News Franken-CrossFire: Radeon RX 5600 XT Joins Radeon RX 5700

[Admin]

CrossFire is gone, but we still have multi-GPU.

Franken-CrossFire: Radeon RX 5600 XT Joins Radeon RX 5700 : Read more

 

[techgeek]

Being that this is supported in DX12, one would assume this could be done on nVidia hardware. I wonder if they'll try to benchmark this on nVidia hardware. nVidia still technically "supports" SLI, but not many games support it, especially new titles. I just wonder how the scaling compares between SLI and mGPU?

As for the article, scaling isn't terrible. I like the uplift in the .1% in Rise of the Tomb Raider, because this is where gains are the most important.

I wonder if they were looking for micro-stuttering? As this seemed to plague multi-GPU systems, especially CrossFire, that's why we got FCAT from nVidia which exposed this.

 

[cusbrar2]

I wonder if they were looking for micro-stuttering? As this seemed to plague multi-GPU systems, especially CrossFire, that's why we got FCAT from nVidia which exposed this.

The stuttering exhibits on Navi hardware even with 1 GPU so I don't think that is part of it, once it is fixed for single GPUs it should also go away for mGPU setups... Note that SLI almost always fails to impove 1% lows or min frametimes, while a very good increase is seen here. You'll see games in SLI configurations with 40FPS 1% lows pushing 160FPS etc...

 

[dehdstar]

emGPU was such a let down. This is because software companies are too busy pointing the finger at the other as to how it's actually going to be implemented. The result is a stand off.

If you ask me Microsoft should have some sort of system built in so that any dedicated GPU can intuitively begin processing, so long as they are like API cards. It shouldn't be up to the devs, as this just results in some games having the feature and others not. Which means we have wasted silicon sitting in our systems most of the time and inconsistent performances that range from nearly twice the performance to single GPU performance. A clever API should even be able to take an old game and toss extra GPUs at it, as if it were seen as just one (bigger) GPU. It should just be an OS function, in other words. Microsoft can do it, I mean virtual OS's and machines already essentially do this. You can run a server loaded with massive Quadros or Teslas and tell the software how much GPU to dedicate to each workstation. That's got to be harder than an OS just auto-delegating the task to apps, as it would a CPU.

 

[dehdstar]

Being that this is supported in DX12, one would assume this could be done on nVidia hardware. I wonder if they'll try to benchmark this on nVidia hardware. nVidia still technically "supports" SLI, but not many games support it, especially new titles. I just wonder how the scaling compares between SLI and mGPU?

As for the article, scaling isn't terrible. I like the uplift in the .1% in Rise of the Tomb Raider, because this is where gains are the most important.

I wonder if they were looking for micro-stuttering? As this seemed to plague multi-GPU systems, especially CrossFire, that's why we got FCAT from nVidia which exposed this.

Depends on how eMGPU is leveraged. Since it's evidently up to the devs to implement, many of them are just going to gravitate to AFR methods, such as SLI/ Crossfire because it's easier. But it's the poorest use for two GPUs. All it's doing is making one GPU render frame 1, while the other renders frame 2. This means you're paying an entire GPU price for at best? 10-20% real world performance.

The end goal with eMGPU was a literal compute pool. Something like the old Voodoo SLI, or even SFR. In the old days? Especially going back to Voodoo SLI? The goal was to create performance that was nearly 2x that of a single GPU. One Voodoo GPU would render the top half and the other the bottom half and there was I think a VSYNC-like feature called Z-sync, or Z-buffer to sync the two GPUs. Around 2008(?) SFR was born, kudos to Nvidia. It attempted to bring back that concept, but was scrapped due to "syncing" issues, allegedly. Honestly? I think that GPU vendors just didn't like people buying multiple cards and not having to upgrading for another 5 years lol.