Take a look at this excellent review by AnAndTech, right here.
During my hunt for the 3950X, I watched as NewEgg, Amazon, B&H Photo and EBay, all increased their prices. After a little over two weeks, NewEgg had raised their price from $750 to $899 and Amazon is consistently selling them for over $1000. This speaks to two things, the demand for these processors, as well as the supply (from TSMC). It’s hard to fault these companies for raising it $50-$100. However, once you start charging 120% of the MSRP (and more), then I have an issue with it. Problem is, they are still selling them out, so prices won’t drop anytime soon.
Some readers are probably wondering, but what about Intel and the i9 9900KS? Well, there are a lot of reasons why I decided to go with AMD’s 3950X instead. The main reason is the plethora of things I do with my computer, many of which are related to this site. Some examples: Audio & Video encoding/decoding, PSpice, Matlab, SolidWorks, C Lion, RAW Photo Editing, Gaming, VR and more. If you are solely gaming with your PC, the 3950X is probably not the best use of your money. At this time, you’re better off spending less on the CPU (maybe a 3900X or 9900KS) and buying a better GPU instead.
3950X and Memory Bandwidth
One of the biggest concerns for a lot of engineers looking at the 3950X design is how it deals with memory. The CPU has 16 cores that need to be fed by Dual Channel RAM. The absolute maximum memory throughput is 82.963 GB/s. This leaves only 5.2 GB/s for each core, which is only a little more than (Single Channel) DDR2-667! AMD is able to get around this, by using their Infinity Fabric technology. We also have much larger cache levels now: L1 Cache = 1MB | L2 Cache = 8 MB | L3 Cache = 64 MB. Thus, this allows for many instructions
Note on CL14/CL16: CL = Column Access Strobe (CAS). This is the most important memory timing to adjust. Keep in mind, sometimes it cannot be changed at all, it depends on the RAM kit and motherboard. For a good article on memory timings, see this.
- CL: CAS Latency. The time it takes between a command having been sent to the memory and when it begins to reply to it. It is the time it takes between the processor asking for some data from the memory and then returning it.
- tRCD: RAS to CAS Delay. The time it takes between the activation of the line (RAS) and the column (CAS) where the data are stored in the matrix
- tRP: RAS Precharge. The time it takes between disabling the access to a line of data and the beginning of the access to another line of data.
- tRAS: Active to Precharge Delay. How long the memory has to wait until the next access to the memory can be initiated.
- CMD: Command Rate. The time it takes between the memory chip having been activated and when the first command may be sent to the memory. Sometimes this value is not announced. It usually is T1 (1 clock cycle) or T2 (2 clock cycles)
Below is a performance chart showing the difference between DDR4-3600 CL16 and DDR4-3600 CL14 with the 3950X. You can find the entire article at TechSpot, right here!
You can see that on average, the 3950X gained 6% more performance by changing from 3600 CL16 to 3600 CL14. If you have a RAM kit that uses Samsung “B-Dies” then you will most likely want to run at 3200 MHz CL14, with optimized timings for best performance. Many people have reported this is the case with those chips, regardless of advertised speed. This suggests they may not have been capable of their reported speeds, without a big shift in timings.
Above you can see the slide that caught everyone’s attention during AMD’s presentation, it’s about memory performance/latency vs. the speed of the RAM. With how AMD’s Inifity Fabric works, there are RAM speeds that fall into 1:1:1 mode and speeds that fall into 2:1:1 mode (note the latency in 2:1:1 mode). There are actually three separate clocks involved here, of course you have UCLK (Memory Clock). You also have the MCLK (Memory Controller clock), and the FCLK (Infinity Fabric Clock). DDR4-3733 is the highest speed that will remain 1:1:1 mode. According to AMD though, this small shift in latency can be easily overcome.
Enabling 2:1 mode crosses clock domain boundaries, imparting a DRAM latency penalty of approximately 9ns that may be overcome with additional memory clocks, higher CPU frequencies, or sub-timing adjustments – AMD
Below you can see two of the most popular benchmarks,
As you can see, AMD has made significant strides since the 2950X Threadripper came out! We can see a consumer part (3950X) besting a professional part (2950X). AMD truly blurs the line between consumer and professional parts, with their 3950X. This is exactly what has put Intel into a position they had to cut prices across the board, multiple times.
ASUS STRIX X570-E Motherboard
When I went to look for motherboards I was pleasantly surprised to find a pretty good selection of brand new 5th generation Ryzen boards. I ended up going with the ASUS X570-E mainly because it had a $75 deal for Black Friday and the reviews are very good. There are better performing motherboards out there, but it’s hard to beat the X570-E for a sub $300 motherboard.
I really wanted to go with ASRock, I had been using them since 2004. My current desktop has an ASRock Z77 OC Formula, which is an excellent motherboard all around. The ASRock X570 Taichi is definitely a good board, but it wasn’t marked down at the time, like the X570-E.
There are a lot of good reviews out there of the ASUS X570-E, but one of the best is by Luke Hill over at KitGuru. You can find his full review right here.
As you can see, the ASUS Strix X570-E trades blows with some of the top boards from Gigabyte, MSI and ASRock. This kind of behavior is expected, as board manufacturers solve design issues in many different ways. Overall though, for the money, it’s hard to beat this offering by ASUS (when it’s on sale). If the ASRock X570 Taichi was the same cost, I’d probably have bought that one.
I’ll be adding a little more to this as time goes on and once the 3950X is here, I’ll do a full review with a dedicated page or two. ETA on my 3950X is roughly right around New Years, so stay tuned!