. 

[hobobobo]

13 minutes ago, Trixanity said:

Vega on 7nm will be the pipe cleaner. Zen2 will not be on TSMC. Where are you getting all this? 

 

AMD has talked about splitting their orders between GloFo and TSMC based on product lines. There may be some Zen products made by TSMC but it's more likely that it'll be GPU products. No details on that though. Given the teams and experience working at either place, I'm pretty confident AMD will give GloFo full custody of their CPU business. It's after all IBM people in charge of that and they're going specifically for high frequency. 

Oh yeah, vega 7nm, kinda forgot about it since 12nm got cancelled. I honestly cant remember, but seems logical with Vega being pipecleaner, previous GloFo fuck up with 14nm, TSMC track record for new processes and the emphasis on the CPU side. If i find where i read it, ill edit it in. Perhaps it was Jim's video, if it was - its just speculation.

[Trixanity]

1 hour ago, hobobobo said:

Oh yeah, vega 7nm, kinda forgot about it since 12nm got cancelled. I honestly cant remember, but seems logical with Vega being pipecleaner, previous GloFo fuck up with 14nm, TSMC track record for new processes and the emphasis on the CPU side. If i find where i read it, ill edit it in. Perhaps it was Jim's video, if it was - its just speculation.

TSMC doesn't have any experience with high performance x86 CPUs at 4+ GHz as far as I know. The 14nm process was licensed from Samsung and tweaked to better fit desktop processors (unless you refer to the initial 14nm process GloFo scrapped). The 7nm was made by IBM and sold along with their teams and facilities to GloFo and the key people working on it are from IBM. It seems like GloFo is getting on top of things. There is reason to believe they could be dominating certain areas within a couple of years.

 

The only info we have is a brief answer in an interview with Lisa Su 

Quote

So in 7nm, we will use both TSMC and GlobalFoundries. We are working closely with both foundry partners, and will have different product lines for each. I am very confident that the process technology will be stable and capable for what we’re trying to do.

That comment was made after talking about GPUs and bringing Vega to 7nm so there is no evidence pointing to TSMC doing CPUs. Given AMD's CPU strategy they're only doing two (perhaps three this time) chip designs and scale through infinity fabric. It would be troublesome to do designs on different nodes. It's easier to give GloFo the mainstream GPUs and then hand TSMC the enthusiast GPUs. That's something TSMC does well although we don't know if GloFo and their 7nm process handles GPUs better than previous efforts. The 14nm GPUs was GloFo's first rodeo in that regard I believe.

 

It's clear that TSMC got a better handle on GPUs as they've been at it for many years and that the node licensed from Samsung was hard to salvage for high performance desktop chips.

 

That may change or it may not. Time will tell.

[AresKrieger]

It's rather funny to I've had reason to want to like AMD but their products always die on me so its rather a mute point, either way I do hope Vulcan succeeds despite preferring Nvidia I despise windows 10 thus any good news for vulcan is much approved by me.

[M300843499]

9 minutes ago, AresKrieger said:

It's rather funny to I've had reason to want to like AMD but their products always die on me so its rather a mute point, either way I do hope Vulcan succeeds despite preferring Nvidia I despise windows 10 thus any good news for vulcan is much approved by me.

Well i hate both companies, Radeon 7790 died on me, GTX 480 died on me.

[ivan134]

10 hours ago, Terryv said:

Did you even read what I wrote. Adoption of a new API takes time. not to mention that DX9-10-11 each have their share of shitty games, lots of them in fact. DX11 released in 2009, Its time we move on to something new.

 

Just because you don't like the current vulkan titles doesn't make them "not games"

You're new here. He's a fanboy who hates anything that makes AMD look good.

[Terryv]

39 minutes ago, ivan134 said:

You're new here. He's a fanboy who hates anything that makes AMD look good.

understood, thanks for the heads up

[leadeater]

10 hours ago, Nicnac said:

Sympathy points for AMD here! what a shame I still have a HD 6970 lying around..... ^^

My friend is still using my old 6970 now, surprising that it can actually run stuff rather well just not on ultra obviously or above 1080p.