Radeon VII Gets Pro Driver Support

AlphaAtlas

[H]ard|Gawd
Staff member
Joined
Mar 3, 2018
Messages
1,713
AMD released a new version of their Radeon Pro Software last week. Among several new features and fixes aimed at workstation users, AMD mentioned that "Radeon Pro Software for Enterprise 19.Q1 now provides support for AMD Radeon Products*," and today, HardwareLuxx says that means the new Radeon Pro Driver supports Radeon VII. The publication notes that the Radeon VII's FP64 performance is only half that of the Professional MI60 variant, while RX Vega has a fraction of the FP64 power, making this card a particularly interesting option for high-precision compute workloads. This is apparently part of AMD's "One Driver" program, which will allow professional apps to run on Instinct, FirePro, or "select AMD Radeon products." Thanks to cageymaru for the tip.

In addition, AMD announced that the Radeon Pro Software Enterprise Edition 19.Q1 or later release will also support the Radeon VII . In the first generation of Vega, only the Frontier editions were supported by the Radeon Pro software, while the Radeon RX Vega models had to do without the Enterprise driver. However, it is unknown what exactly the Radeon Pro Software Enterprise Edition will be released for the Radeon VII.
 
This is more in line with what I had hoped for Radeon VII--a workhorse and beast that would do compute, AR and graphics design for way less than a professional graphics card--yes, the poor man's Titan if you will--with most of the guts of the card dedicated to the processing and not to RTX, DLSS and things that are really slow to take off and mostly directed at gaming solutions. Likely not the best solution for gamers but a boon to many other users. Just curious what the M150/160 buyers must be thinking after putting down the big bucks for their cards and now getting a lot of the same results with the less expensive little brother.

Curious to see not-gaming benchmarks, firm pricing and availability.

Now reading that the Radeon VII has no UEFI support! A BIOS update should/might fix it but that is a big job for lots of folks--guess no one plugged in the VII and checked if everything worked OK before shipping out. Sure to be RMA on these and hopefully the BIOS will be updated before any more ship out. Bit of a rush job on these cards it seems.
 
Last edited:
I had a feeling they may go this route, since gaming performance lags but it has compelling stats for some non-gaming applications.
 
So this doesn't have anything to do with it's gaming performance, this just means that Enterprise drivers for the high end workstation cards (not game optimized drivers by any stretch) now also support regular Radeon products? That's how I understand it.
 
Can i jsut say its horseshit that everything is fucking whitelisted now. We have entered an era where what the device is capable of doesnt matter, jsut what others decide you can use it for...
 
Can i jsut say its horseshit that everything is fucking whitelisted now. We have entered an era where what the device is capable of doesnt matter, jsut what others decide you can use it for...

This has been the case for a long time, nothing new. You get what you pay for-- just like that Uber mode key you can buy from Bugatti for your Veyron.
 
That 5% performance doesn't seem as severe now, knowing you will most likely be able to sell this card at a FAR higher price than the RTX 2080.
 
That 5% performance doesn't seem as severe now, knowing you will most likely be able to sell this card at a FAR higher price than the RTX 2080.

Just pulled the trigger for this reason. If I don't like it I probably still get a +ROI anyway.
 
I am waiting to hear back from my Citrix rep to find out if this will work for the Virtual Apps platform, if so it may solve a few of my problems, the first being my Quadro's still aren't here and its been a month with no word on when they will actually be available. It's the memory I require most not necessairly the performance. If I can slap 3 of these in I come about even where those two RTX 6000's would have gotten me at a much lower price point. The added juice from powering and cooling these won't even touch that cost margin, so fingers crossed it is even possible given the new stuff :D because if it is I will cancel that order in a heartbeat.


Update to this, the Citrix people don't know, their engineers who were on the call have had mixed results on the AMD side and as a whole can't recommend them. They are recommending I go with the Tesla P100 for the work load I am looking at and use some of the configurations to implement memory scaling so it is shared better. They say it costs more but is rock solid and just works for months on end with out having to be looked at where they found the AMD stuff needs babysitting.
 
Last edited:
So I haven't looked too far into just how alike the MI60 and the Radeon VII are, but I wonder how likely or possible it might be to enable full FP64 performance with a BIOS mod? I think this same question has been asked around the web. Getting full FP64 support won't help games much, but for other applications, it might be a real selling point.
 
I am waiting to hear back from my Citrix rep to find out if this will work for the Virtual Apps platform, if so it may solve a few of my problems, the first being my Quadro's still aren't here and its been a month with no word on when they will actually be available. It's the memory I require most not necessairly the performance. If I can slap 3 of these in I come about even where those two RTX 6000's would have gotten me at a much lower price point. The added juice from powering and cooling these won't even touch that cost margin, so fingers crossed it is even possible given the new stuff :D because if it is I will cancel that order in a heartbeat.

Sounds like it will have enterprise virtualization / sr-iov... Curious to hear what your rep says.
 
the only thing this card has going for it is 4k editing.
Total failure for its price.
 
This is more in line with what I had hoped for Radeon VII--a workhorse and beast that would do compute, AR and graphics design for way less than a professional graphics card--yes, the poor man's Titan if you will--with most of the guts of the card dedicated to the processing and not to RTX, DLSS and things that are really slow to take off and mostly directed at gaming solutions. Likely not the best solution for gamers but a boon to many other users. Just curious what the M150/160 buyers must be thinking after putting down the big bucks for their cards and now getting a lot of the same results with the less expensive little brother.

Curious to see not-gaming benchmarks, firm pricing and availability.

Now reading that the Radeon VII has no UEFI support! A BIOS update should/might fix it but that is a big job for lots of folks--guess no one plugged in the VII and checked if everything worked OK before shipping out. Sure to be RMA on these and hopefully the BIOS will be updated before any more ship out. Bit of a rush job on these cards it seems.

FE has no non uefi support either.

Pro drivers suck.
 
This has been the case for a long time, nothing new. You get what you pay for-- just like that Uber mode key you can buy from Bugatti for your Veyron.
The practice is growing, along with my ire about it. Did you think this would help? Artificial segmentation is bullshit.
 
The practice is growing, along with my ire about it. Did you think this would help? Artificial segmentation is bullshit.


How so? If a manufacturer can make several products out of one, that's better for everyone starting at the top all the way down to the end-user. Volume production saves money.

If you want a card with certified drivers needed for fast high precision work, that generally also comes with top-notch support, then you will pay for it.
If you want a card for primarily playing games on, that includes a basic driver with basic support, then you pay for that.

Warranty and support isn't cheap, and that has to be built into the cost of the part.

We do this in my industry(industrial EOT cranes). We sell a high-end crane for $xx,xxx. Does some nice things, has a nice feature group, etc... We also sell that same exact crane for up to $x,xxx,xxx. Has the same controls package on it, but all the other features are turned on. You can also buy a license for the cheaper crane to turn all that extra fancy stuff on down the road, but you will pay for it.
Most technicians can work on fancy crane 1, but very very few have the skills, training, and know-how to work on the second crane.
It's cheaper(for us and the customers) and faster to build one set of controls to use across multiple product lines instead of building 20 different panels for 20 different configurations.
So there is a cost savings to this method, but your costs for training, engineering, support, etc... has to go somewhere. That somewhere is the top.
I don't know about you, but if I'm buying a video card for gaming, I'm sure as hell not interested in absorbing some of the costs related to the entire product lifecycle for the workstation products.
 
IF the hardware is present, you should be able to utilize it the way you see fit. IM not talking support or anything else like that, merely physical functionality. The driver should not be the differentiator.

I am quite familiar with this sort of practice, going back to mainframes where you literally remove a part or bridge a contact to increase function.

Pretty much every modern medical scanner, the Dr. is charged for each scan. Matterport will sell you a camera, but you have to license the software. There is no way to use it without their service, DJI too. Their handheld gimbals require an account.

There are deep unintended consequences for championing this sort of thing.
 
Last edited:
IF the hardware is present, you should be able to utilize it the way you see fit. IM not talking support or anything else like that, merely physical functionality.


Say you play an MMO like WoW. A new expansion comes out. You have all the game files for the new expansion because that's the way it is. Because Blizzard includes the files for the expansion in the game client, you now have the content(hardware) in your computer. So therefore you should have access to it, simply because it's there?

We should simply flip the switch for a customer that didn't spend the million bucks simply because it's there?

I don't see the logic.
 
See this is the problem, you dont understand the fundamental difference between hardware and software. In the wow example you are paying monthly to access the SERVER. The rest of your argument is so poorly constructed i wont even argue it.


Edit: changed my mind, i will rebut. There has never been an expac made solely of already used assets.
 
See this is the problem, you dont understand the fundamental difference between hardware and software. In the wow example you are paying monthly to access the SERVER. The rest of your argument is so poorly constructed i wont even argue it.


Edit: changed my mind, i will rebut. There has never been an expac made solely of already used assets.


In my example using cranes- The hardware is there. So is the software(must enter license key in the PLC logic to enable). You don't get to use that portion software to run all the extra automation hardware unless you pay more for a license.
That is a perfect comparison- The basic video card shares hardware with the premium product, that is differentiated by software. That software happens to come with other goodies like extra support. You need the software to run the hardware.
 
Just shows you how many artificial limitations are placed upon consumer hardware. However, this is a generous move by AMD.
 
Why even bother? With DXVK, Steam Play and native gaming releases now, there's far fewer games I _can't_ play on Linux (without Windows) than _can_.

I'm running Ubuntu as my primary OS with a windows VM for gaming / webex. We will see how this does. Super excited about sr-iov possibilities.
 
Sounds like it will have enterprise virtualization / sr-iov... Curious to hear what your rep says.
Just got off the phone with my rep and an engineering team, they can't currently recommend the AMD cards for a multi user heavy workload environment, they say that those systems require far too much babysitting and tinkering and they are saying I ditch the Quadro option and go for the Tesla P100's, it costs more but I can just run them for months on end with out even looking at them. But supposedly because of some of the features on the Tesla cards it will do a better job of memory scaling so I should see less usage per user on the memory front as it has some reporting capabilities back to the Xen App server so it can do some back end micro managing.
 
Why even bother? With DXVK, Steam Play and native gaming releases now, there's far fewer games I _can't_ play on Linux (without Windows) than _can_.

Right now GPU Passthrough is a bit easier (or at least it feels that way). We're getting damn close, though!
 
Go use Lutris already you silly goose! Seriously, you're actually making it harder than it really is.

Right now GPU Passthrough is a bit easier (or at least it feels that way). We're getting damn close, though!
 
Back
Top