MP 7,1 What is faster for FCPX. Base 7,1, or Fully Loaded 6,1?

[Korican100]

I have a 6,1 right now. I swapped the base CPU for a 3.3 GHz 8 core Xeon E5-2667 v2, and it has 64gb RAM, with Dual D700 AMD's.

The 7,1 as we know comes with 8 code 3.5ghz, with 32gb ram, and the AMD Radeon Pro 580X as base.

If I compared these two rigs, what would be faster in terms of video editing?

Processing effects, exporting/rendering, stabilization etc

I ask because I want to know how much I will need to upgrade the config from base once released. Or can I keep it simple for now, and then build as necessary going forward.

 

[h9826790]

Should be base 7,1.

1) CPU is much newer, even same clock speed, will be significantly faster.

2) A single RX580 is about 2x 7970. However, single more powerful GPU always better than 2x weaker GPU. Also, single 8GB VRAM is better than 6+6 mirroring.

3) much newer architecture with better I/O

4) Faster storage

5) Faster RAM. However, this depends on your workflow. If you often need more than 32GB RAM, then 64GB 6,1 should have advantage.

6) Have full H264 / HEVC hardware acceleration. This will kill the 6,1 on H264 / HEVC editing and exporting (there is no need to transcode to ProRes before editing. And the HEVC exporting performance can easily 10x better than the 6,1).

 

[casperes1996]

2) A single RX580 is about 2x 7970. However, single more powerful GPU always better than 2x weaker GPU. Also, single 8GB VRAM is better than 6+6 mirroring.

I agree with pretty much everything you wrote, but I'll add that when working with lower level APIs like Metal, DX12, Vulkan, and even with a lot of work APIs like OpenGL, you don't need to mirror the VRAM on the GPUs. You can set them to do completely independant tasks on their own VRAM, instead of the traditional Cross-Fire style split render setup

 

[h9826790]

My answer only base on OP’s question, target FCPX, in my own tests (I ran dual 7950s few years back), FCPX always mirroring the VRAM.

May be it’s better now, especially the latest FCPX has a new Metal engine. But I already sold my 7950s, can’t test it anymore.

 

[Korican100]

2) A single RX580 is about 2x 7970. However, single more powerful GPU always better than 2x weaker GPU. Also, single 8GB VRAM is better than 6+6 mirroring.

My answer only base on OP’s question, target FCPX, in my own tests (I ran dual 7950s few years back), FCPX always mirroring the VRAM.

Forgive me for my ignorance guys. Is the 7970 the equivalent of D700 Firepro?
Also, could you expand on mirroring the VRAM, and how 6+6 (12gb) is worse than 8gb?

 

[h9826790]

Forgive me for my ignorance guys. Is the 7970 the equivalent of D700 Firepro?
Also, could you expand on mirroring the VRAM, and how 6+6 (12gb) is worse than 8gb?

D700 is the same chip as 7970, however, due to the thermal / power restrain, the D700 has lower clock speed.

6+6 mirroring means both 6GB VRAM contain the same data. So, no difference then single 6GB. Since the 2nd GPU cannot use the VRAM from the 1st GPU, so it must keep its own copy of data on its own VRAM.

Therefore, despite the nMP has 12GB VRAM in total, but it will run out of VRAM before the RX580 hit its 8GB limit (when the apps can only use the VRAM in mirroring).

 

[casperes1996]

My answer only base on OP’s question, target FCPX, in my own tests (I ran dual 7950s few years back), FCPX always mirroring the VRAM.

May be it’s better now, especially the latest FCPX has a new Metal engine. But I already sold my 7950s, can’t test it anymore.

Fair enough. I've never actually had the pleasure of running Final Cut on a dual GPU machine. My best Final Cut machine to this day was my old iMac with the R9 M295X. Waiting for the next iMac refresh to get a new one. But I have done some GPU programming (very little), and thought I'd chip in that it is indeed possible to address the GPUs individually.

Also, could you expand on mirroring the VRAM, and how 6+6 (12gb) is worse than 8gb?

Well when you mirror it you don't get 12GB of usable space. You get 6. Both cards will keep the exact same stuff in their respective 6GB. Hence they mirror each other

 

[Selsk]

^Really? The 7970 is only 3gb, the D700s are 6gb each.

"D700
So here’s the easiest part. The D700 has exact spec matches to the FirePro W9000, AMD’s highest performing workstation class GPU. Notable specs for both the D700 & W9000 include:

• 2048 Stream Processors (texture mapping units/unified shaders)
• 384-bit memory bus width
• 264 GB/s memory bandwidth
• 6 GB Vram GDDR5

The W9000 specs at 4 teraflops single precision while Apple has the D700 at 3.5 teraflops. This could be because of a frequency delta for the core clock. On the W9000 it specs at 975 MHz core clock speed. Perhaps on the D700 it runs at 900? D700 is a Tahiti-based GPU."

 

[CC88]

I think the base 7.1 will be faster both in single and multi core processing and gpu. Also it will be upgradable in a simple way.

 

[casperes1996]

^Really? The 7970 is only 3gb, the D700s are 6gb each.

"D700
So here’s the easiest part. The D700 has exact spec matches to the FirePro W9000, AMD’s highest performing workstation class GPU. Notable specs for both the D700 & W9000 include:

• 2048 Stream Processors (texture mapping units/unified shaders)
• 384-bit memory bus width
• 264 GB/s memory bandwidth
• 6 GB Vram GDDR5

The W9000 specs at 4 teraflops single precision while Apple has the D700 at 3.5 teraflops. This could be because of a frequency delta for the core clock. On the W9000 it specs at 975 MHz core clock speed. Perhaps on the D700 it runs at 900? D700 is a Tahiti-based GPU."

That's all the same for the 7970 aside from the amount of memory. The Radeon 7970, is the Radeon variant of the FirePro W9000/D700

 

[h9826790]

^Really? The 7970 is only 3gb, the D700s are 6gb each.

"D700
So here’s the easiest part. The D700 has exact spec matches to the FirePro W9000, AMD’s highest performing workstation class GPU. Notable specs for both the D700 & W9000 include:

• 2048 Stream Processors (texture mapping units/unified shaders)
• 384-bit memory bus width
• 264 GB/s memory bandwidth
• 6 GB Vram GDDR5

The W9000 specs at 4 teraflops single precision while Apple has the D700 at 3.5 teraflops. This could be because of a frequency delta for the core clock. On the W9000 it specs at 975 MHz core clock speed. Perhaps on the D700 it runs at 900? D700 is a Tahiti-based GPU."

Same “chip”, device id 6798 for all 7970, D700, W9000.

RX580 4GB is RX580

RX580 8GB is also RX580

Same chip, same device id. VRAM size is not related.

Anyway, 7970 also has 6GB version.