3DS Render Server Build AMD

Satoshi · October 25, 2019

We're looking to replace our PowerEdge FC430's with dual Intel Xeon E5-2680 v3 rendering nodes in our Max 3ds render farm (we render 3Ds max files that include vray, phoenix, Forest pack plugins). We also have a handful of AS-2022TG-HTRF with dual AMD Opteron Processor 6272 that were the predecessors to the PowerEdge. From our benchmarks most of our rendered job efficiency depends on number of physical cores so we're emphasizing this upcoming upgrade/rollout on Cost per Cores.

We've been discussing racking desktop CPU's (threadripper 2990wx, Ryzen 9 3900x) and also looking at Epyc dual CPU servers (7401, 7551) and would like to see if anyone has any input/experience or suggestions.

We're currently bench marking the Ryzen 9 3900x and the dual xeon machines in the office, it did slightly lower than an i9-7900x series on a v-ray benchmark and we'll be bench marking actual 3DS Max renders over the weekend to compare the machines.

We've just begun the process and have limited down time to invest into researching this project so we're posting this for any advice/opinions/assistance, which would be GREATLY appreciated and hopefully we can help others who may be in the same boat.

Firewrath9 · October 25, 2019

look at servethehome.com, they have great reviews. Currently the 7702p has better performance than dual 8280's at a fraction of the price.

https://www.servethehome.com/amd-epyc-7702p-review-redefining-possible-at-64c-per-socket/

(They don't have v-ray benchmarks but there is c-ray which is not really representative but is still a raytracing benchmark)

if you look at the price per core, the Epyc 7282 stands out, as well as the 7552 and 7452 for dual socket systems. I think that single-socket blade/1U servers would be the best performing. Not to mention that epyc stomps intel in efficency too, it only drew 270W peak

Quote

We tested these in a number of configurations. The lowest spec configuration we used is a Supermicro AS-1014S-WTRT. This had two 1.2TB Intel DC S3710 SSDs along with 8x 32GB DDR4-3200 RAM. One can get a bit lower in power consumption since this was using a Broadcom BCM57416 based onboard 10Gbase-T connection, but there were no add-in cards.

Even with that here are a few data points using the AMD EPYC 7702P in this configuration when we pushed the sliders all the way to performance mode and a 200W cTDP:

Idle Power (Performance Mode): 104W

STH 70% Load: 245W

STH 100% Load: 261W

Maximum Observed Power (Performance Mode): 273W

As a 1U server, this does not have the most efficient cooling, still, we are seeing absolutely great power figures here. The impact is simple. If one can consolidate smaller nodes onto an AMD EPYC 7702P system, there are power efficiency gains to be attained as well.

Next, let us look at our performance benchmarks before getting to market positioning and our final words.

However try to get quotes for both systems, and then compare performance as apparently vendors charge vastly different prices than what are listed.

Satoshi · October 28, 2019

Very helpful post/info. We've been discussing the depreciation value of the high end products we've previously purchased and how quickly something newer and cheaper ends up outperforming it in 1-3 years, but that's the nature of our business. Our company has made it a habit of going with the cheaper options regardless of performance so I don't know if we'll be able to justify a server/build where the cpu is $7500 alone. To us, it obviously makes sense as the last time the company did a server upgrade they spent roughly around $40,000 in mediocre equipment and with a dual 7742 node we would literally more than double our current processing/rendering power for half that cost (dual 7742) while drastically lowering our footprint. The only negatives are the obvious initial cost and depreciation.

We will be further looking into this but thank you for the information you provided as it has helped us consider the "high end" option.