Precision 7530 & Precision 7730 owner's thread

So now I have three different 16 GB 2666 MHz DDR4 ECC SODIMMs in mind:

Kingston KSM26SED8/16ME

Samsung M474A2K43BB1-CTD

Micron MTA18ASF2G72HZ-2G6

Speak of the devil; I was typing up my post just as yours came in. If you want 3.2 GHz RAM, it will likely be cheaper to simply buy it aftermarket (G.Skill, Corsair Vengeance, etc.) and use it in the Precision, as @anuraj1 has demonstrated earlier in this thread.

I'm looking for the fastest ECC memory I can find, in the hopes that I can overclock it (and tighten timings) so that I can eke out a little more performance. If not, I still get decent performance over 2.4 GHz ECC RAM. Ideally I would like the Samsung modules I've mentioned (because they use the fabled B-die), but I cannot find them for sale anywhere.

 

Isn't ECC RAM generally about 5% slower than regular RAM (at equal frequency) ?

 

Running multiple VMs on my 7730 and before m6800 with none ECC RAM and never had any reliability issues.

 

According to Techspot, which tested DDR3 ECC UDIMMs vs non-ECC UDIMMs, the ECC RAM was barely noticeably slower than the non-ECC RAM. At any rate, the generally accepted figure is 2% slower. I can give that up for the order of magnitude improvement in memory failure rates.

 

Has this been confirmed by Dell? Does this mean that if I buy a 7730 with i9 now, will I be able to add 3200MHz RAM at a later date or would the machines be slightly different? If not it does make sense to wait a month or two.

What exact benefits would there be between HyperX Impact 3200MHz DDR4 CL20 SODIMM and HyperX Impact 2666 MHz DDR4 CL15 SODIMM?

Most of my work is with high resolution images, HD and 4K video.

 

In memory-intensive workloads—precisely what you've mentioned—video/media processing and transcoding, photo processing, layering, doctoring, you will experience a slight increase in performance using the latter memory kit, as the CAS latency of the former kit really is rather poor for such a high bandwidth.

The actual latency (in nanoseconds) is calculated by 1 ÷ (RAM speed in MHz ÷ 2 ÷ 1000) × CL number.

So the 3200 MHz kit has 12.5 ns of latency, while the 2666 MHz kit has ~11.25 seconds of latency.

 

True latency in ns for DDR is [1 ÷ (MHz/2)] x CL x 1000. Easier way to compare two modules is MHz/CL and whichever is bigger has the higher effective bandwidth, if they're the same then the higher MHz one is better.

 

Thanks. I am now at the screen decision for the 7730.

The 17.3" UltraSharp UHD IGZO 3840x2160 sounds nice but I wonder whether going for the 17.3" UltraSharp FHD IPS 1920x1080 would be the better overall option? It allows me to have WWAN, will still look great for all uses and while I do edit 4K video from time to time, most of the video I work with is still 1920x1080.

What do you think? Any comments or thoughts you would like to share?

The exact 1920x1080 screen specs are: 17.3" UltraSharp FHD IPS 1920x1080 AG, NT, Cam/Mic, w/Prem Panel Guar 72% color gamut.

Thank you.

 

Two advantages with the 4K display:

1. It has a wider colour gamut than any of the 1080p displays;

2. Since it's so pixel-dense, you can work on even your 1080p videos at 100% zoom, and still have space for the user interface.

 

Fill in the equation with realistic numbers.
for DDR4-2600 CL19:
[1 / (2600MHz / 2)] * 19 * 1000
= [ 1 / 1300MHz ] * 19000
= 769ps * 19000 = 14.615us

for DDR4-2600 CL17:
[1 / (2600MHz / 2)] * 17 * 1000
= [ 1 / 1300MHz ] * 17000
= 769ps * 17000 = 13.077us

for DDR4-3200 CL19:
[1 / (3200MHz / 2)] * 19 * 1000
= [ 1 / 1600MHz ] * 19000
= 625ps * 19000 = 11.875us


Simply dividing clock frequency by CL timing number does return a number with correct dimension (MHz divided by dimensionless number is still MHz), but if it didn't it was still a valid thing to do; it'd be known as a figure of merit.
2600 / 19 = 136.842
2600 / 17 = 152.941
3200 / 19 = 168.421
Where a higher figure of merit indicates a better performance.

Edit:
The crucial webpage you linked to list this latency equation:
true latency (ns) = clock cycle time (ns) * number of clock cycles (CL)
That's exactly the same formula, with cycle time instead of frequency. After all, cycle time = 1 / frequency.