M380 M390 M395 M395X Thread

[gamervivek]

Here's the R9 M390 from the teardown photos on OWC's blog:


Compare to any photo of Pitcairn or its many rebadges. That is, beyond a shadow of a doubt, Pitcairn.

You can give similar hardware new device IDs as you please, but if you give dissimilar hardware the same device ID, things go pear shaped.

Neptune is not Pitcairn. I'm not going to repeat all of the stuff I've said already. That you still consider it to be a Pitcairn rebadge beyond a shadow of doubt by looking at the gpu die and comparing it to pitcairn rebadges(and how do you know they were rebadges? device ID? On no, it's handwaved away!) is enough for me to stop bothering.

As for Dave Buamann's GCN2 being completely different from Ryan Smith's GCN1.1, no they aren't so.

OpenCL 2.0 compliance occurs with Bonaire and later.

https://forum.beyond3d.com/threads/...ation-rumor-thread.55600/page-90#post-1851125

 

[repoman27]

Neptune is not Pitcairn. I'm not going to repeat all of the stuff I've said already. That you still consider it to be a Pitcairn rebadge beyond a shadow of doubt by looking at the gpu die and comparing it to pitcairn rebadges(and how do you know they were rebadges? device ID? On no, it's handwaved away!) is enough for me to stop bothering.

I was looking at the die and the package, but mostly at the layout of the passives on the interposer, those are the dead giveaway. Once again, I think we're having a communication issue because you have a much more specific definition of Pitcairn than I do. I'm just talking about the silicon itself, not which fuses have been blown or features have been enabled through firmware or drivers.

 

[roaltar]

Some excellent benchmarks from barefeats here:
First, 5k resolution test which shows that clearly, just as we thought, not even m395x is able to handle 5k. And the difference in performance between m395 and m395x is negligible.

http://barefeats.com/imac5k17.html


Secondly, test of all the new GPUs from m380 to m395x on 1440p resolution

http://barefeats.com/imac5k15.html

Pls note, that the m395x machine was equipped with double the ram (16 gigs) faster processor 4.0 and SSD faster storage. That can, theoretically, influence the gpu performance in terms of fps. Again, the difference between m395 and m395x is, imo not worth the 250$

Cheers.

 

[magbarn]

Some excellent benchmarks from barefeats here:
First, 5k resolution test which shows that clearly, just as we thought, not even m395x is able to handle 5k. And the difference in performance between m395 and m395x is negligible.

http://barefeats.com/imac5k17.html


Secondly, test of all the new GPUs from m380 to m395x on 1440p resolution

http://barefeats.com/imac5k15.html

Pls note, that the m395x machine was equipped with double the ram (16 gigs) faster processor 4.0 and SSD faster storage. That can, theoretically, influence the gpu performance in terms of fps. Again, the difference between m395 and m395x is, imo not worth the 250$

Cheers.

Thanks for the post. Ironically, based on gpu compute tests, the users who are going to use the iMac for work have more to gain from the m395x than gamers. The m395x upgrade won't make a game that's unplayable on the m395 playable at 5k or 2k resolutions.

PS. That iris pro is a cruel sick joke on iMac 4K users. Can't even push half the pixels of its 4K screen!

 

[TechZeke]

Some excellent benchmarks from barefeats here:
First, 5k resolution test which shows that clearly, just as we thought, not even m395x is able to handle 5k. And the difference in performance between m395 and m395x is negligible.

http://barefeats.com/imac5k17.html


Secondly, test of all the new GPUs from m380 to m395x on 1440p resolution

http://barefeats.com/imac5k15.html

Pls note, that the m395x machine was equipped with double the ram (16 gigs) faster processor 4.0 and SSD faster storage. That can, theoretically, influence the gpu performance in terms of fps. Again, the difference between m395 and m395x is, imo not worth the 250$

Cheers.

The 5K test is particularly interesting since it seems like extra VRAM isn't much help at all. I would have expected a much bigger percentage difference.

I'm not even sure why the M395X carries a $250 premium. It made sense on the M295X vs the M290X since there was a significant difference in architecture and performance between the two.

Past GPU upgrades were all $100-$150.

 

[magbarn]

The 5K test is particularly interesting since it seems like extra VRAM isn't much help at all. I would have expected a much bigger percentage difference.

I'm not even sure why the M395X carries a $250 premium. It made since on the M295X vs the M290X since there was a significant difference in architecture and performance between the two.

Past GPU upgrades were all $100-$150.

That's what I've been saying all along the m395x doesn't have the horsepower to take advantage of the extra 2gb of vram. Pc gaming laptop makers do it all the time for spec wars where they pair up a 980M with 8gb of vram.

 

[joema2]

...the difference between m395 and m395x is, imo not worth the 250$...

Excluding the one anomalous test where the 395X was slower, the average improvement over the M395 was 16%. The real question is how much your particular tasks are improved, and that will be different for each person's workflow.

I don't have a M395 to test but in some GPU intensive video editing tests my iMac with M395X is much faster than my 2013 model with nVidia GTX-780m. In the BruceX benchmark the M395X is 1.8 times faster.

 

[twilexia]

Excluding the one anomalous test where the 395X was slower, the average improvement over the M395 was 16%. The real question is how much your particular tasks are improved, and that will be different for each person's workflow.

I don't have a M395 to test but in some GPU intensive video editing tests my iMac with M395X is much faster than my 2013 model with nVidia GTX-780m. In the BruceX benchmark the M395X is 1.8 times faster.

What was your BruceX benchmark for m395x? on my i7 1TB SSD M395 16gb ram it was 29s

 

[joema2]

What was your BruceX benchmark for m395x? on my i7 1TB SSD M395 16gb ram it was 29s

Mine was 19 seconds, although subsequent runs were as fast as 18 sec. That is exporting to ProRes 422 at 5120x2700. The actual output file size was 70MB and Finder Info shows 5120x2200, duration 00:02. Configuration:

2015 iMac 27, 4Ghz i7-6700K, 32GB, 1TB SSD, M395X
OS X 10.11.1, FCP X 10.2.2

 

[twilexia]

Mine was 19 seconds, although subsequent runs were as fast as 18 sec. That is exporting to ProRes 422 at 5120x2700. The actual output file size was 70MB and Finder Info shows 5120x2200, duration 00:02. Configuration:

2015 iMac 27, 4Ghz i7-6700K, 32GB, 1TB SSD, M395X
OS X 10.11.1, FCP X 10.2.2

Fantastic... I think I had the same resolution (ProRes 422). So it's about a 33% decrease in time. Do you think it is primarily attributed to the 2GB extra VRAM?

 

[joema2]

Fantastic... I think I had the same resolution (ProRes 422). So it's about a 33% decrease in time. Do you think it is primarily attributed to the 2GB extra VRAM?

I really don't know if it is VRAM, or other internal GPU features. Has anyone with M395 run and posted results from the various GPUTest benchmarks? That might give a broader result. Attached is a file with my various benchmark results.

 

[twilexia]

I really don't know if it is VRAM, or other internal GPU features. Has anyone with M395 run and posted results from the various GPUTest benchmarks? That might give a broader result. Attached is a file with my various benchmark results.

I'll run these when I get home. For the GPU and FCPX tests, are these all individual tests? Is there a template for the FCPX tests so I can run them on the same parameters as you?

 

[magbarn]

Fantastic... I think I had the same resolution (ProRes 422). So it's about a 33% decrease in time. Do you think it is primarily attributed to the 2GB extra VRAM?

It's either the test or the extra vram. The m395 is a cut down version of the m395x and it's not a 33% cut down.

 

[koyoot]

3.7 TFLOPS vs 3 TFLOPs. M395X has 1/4th more compute power and double the VRAM.

 

[joema2]

I'll run these when I get home. For the GPU and FCPX tests, are these all individual tests? Is there a template for the FCPX tests so I can run them on the same parameters as you?

Unfortunately I was not planning on it being a formal test; I just ran the FCP X effects on a 4k clip from an AG-DVX200. I used all the defaults for each test but I can't remember the exact clip I used. We would both have to use the same clip for a comparison, and the same test procedure -- timeline rendering vs exporting.

If you want to pursue it, I can re-run all the tests on a common clip we decide on. It could be one we download from Vimeo or someplace; it doesn't have to be any particular one. However I'd recommend it be one with an approx rendering time of in the 1-2 min range to make testing easier. If you have any suggestions or preferences for a test clip, let me know.

The Beauty Box and Flicker Free tests are plugins from Digital Anarchy; the others are built-in FCP X effects.

I just tried the FCP X plugin Photon Pro plugin Photon Pro video noise reduction. It is very GPU-intensive. It is $29 on the App Store.

 

[twilexia]

Unfortunately I was not planning on it being a formal test; I just ran the FCP X effects on a 4k clip from an AG-DVX200. I used all the defaults for each test but I can't remember the exact clip I used. We would both have to use the same clip for a comparison, and the same test procedure -- timeline rendering vs exporting.

If you want to pursue it, I can re-run all the tests on a common clip we decide on. It could be one we download from Vimeo or someplace; it doesn't have to be any particular one. However I'd recommend it be one with an approx rendering time of in the 1-2 min range to make testing easier. If you have any suggestions or preferences for a test clip, let me know.

The Beauty Box and Flicker Free tests are plugins from Digital Anarchy; the others are built-in FCP X effects.

I just tried the FCP X plugin Photon Pro plugin Photon Pro video noise reduction. It is very GPU-intensive. It is $29 on the App Store.

I don't have any preference for the test clip. Happy to use whichever one you choose. For the tests I prefer to use free plugins and built in effects . Not sure how GPU would affect noise reduction?

 

[magbarn]

3.7 TFLOPS vs 3 TFLOPs. M395X has 1/4th more compute power and double the VRAM.

If that's the case then why in Barefeats' benchmarking the difference only 10-18% difference between the two in their FCPX/openCL testing?

Unfortunately I was not planning on it being a formal test; I just ran the FCP X effects on a 4k clip from an AG-DVX200. I used all the defaults for each test but I can't remember the exact clip I used. We would both have to use the same clip for a comparison, and the same test procedure -- timeline rendering vs exporting.

If you want to pursue it, I can re-run all the tests on a common clip we decide on. It could be one we download from Vimeo or someplace; it doesn't have to be any particular one. However I'd recommend it be one with an approx rendering time of in the 1-2 min range to make testing easier. If you have any suggestions or preferences for a test clip, let me know.

The Beauty Box and Flicker Free tests are plugins from Digital Anarchy; the others are built-in FCP X effects.

I just tried the FCP X plugin Photon Pro plugin Photon Pro video noise reduction. It is very GPU-intensive. It is $29 on the App Store.

I didn't realize that test required you to put in your own video, in that case you can't compare 19 to 29 seconds. Agree that the same clip should be run.

 

[twilexia]

If that's the case then why in Barefeats' benchmarking the difference only 10-18% difference between the two in their FCPX/openCL testing?


I didn't realize that test required you to put in your own video, in that case you can't compare 19 to 29 seconds. Agree that the same clip should be run.

For the BruceX benchmark it's the same video.

 

[jerwin]

If that's the case then why in Barefeats' benchmarking the difference only 10-18% difference between the two in their FCPX/openCL testing?

(The tests are less than perfectly parallel. Conceivably, though, someone could come up with a scenario where having less than 3-4 GB promoted thrashing.

There are more detailed models in Luxmark than ball. My own video card cannot render the Hotel Lobby scene, though this may simply be a driver bug that affects all macs, or all amd macs rather than a demand for more vram

 

[joema2]

I don't have any preference for the test clip. Happy to use whichever one you choose. For the tests I prefer to use free plugins and built in effects . Not sure how GPU would affect noise reduction?

OK let's use this one at 1080p. It is 53 sec, about 30MB. Main page:

Direct download link: HD 1080p

 

[twilexia]

OK let's use this one at 1080p. It is 53 sec, about 30MB. Main page:

Direct download link: HD 1080p

Fantastic. I'll run this when I get home.

I don't know FCPX settings all too well but the for BruceX benchmark I recalled that I turned off background Render. If there's other settings you have in mind let me know. For the FCPX effects, would the testing process go something like right click video clip - apply effect - start stopwatch and stop once the render is finished?

 

[joema2]

Fantastic. I'll run this when I get home.

I don't know FCPX settings all too well but the for BruceX benchmark I recalled that I turned off background Render. If there's other settings you have in mind let me know. For the FCPX effects, would the testing process go something like right click video clip - apply effect - start stopwatch and stop once the render is finished?

Yes that's correct. Just turn off FCP X background rendering for BruceX. Ideally already have Quicktime launched so it doesn't wait for that, but I don't think it makes much difference.

For the other tests, apply the effect then render with CTRL-R, start the timer, wait for the background task display below the viewer to reach 100%, then stop the timer.

The only exception is stabilization, which has both an analysis pass and a rendering pass. For simplicity just wait until analysis is done then do CTRL-R and time the render pass.

To re-do most of these you have to delete the render files by clicking on the library then doing Files>Delete Generated Library Files, and select Delete Render Files.

If the execution time is too short we may have to use the 4k file but let's try 1080p for now.

 

[jerwin]

The configuration Guide for DaVinci Resolve recommends a D700 over a D500 specifically because of VRAM-- when working with UHD video.

http://documents.blackmagicdesign.c...e9/DaVinci_Resolve_12_Configuration_Guide.pdf

 

[huffy15]

The only thing I care about the 395x is the 4gb of ram. To 5k retina use almost 1gb itself so another 3gb won't be too much.

 

[jerwin]

To 5k retina use almost 1gb itself

quite the assumption! Pray, tell us how this is calculated!