I’m building a new desktop system soon after Ivy Bridge is released and want it to optimized for decent performance with CS5.5/6 with the focus on PPro.
I’ve not used PPro before so not sure how well it scales with RAM or how best to allocate multiple drives to maximize performance. I’m using a Panasonic SDT750 shooting in 1080i and 1080P and 2D only.
So my questions are:
I’m hoping to get away with 8 or 16GB DDR3-1600 so if more is recommended it would be good to know how much of a gain you can achieve by going higher than that.
I have a Crucial C300 SSD 128GB, a WD 2TB Green SATA 5,400 rpm and potentially a Samsung PM830 SSD 256GB. I’m (naively) thinking of the following scenario:
A) Crucial C300 – O/S and applications: pagefile disabled!
B) Samsung PM830 – Working Projects and Media.
C) WD Green – Previews and exports.
Questions stemming from this allocation:
As you can see I’m new to this so would be grateful on help in allocating resources. I’m happy to make changes as long as they are within my thermal/noise budget and offer reasonable value for money; silent computing is a limiting factor for my design.
I’m not looking for advice on CPU or GPU
1. I had already looked at the PPBM5 site and noticed that 32GB was very common. But I try not to assume anything so still want to know how much difference 32GB would be over 16GB; has anyone made that comparison?
2. I understand the benefits of SSDs and see that many people use HDD RAID setups so I’m trying to understand which aspects of the process particularly benefit from good read or write access times or transfer rates. i.e. I can’t allocate resources efficiently unless I know where the potential bottlenecks are. So which are the critical areas for allocating fast storage to in order of importance and which are related to read and which write speed?
3. The reasons I wondered whether 5,400 rpm HDDs might not be a significant bottleneck for Exports/Previews are a) other areas of the work-flow are the bottleneck in these situations. Is this the case when working with AVCHD? b) The transfer rate is a factor of the spindle speed and aerial density and sometimes faster drives have lower densities which reduces their advantage.
4. AFAIK it’s okay to use SSDs for constant reads and it’s only writes that wear out the drives. That’s why I was going to allocate an SSD to hold project data that is read only. Are you suggesting otherwise?
That graph perplexes me as for H.264 there is little difference when going from 4 to 16GB and then a big jump from 16 to 32GB:
4GB (2133) 61
16GB (2133) 57
16GB (1333) 61
32GB (1333) 52
It just seems a very unusual way for performance to scale!
I can not explain that almost insignificant anomaly. The two plots were definitely gathered at different times and so might have other external factors coming into play. When I initially gathered the second table it was all done at one sitting. Later I got the 32 GB (4 x 8GB) set of memory and for 16 GB I used a 4 x 4GB different set of modules, possibly resulting in different timings, but I thought that it would be a better comparison that just running a test with only two modules. But all the numbers are really close so without many, many more repeated runs it would be difficult to say anything except with our PPBM5 H.264 timeline encoding is really memory insensitive. Repeatability in these tests is affected by the coarseness of the one second Windows clock so any answer is +/- 1 second and then since all tests involve writing to asynchronous disk drives you have another variable that affects each and every run.
Generally to establish a real baseline measure I have run 10 complete benchmarks and do a round-off average, and I have on occasion thrown out an an "unexplained" deviation.
But put it this way, more often than not I have seen better results of 32 GB of slow memory than 16 GB of faster memory. And I am overclocking the CPU extensively. This winter in my cold basement I have been consistently running at 5.0 GHZ with that 32 GB of DDR3-1333. Now if the price of RAM every bottoms out again and I could afford 4 each 8 GB fast RAM modules I of course would be very happy.
Bill, I’m not sure why you say it’s almost insignificant!
4 to 16GB = 61 to 57 seconds = 7% less time whilst quadrupling the memory
16 to 32GB = 61 to 52 seconds = 15% less time whilst doubling the memory.
I’m leaning towards 16GB of DDR3-2133 which is roughly half the price of 32GB DDR3-1600 and seems a good trade off; this is with LGA1155.
I’m not chasing every last drop of performance at any price.
I just need to understand now how best to utilise storage to narrow the system build down.
Thanks for the PPBM5 vs. RAM results table!
1. 16GB does sound like a good choice for your build; you may even want to get a slower speed and 1.5v RAM. I am assuming you are going for 1.6v on the DDR3-2133
2. For your "A" drive, great. For your "B" drive, a SSD is really not a good choice at all; suggest you sell it and fund a new 2TB or 3TB 7200 rpm drive. Now for your "C" drive, WD Greens (less than 7200 rpm) certainly are not a good choice for a new purchase for a PPro rig, but since you have a big one already, why not just try it and change it out later if it does not produce acceptable results.
3. See #2. Also, regarding your mention of heat issues (not sure why though for a desktop?), external eSATA solutions are excellent, and USB 3.0 would be OK.
4. No, and trust me I've tried!