AMD unveils new Radeon Pro Card

[GOTSpectrum]

Quote

AMD unveils Radeon Pro workstation graphics cards for visualization market

 

Quote

Pandher claimed that the new cards can beat the performance of the rival Nvidia Quadro P5000 and P4000 graphics cards in media and entertainment applications.

So AMD has released a new Pro GPU that is supposed to out perform quadros in a number of areas, this will be interesting to see if it is true as with the lack of CUDA support they have to have limited read in the field right? 

 

Quote

He said the cards can be used in creating high-end animations or virtual reality applications, as well as traditional computer-aided design (CAD) applications. The cards start at $1,000, and preorders will begin on August 13 on Newegg.com.

Quote

AMD said that visualization demand is rising in markets such as media, manufacturing, and architecture. The WX 8200 cards have 56 compute units that can perform at up to 11 teraflops

So basically it is a VEGA 56 with a 1000USD price tag... Lets hope they get the certification to back up the price

 

Quote

A number of software partners — from Autodesk to Adobe — have certified that their software works with the new cards.

This isnt exactly showing of just how many have certified the product but it looks promising at least. 

 

 

 

Over all I think RTG are pushing in the right direction, Vega is known to be a low margin product, if they can push them margins up I'm all for it as they need the RnD funding like mad. The only problem is will anyone actually use them, we can only wait and see. Id really like to see Linus get his hands on one and see exactly what it can do. But we all know that will be unlikely. I really want to see AMD think about building their own severs, EPYC and radeon in one box and sell them for pretty cheap. It would make for an interesting take on the market for sure. Im thinking like how the Gaming bundles were done. 

 

 

https://venturebeat.com/2018/08/12/amd-unveils-radeon-pro-workstation-graphics-cards/

 

 

[Guest]

I think this could be a pretty good card, in the benchmarks of the vega 64 and such, it was already seen that these cards perform good at things like cad and rendering.

And you don't necessarily need CUDA cores if the app supports OpenCL too, an most do.

[GOTSpectrum]

1 minute ago, timl132 said:

I think this could be a pretty good card, in the benchmarks of the vega 64 and such, it was already seen that these cards perform good at things like cad and rendering.

And you don't necessarily need CUDA cores if the app supports OpenCL too, an most do.

Tbh I love my Vega 64, It chews through a lot of things, But in apps that are CUDA and openCL accelerated, The CUDA is usually way faster, even my 980ti can out perform the vega64 in a few tasts... 

[Guest]

5 minutes ago, Ben Quigley said:

Tbh I love my Vega 64, It chews through a lot of things, But in apps that are CUDA and openCL accelerated, The CUDA is usually way faster, even my 980ti can out perform the vega64 in a few tasts... 

That is true, probably because it's optimized for nVidia hardware. I thought Linus did a video on the Vega cards, and that it was better in some cases, but worse in other cases.

I think if apps start also optimizing for AMD, it will improve performance a lot.

[GOTSpectrum]

1 minute ago, timl132 said:

That is true, probably because it's optimized for nVidia hardware. I thought Linus did a video on the Vega cards, and that it was better in some cases, but worse in other cases.

I think if apps start also optimizing for AMD, it will improve performance a lot.

The thing is for a very long time, since terrascale or so RTG has gotten its performance by brute forcing its way, Throwing TFLOPS around like they mean nothing, where as Nvidia like to throw money at gameworks, where things like hairworks just kill AMD perf. If they could use them TFLOPS equally wed live in a very interesting world. 

[Bcat00]

It be fun to see how to goes up against the quado card but i'm willing to bet it be a half-half win with some tests being better for AMD and the other half going to Nvidia

[GOTSpectrum]

1 minute ago, Bcat00 said:

It be fun to see how to goes up against the quado card but i'm willing to bet it be a half-half win with some tests being better for AMD and the other half going to Nvidia

I think you are right here,, But with this costing half as much as a P5000, you can buy twice as many for the same cost... So for smaller businesses, it may make sense just to stick two of these in a machine vs one P5000. That is, if they are in fact similar in performance. 

[leadeater]

4 minutes ago, Ben Quigley said:

I think you are right here,, But with this costing half as much as a P5000, you can buy twice as many for the same cost... So for smaller businesses, it may make sense just to stick two of these in a machine vs one P5000. That is, if they are in fact similar in performance. 

Or even be able to get a proper workstation card with certified drivers at all.

[AlTech]

Oooh. Noice

[GOTSpectrum]

1 minute ago, leadeater said:

Or even be able to get a proper workstation card with certified drivers at all.

I didnt think about that, I dont know much about certified gear, But I would think NV would have cheaper offerings too right??

[GOTSpectrum]

1 minute ago, AluminiumTech said:

Oooh. Noice

Agreed, 1k for a double digit TFLOP card with certified drivers is Nice indeed. If you go a few years back most wouldn't think AMD would be competing like this. 

[leadeater]

9 minutes ago, Ben Quigley said:

I didnt think about that, I dont know much about certified gear, But I would think NV would have cheaper offerings too right??

Low end Quadros are terrible, most people just buy GTX cards of the same price and get typically double to triple the performance with the odd case coming about about the same. Only a select few things must have a workstation card with certified drivers to work at all i.e. SolidWorks Realview or Shaded /w Edges.

 

Quote

Only the Quadro cards have hardware-accelerated, antialiased points & lines for example. This is why the GeForce cards don't do too bad in just shaded mode. But turn on the edges and the GeForce cards are not accelerating those lines like a Quadro will. This kills performance on typical GTX cards.

https://www.pugetsystems.com/labs/articles/SOLIDWORKS-2018-GPU-Comparison-What-Is-the-Meaning-of-This-1112/

[GOTSpectrum]

1 minute ago, leadeater said:

Low end Quadros are terrible, most people just buy GTX cards of the same price and get typically double to triple the performance with the odd case coming about about the same. Only a select few things must have a workstation card with certified drivers to work at all i.e. SolidWorks Shaded /w Edges.

 

https://www.pugetsystems.com/labs/articles/SOLIDWORKS-2018-GPU-Comparison-What-Is-the-Meaning-of-This-1112/

Interesting, So do you think this could actually be a hit at the given price point? 

[leadeater]

Just now, Ben Quigley said:

Interesting, So do you think this could actually be a hit at the given price point? 

Yep, as long as people know the product exist which is actually a surprisingly big problem.

[GOTSpectrum]

Just now, leadeater said:

Yep, as long as people know the product exist which is actually a surprisingly big problem.

Maybe sending out a bunch of the cards for free will help, like they did with consumer hardware to game devs. 

[schwellmo92]

The HBM2 is Hynix’s 2.4Gbps product, I would be interested in seeing how far someone can push it.

[Trixanity]

3 minutes ago, schwellmo92 said:

The HBM2 is Hynix’s 2.4Gbps product, I would be interested in seeing how far someone can push it.

It's clocked at 2 Gbps out of the box though so I think it'll be a lottery. Still higher than existing Vega cards though, so we'll see it perform better in some scenarios solely because of increased memory bandwidth. Will definitely be interesting in the benchmarks to see it outperform Vega 64/WX9100 and by how much.

[Dabombinable]

But in which specific circumstances has the GPU being able to hit 11Gflops?

[GOTSpectrum]

6 minutes ago, Dabombinable said:

But in which specific circumstances has the GPU being able to hit 11Gflops?

Correction - TeraFLOPS 

 

I believe FLOPS are a relatively standard across devices in terms of its definition... @leadeater Is this correct or am I misunderstanding? 

 

 

Unless you are asking in regards to precision... 

[GOTSpectrum]

8 minutes ago, Dabombinable said:

But in which specific circumstances has the GPU being able to hit 11Gflops?

If you are talking about precision then this will help. 

 

https://www.amd.com/en/products/professional-graphics/radeon-pro-wx-8200

 

 

[schwellmo92]

25 minutes ago, Trixanity said:

It's clocked at 2 Gbps out of the box though so I think it'll be a lottery. Still higher than existing Vega cards though, so we'll see it perform better in some scenarios solely because of increased memory bandwidth. Will definitely be interesting in the benchmarks to see it outperform Vega 64/WX9100 and by how much.

The memory will at least hit 2.4Gbps with overclocking as that’s what it’s rated for, how far past that is lottery.

[The Benjamins]

49 minutes ago, leadeater said:

Low end Quadros are terrible, most people just buy GTX cards of the same price and get typically double to triple the performance with the odd case coming about about the same. Only a select few things must have a workstation card with certified drivers to work at all i.e. SolidWorks Realview or Shaded /w Edges.

 

https://www.pugetsystems.com/labs/articles/SOLIDWORKS-2018-GPU-Comparison-What-Is-the-Meaning-of-This-1112/

Their are third party patches to make solidworks realview work on gaming cards. And realview doesn't even help much with modeling/rendering anyways.

[leadeater]

9 minutes ago, Ben Quigley said:

I believe FLOPS are a relatively standard across devices in terms of its definition... @leadeater Is this correct or am I misunderstanding? 

Standard but a bit meaningless, they are just hardware specs of best case with all GPU resources fully utilized running an optimal workloads. They are also just calculations rather than actual measurements.

 

Depending on what you are actually running a 11 TFLOPs GPU could be executing something at only 100 GFLOPs, that's an actual real situation.

[leadeater]

Just now, The Benjamins said:

Their are third party patches to make solidworks realview work on gaming cards. And realview doesn't even help much with modeling/rendering anyways.

The link to the patch is given at the top of the article . They don't recommend it though.

[The Benjamins]

Just now, leadeater said:

The link to the patch is given at the top of the article . They don't recommend it though.

Oh cool, I used it it the past, but I don't see a need for realview.