AMD NAVI and Ryzen COMPUTEX Re-Cap and Analysis

David_Schroth

Administrator
Staff member
Joined
Apr 23, 2019
Messages
1,279
Points
113
In this editorial, Brent Justice recaps the AMD COMPUTEX announcements and gives his opinion on where NAVI and Ryzen are headed for the gaming enthusiast.
 
Great article.

If Zen 2 performs as well as they are hyping, then I get the feeling it's going to be the FX 64 of our generation. It's been a long time coming.
 
I wonder if performance on Zen2 will be as significantly affected by memory speed as Zen and Zen+? I hope that gets tested out once these are available for review.
 
Nice see AMD iterating hard on this already pretty solid design. Can't wait to see how it really performs.
 
jardows said:
I wonder if performance on Zen2 will be as significantly affected by memory speed as Zen and Zen+? I hope that gets tested out once these are available for review.
Click to expand...

Probably. Although, I suspect over a certain speed you hit a point of diminishing returns quickly. Lower latency is probably more important past a certain speed. I found this out doing some recent testing on Ryzen and Threadripper. Granted, you do want memory speeds over DDR4 3000MHz, but this is true of Skylake and newer architectures from Intel as well. The difference, is that Intel systems don't care much about lower latency and AMD systems do.

For example, my Threadripper 2920X actually performs better in most tests using DDR4 3000MHz modules which are CL16 modules than it does using DDR4 3600MHz modules at 3400MHz which have timings of 18,19,19,39@1T.
 
I've heard memory latency is important as well with Ryzen 2, like you said, once you get to a certain bandwidth point, then latency should not be overlooked.
 
I'm actually doing some testing on that with existing Ryzen 2000 series CPU's.
 
Brent_Justice said:
That would make for a great review, I'm interested in the results
Click to expand...

That's actually the plan. This is something I wanted to investigate in detail, but never did. It's been my observation that over the years, Intel's CPU's have craved bandwidth and that tighter timings made little impact to performance on Intel platforms. AMD has been somewhat opposite to Intel in this regard. That is, its CPU's are less effected by bandwidth but tighter timings had a sometimes substantial impact. Obviously, we know Skylake and newer Intel CPU's do benefit from RAM speeds over 3000MHz, the same as Ryzen does. but where is the point of diminishing returns, and at which point does latency come into the picture?

While I was doing the first motherboard review for this site, I tested Threadripper and switched out my DDR4 3000MHz modules with tighter timings for DDR4 3600MHz modules with ****ty timings. I actually lost performance in some tests running at DDR4 3200MHz or even higher speeds. That's what prompted me to look deeper into the issue. So I've got a Ryzen 2700X on the bench and an Intel Core i5 9600K system for comparison. I'm going to use the lowest latency RAM I've got and compare it against the highest bandwidth RAM I've got and see what happens in various applications, benchmarks and games.

It will be interesting to see what effect this has in different tests and I want to revisit the topic with the Ryzen 3000 series when we get one.
 
Dan_D said:
That's actually the plan. This is something I wanted to investigate in detail, but never did. It's been my observation that over the years, Intel's CPU's have craved bandwidth and that tighter timings made little impact to performance on Intel platforms. AMD has been somewhat opposite to Intel in this regard. That is, its CPU's are less effected by bandwidth but tighter timings had a sometimes substantial impact. Obviously, we know Skylake and newer Intel CPU's do benefit from RAM speeds over 3000MHz, the same as Ryzen does. but where is the point of diminishing returns, and at which point does latency come into the picture?

While I was doing the first motherboard review for this site, I tested Threadripper and switched out my DDR4 3000MHz modules with tighter timings for DDR4 3600MHz modules with ****ty timings. I actually lost performance in some tests running at DDR4 3200MHz or even higher speeds. That's what prompted me to look deeper into the issue. So I've got a Ryzen 2700X on the bench and an Intel Core i5 9600K system for comparison. I'm going to use the lowest latency RAM I've got and compare it against the highest bandwidth RAM I've got and see what happens in various applications, benchmarks and games.

It will be interesting to see what effect this has in different tests and I want to revisit the topic with the Ryzen 3000 series when we get one.
Click to expand...


Can't say I've seen much of anything on this in a while. Corsair did a quick PR video several years ago, but with a new uarch in Zen, it's time to revisit that and see what's up.
 
alxlwson said:
Can't say I've seen much of anything on this in a while. Corsair did a quick PR video several years ago, but with a new uarch in Zen, it's time to revisit that and see what's up.
Click to expand...

I thought so too.
 
Dan_D said:
... So I've got a Ryzen 2700X on the bench and an Intel Core i5 9600K system for comparison. I'm going to use the lowest latency RAM I've got and compare it against the highest bandwidth RAM I've got and see what happens in various applications, benchmarks and games.

It will be interesting to see what effect this has in different tests and I want to revisit the topic with the Ryzen 3000 series when we get one.
Click to expand...

Now that's content of the sort that makes one want to contribute $$. (Which I have just done.) Looking forward to the eventual results.
 
I'm definitely going to get that done. However, the scope of that has extended beyond what I originally intended as far as testing goes. Its necessary to come to meaningful conclusions. Beyond that, my cooling system sprung a leak and drenched my Intel test system. So I had to take that apart and clean it up. It was pretty bad, but it could have been worse. ASUS' thermal armor actually protected the board from getting more wet than it did.

Unfortunately, much of the fluid ended up in the CPU socket itself. Everything worked, but I had to order new tubing and fittings as well as replenish my supply of fluid. And non-conductive fluid does work. I can confirm this. The system more or less worked while wet, albeit excluding BSOD'ing from the CPU getting too **** hot. The video card got soaked too. Fortunately, I hadn't thrown my one 2080 Ti in it yet.
 
You must log in or register to reply here.
Become a Patron!
Back
Top