Next Gen AMD GPU's - 7nm VEGA refresh & AMD 600 series Project "Zen"

General Forums Latest News
21

No, AMD GPUs never been ahead of Nvidia because of CUDA and the fact that AMD’s OpenCL drivers/implementation always been a big pile of garbage. You don’t have better compute if no one can use it effectively. Ask Luxrender devs the amount of hacking that AMD’s OpenCL required to make their software work.

Most GPGPU software is written in CUDA because of this. It is beyond me why people say that AMD has more TFLOPs or whatever when there is no software that can take advantage of it.

Nvidia is light years ahead in compute. That’s why they can charge $10K for a Tesla P100. That’s why they hold a virtual 100% of the GPU computing market with an iron fist.

22

Well that’s a load of… delusional text. Would you care to support those beliefs.

IIRC both suck and both are great from what i experienced with use, with LuxCore team & users one their forums and other OCL users and abusers elsewhere.

NV can charge because it’s high tech in high demand and in low production… 10k is peanuts with the right mind behind the board. It is as it is because it’s dumb tech. And you’ll soon find out.

Why do i even bother…

23

Regarding the GPP once again; AMD is not letting it stop them from getting Radeons into the hands of consumers (even amid the mining craze).

Their new “Combat Crates” actually uses the new Armor name from MSI (so they’re using a vendor in the GPP).

To note, a concept known as bundling has been used before to try to fight off the miners, but this is perhaps the first time the GPU came with the guts for a whole PC (these crates do contain first generation Ryzens, but you can’t fault AMD for trying to unload what remains after the manufacturing was stopped).

24

I recommend you look at Blenders own tests. as i posted a few times and is part of my signature. Vega 64 is slightly behind GTX 1080ti, and far ahead of GTX 1080. So your point is invalid Esparadrapo

25

You said “have always been ahead” which is false.

AMD didn’t work with Cycles until recently. After that it took a long time to achieve feature parity and even then the performance wasn’t there till much later.

I’m worried you’re developing Alzheimer.

26

AMD worked properly for some time now in Blender, ever since AMD assisted in the kernel split. And in other engines for far longer (eg. LuxRender).

Additionally if you plan to QUOTE me, please include the key part of the sentence and not just a small part.

I said, ahead in COMPUTE, not ahead in Blender.

Nvidia supported Blender for long time, supplying GPU’s and assisting in code. AMD did nothign for a long time. Only untill the AMD kernel split project few years back. After that AMD GPu’s worked, and very quickly Blender Team managed to make Radeons outperform the GTX. This is simply part of the development work needed.

Now as for what i quoted as “always”, relating to COMPUTE activities. You can look at miners as prime example of computer power. GTX cards weren’t able to compete. Only recently there were some cryptocoins where GTX was finally useful at mining.

Finally, OpneCL was developed by Apple. And even though i dont’ like Apple, they usually dont’ place crappy hardware in their devices. Ok they did, but that was a Nvidia card and their issue with solder used making their GPU’s fail at a… nearly catastrophic rate (big lawsuit, fun read )… Ever since that AMD became primary supplier, and their OpenCL support exploded (in a positive way)

Now for Gaming, I won’t dispute, that generally Nvidia is ahead or equal (for most generations - excluding the GTX 5xx - don’t recall other gens).

Currently Nvidia simply doesn’t feel the pressure to properly support OpneCL, as they currently own most of the compute market.

Lack of competition does that. I hope that the new 7nm and the AMD 600 (radeon zen) will bring some needed competition like AMD did with the CPU market. That is the point of this post.

Now you are free to disagree, and I’m actually eager to better understand what is your opinion but I need more details/evidence

27

Err… There are plenty of sites out there where you can partake in the endless battle between Nvidia and AMD fans.

What I do know is that if AMD wants to catch back up with Nvidia, then they will have to employ the same technology they developed for the Ryzen processors (so I want AMD to ensure that the 600 series is a multi-die model with Infinity Fabric, XFR 2, and Precision Boost 2). We also know by now that a couple of their Ryzen guys are being sent to oversee the development of the new Radeons, so hopefully that will be the case.

28

With OpenCL C++/C 2.2 being mainlined into LLVM/Clang I fully expect more improvements to the OpenCL stack from the AMD devs who donated it to the community in the first place.

29

I think you misspelled “all”. Anyway you’re just agreeing with me. You don’t hold that much market share if your products are actually worse than the competition.

You’re looking at numbers while customers want to actually put those GPUs to work and it is impossible or extremely difficult with AMD.

“But AMD works with Cycles and Luxrender!”. Who the hell cares, really?

The point is that most GPGPU software is developed for CUDA or doesn’t work on AMD GPUs. AMD would need a decade of trouncing performance for their GPUs and actual support for GPGPU software to change it. Full stop.

30

Agreed with you Ace on all your points.

Now, I’m wondering how they plan to implement the infinity fabric as I expect that it would have far larger impact then in CPU’s. Either far wider then what the CPU has and definitely at lower latency.

I wonder if they will manage to make this multi-die fusing invisible to the host, essentially bypassing xFire. Or if it will still require work on the game developer side.

Either way, as apparently the Radeon Zen project kicked of in October of last year, with a due date of 1 year, that means we should be getting much more information within the upcoming months.

31

AMD has 7nm GPU’s being tested in their labs right now.

However, the leap from lab to consumer product can take a while (so I wouldn’t get hopes up for a Pascal/Turing killer this year). It’s still good to see that AMD is serious about improving the Radeon Line.

32

Currently rumor until proven otherwise but WCCFtech has posted a pretty sizable finding about Vega 20 and Polaris 10

33

If AMD pulls off a 70 percent performance boost compared to Vega 64, then there’s suddenly competition with Nvidia again (which hopefully means lower GPU prices for everyone).

The effort to ‘zen’ Radeon hasn’t been around very long, but if this is any indication, then they’re moving pretty fast.

34

Totally missed this article.

As for 70% improvement… I have my doubts. That level of re-engineering would take a long time. Just see how long the Ryzen was in development but it did provide a nice 40%+ boost over it’s predecessor.

But as mentioned, if indeed they did get that much performance boost, they need to release that to us ASAP. It would finally bring some refreshing competition to the market as much as Ryzen did in the CPU area.

35

I also missed this article. However, I had no problem with install AMD stuff besides the configuration standard for linux. Even get the AMD catalyst system up and running on Ubuntu, Chrome OS, Chromium and etc… Ubuntu Studio for Artist appears to take snap shots and also act like a mainframe with a pretend boss also.

36

Some news on the GTX 1060 TI refresh and the RX 580 refresh

37

The RX 580 is referring to something different to the topic of this thread.

in China, AMD is rebranding the RX570 as the RX580. Which is different then the rumoured Polaris 30 RX590 rumours

Worst case, AMD and Nvidia have been doign this too often, mostly in China (rebradnign lower speed cards as faster ones… why… no clue but it is stupid…

1 Like
38

So, 7nm VEGA this quater… granted not for the regular folks, but nice to see that 7nm is coming out very soon.

1 Like
39

alleged benchmakrs fo the Vega 20 (7nm). though this will not have a video out port, as it is targeted at compute market, it does show performance improvements and I wonder how it stands on power. (assuming this is Vega 20, and not Navi … or maybe it is?)…

Still I do wonder what it’s compute power ill be.

40

You could always buy a vega52 and do the well known power mod (registry edit that lets the card draw 200 watt more power).gamers nexus did this and got rtx 2070 performance out the card, but i wouldnt expect it to last for years when over speced like that, but for price point could be worth the risk. Dont blame me if you catch the thing on fire.