New Build — i9-7900X or Threadripper?

[Trevor.Dennis]

In all the years of using custom computer builds, I have never agonised for so long, or changed my mind so often.  The reviews and bencharks are all over the place for a start.  Puget Systems and Toms Hardware are the most reliable, and with by far the majority of my workload being with the lightly threaded Photoshop, I'd decided on the 7900X's higher clock speeds and relatively low core count.  Keeping it's temperature under control seems to be an issue.  Runing it flat out stands a chance of it getting hot enough to throttle back even with a big 280mm AIO liquid cooler, but Photoshop is hardly going to stress it.

Then someone pointed me at this video

It does not show any Photoshop benchmarks, but my thinking had been to optimise for Photoshop, and put up with a slight compromise with apps like Premiere Pro. but this site has Threadripper with a significant advantage with the higfhly threaded apps — and it does not get nearly so hot.

I'd found a nice X299 mother board in the ASRock Fatal1ty Professional Gaming i9, which has a ton of SATA and USB3 ports, and a better than average VRM with 13 phases, but I liked the MSI X399 Pro Carbon even more.  There wouldn't be much in it cost wise as the X299 MB is a bit cheaper, but I am not going to worry about a few hundred dollars either way after all this hassle.

There must be people who have built both Threadripper and 7900X systems for Photoshop by now, so if you are one of them, and read this, I would dearly love some feedback.  Especially if you have the i9-7900X and can tell me if keeping it cool is as big a problem as they are saying, and if not, how are you managing it?

Thanks in advance for any help.  I am desperate to get tis nailed down and start the build.

[davescm]

I can't help you Trevor - I'm going to use you as the guinea pig before my next build . I'm interested in the answers though.

Dave

[Trevor.Dennis]

I'm feeling disinclined to abandon the findings of Puget Systems and Tom's Hardware for some bloke from Oz.

Puget Systems Skylake vs Threadripper with Photoshop conclusion

And the same thing but with Premiere Pro

These show 7900X with a small but still clear lead over either Threadripper with Photoshop, but unlike the Ozi guy, they show that 7900X is definitely not trounced by Threadripper with Premiere Pro.  In fact it beats it.  I'm going to put this down to a last minute wobble, and stay on track with i9X

7900X

ASRock X299 Professional Gaming i9

64Gb G-Skill Trident Z 3200

Cooler Master Silent Pro Gold 1000W  (or maybey the 1200W)

Corsair H115i (280mm liquid cooler)

Cooler Master Cosmos II (GIANT case)

GTX970 bumped from my current system — I'll maybe look at a GTX1080 next year if nothing comes of Volta

Samsung 960PRO 512Gb M.2 — two to start with.  When I see how they perform I'll decide how to use the third M.2 slot.

I'll bring an 840PRO 250Gb, 840EVO 1Tb, 4HDDs and 8 USB3 externals from this system across to the new build (which is why I wanted all the SATA and USB3 ports)

But I'd still like to hear if it is possible to stop the 7900X from melting.

[davescm]

I'm sure there is plenty room in one of those nearby vine fields for you to build the air-con plant

Joking aside, although most Photoshop use does not stress the CPU for long periods, a 3D render does. So I would want to be sure about the cooling. External liquid cooling?

Dave

[Trevor.Dennis]

I'm going with a 280mm AIO as an initial plan.  If that doesn't cope, then its either delidding the CPU  — a bit scary with a NZ$1500 component — and / or adding another radiator with a custom loop.  But as you say, for the most part it is not going to stressed into getting seriously hot.  If I did as much 3D as you Dave,

I would probably have gone with Threadripper, and probably the 1950X.  Puget Systems is our go to place for hardware used for digital media content creation, but they also have some interesting blog articles.  This entry from the 5th October, resonated with me.  He makes the point that for many years he looked for a component price/performance sweet spot on an individual component basis.  That's how I have also done it, and I bet most people do the same.  He then argues that it would be more valid to set price differences as a percentage of the overall build cost, and I think he is right.   So when you look at the US$180 differential between the 1920X and 1950X.   That's a bit over 20% more cost for x increase in performance, but it would be well under 5% when set against the total build cost.  

'A ha'porth of tar', and all that.