One of the really interesting things about Ryzen, especially when looking at it from a reviewers/testers perspective, is the fact that AMD can have a big influence on performance and/or compatibility rolling out so called AGESA upgrades though BIOS updates from motherboard vendors.

I’ve been lucky enough to start toying with Ryzen months before the actual launch took place and already by then it was clear as day that this platform would require a lot of additional tweaking from AMD as well as from the motherboard vendors if we were to ever see what it’s really capable of. Back then memory compatibility was the least of the problems the platform hat do deal with. Even getting the thing to post was already a challenge of its own not to say changing values in the BIOS. In fact setting hex-values in the BIOS is something I used to do when I was a teenager, which is quite a while ago - actually.

Long story short, since I very first touched an AMD Ryzen platform quite a lot has changed and in my personal opinion we’re nearing a state, where, on the memory side we’re leaving the beta stage. The current beta BIOS for the ASUS Crosshair VI Hero (version number 9945) helps a lot with memory compatibility and to a certain extent also with performance. Nevertheless I don’t even want to touch the topic of overclocking the cores. To say the least, I’m waiting for a few more AGESA updates until I touch air/water cooled overclocking of Ryzen again.

But to finally come to the point of explaining what I’ve done here. As you might have already gathered from what I’ve written so far, I went ahead and got the latest beta BIOS for the ASUS Crosshair VI Hero motherboard, flashed it and checked whether or not memory compatibility got any better. To my positive surprise it actually is better (my expectations are very low these days). Much better in fact. Running stock clocks on the CPU, the two 8GB DDR4-3200MHz Corsair Vengeance LED (Samsung B-Die) sticks I used for testing even passed DDR4-3333MHz with ease and with ease means just setting another memory frequency from the drop-down in the BIOS (screenshot at the bottom). Apart from having a look at compatibility I ran SuperPi 1.5 XS - 32M as well as Cinebench R15 while monitoring power consumption in order to determine how big the performance gain is in relation to the increase in power consumption. That said it’s about time to show you the graphs. If you’re missing the full specs regarding the test system look at the bottom of this page. I moved them there to not have too many distracting objects before the graphs.

CPU Clk	Vcore load SuperPi 1.5 XS - 32M (Volt)	Timings	Mem Clk (MHz)	P. cons. idle (Watt)	P. cons. SuperPi 1.5 XS (Watt)
4.1 GHz on 2 cores	1.199	15-15-15-36-51 1T	2133	54	76
4.1 GHz on 2 cores	1.199	15-15-15-36-56 1T	2400	59	81
4.1 GHz on 2 cores	1.199	16-16-16-39-62 1T	2666	59	82
4.1 GHz on 2 cores	1.395	16-16-16-39-66 1T	2800	60	83
4.1 GHz on 2 cores	1.395	16-16-16-39-69 1T	2933	63	85
4.1 GHz on 2 cores	1.395	16-16-16-39-72 1T	3066	63	87
4.1 GHz on 2 cores	1.395	16-16-16-39-75 1T	3200	61	88
4.1 GHz on 2 cores	1.395	18-21-21-49-78 1T	3333	63	87

This graphs shows the performance gains in SuperPi 1.5 XS - 32M when simply increasing the memory frequency by selecting the different speeds via the drop-down in the BIOS in relation to the increase in power consumption. All timings were set automatically and they’re listed below the graph alongside some other data such as voltages and core frequencies.

Increasing the memory speed from the stock DDR4-2133 to DDR4-3200 resulted in a 3.67% performance gain. This is something but not really the world especially since those not-even-4% come at quite a cost. Bumping the memory frequency has a pretty drastic effect on power consumption of the entire system, which went up by 15.79%.

CPU Clk	Vcore load Cinebench R15 (Volt)	Timings	Mem Clk (MHz)	P. cons. idle (Watt)	P. cons. Cinebench R15 (Watt)
3.7 GHz on 8 cores	1.199	15-15-15-36-51 1T	2133	54	158
3.7 GHz on 8 cores	1.199	15-15-15-36-56 1T	2400	59	167
3.7 GHz on 8 cores	1.199	16-16-16-39-62 1T	2666	59	169
3.7 GHz on 8 cores	1.395	16-16-16-39-66 1T	2800	60	170
3.7 GHz on 8 cores	1.395	16-16-16-39-69 1T	2933	63	173
3.7 GHz on 8 cores	1.395	16-16-16-39-72 1T	3066	63	174
3.7 GHz on 8 cores	1.395	16-16-16-39-75 1T	3200	61	175
3.7 GHz on 8 cores	1.395	18-21-21-49-78 1T	3333	63	175

This graphs shows the performance gains in Cinebench R15 when increasing the memory frequency by selecting the different speeds via the drop-down in the BIOS in relation to the increase in power consumption. All timings were set automatically again and they’re listed in the below the graph abvoe alongside some other data such as voltages and core frequencies - again.

Increasing the memory speed from the stock DDR4-2133 to DDR4-3200 resulted in a 1.62% performance gain. Considering that the power consumption of the entire system went up by 10.76% this is rather disproportionate.

Conclusion

Seeing that AMD, as well as the motherboard vendors, are working hard on improving memory compatibility as well as performance, even after the launch of a product, is certainly a great thing to witness. Most importantly it shows that AMD is determined making Ryzen the mature platform it should be and it deserves to be. Nevertheless I’m disappointed with the performance increase related to the new possibilities we're all receiving by being able to set high memory frequencies easily. Going all the way from DDR4-2133 to DDR4-3200 and only seeing a 1.62% higher score in Cinebench R15 while SuperPi 1.5 XS - 32M benefits by 3.67% sobering – to say the least – is really not what I was hoping for. Putting this in perspective to the increase in system power consumption it almost becomes ridiculous, since this value went up by 10.76% in the case of Cinebench R15 and 15.79% regarding SuperPi 1.5 XS - 32M.

Whether or not you want to run your memory at the highest frequency possible; I’d like to leave that decision to. After having read this article you certainly have the necessary data to take an elaborate decision. Peronsally I run stock values, since I’m not willing to pay extra by the end of the month for an insignificant increase in performance – but hey, that’s my opinion/decision. What is yours?

Specifications Test System

Carte Mère	ASUS Crosshair VI Hero (Beta BIOS Version 9945)
CPU	AMD Ryzen 7 1800X @ Default
Mémoire Vive	Corsair Vengeance LED DDR4-3200C16 (CMU32GX4M4C3200C16R) @DDR4-2133C15 @DDR4-2400C15 @DDR4-2666C16 @DDR4-2800C16 @DDR4-2933C16 @DDR4-3066C16 @DDR4-3200C16 @DDR4-3333C18
Software	SuperPi 1.5 XS - 32M Cinebench R15
Cartes Graphiques	nVidia GeForce GTX 980
OS	Windows 10 x64
Disque système	OCZ Storage Solutions Arc 100 240GB
Alimentation	Seasonic Platinum SS-1000XP / 1000 Watts