AMD Ryzen 9 3900X CPU Review New BIOS Performance Tested

Brent_Justice

Administrator
Staff member
Joined
Apr 23, 2019
Messages
792
Points
93
AMD Ryzen 9 3900X CPU Review New BIOS Performance Tested - We talk about clock speeds and AMD Ryzen 3000 series, how it works, and clear up some information about launch reviews of AMD Ryzen 3000 series CPU performance. We also re-test performance on our AMD Ryzen 9 3900X with a brand new BIOS from MSI to see if Precision Boost 2 clock speeds improve and or change performance. It's a great follow-up to our initial launch review and clears up this issue about potentially low clock speed and performance results at launch time. This is a must read.
 
Wait, so PPT, EDC and TDC are locked? Anyway to know what they are before buying a motherboard so I can game the system :b I really don't care if I eat through power, my electric bill is like 20$ a month and I can afford a few more dollars on power. I just want maximum performance on these puppies. Might wait until the 3950x launch when things seem mature and I can be sure I am getting what I want/need.
 
Wait, so PPT, EDC and TDC are locked? Anyway to know what they are before buying a motherboard so I can game the system :b I really don't care if I eat through power, my electric bill is like 20$ a month and I can afford a few more dollars on power. I just want maximum performance on these puppies. Might wait until the 3950x launch when things seem mature and I can be sure I am getting what I want/need.

No, they are set in stone for the CPU under Precision Boost 2. Those are the values you cannot change. But, when using Precision Boost Overdrive, the motherboard values are used instead of the CPU's OEM programmed values. However, you can access these values and raise them through the UEFI BIOS and through Ryzen Master. In other words, they are locked by default, but overrides exist.
 
So the Destiny 2 fix requires a BIOS level patch?

That seems odd to require that low level of an update for a single software application, and that no other software applications have reported similar issues (or maybe they have they just haven't got the press?)
 
So the Destiny 2 fix requires a BIOS level patch?

That seems odd to require that low level of an update for a single software application, and that no other software applications have reported similar issues (or maybe they have they just haven't got the press?)

Yes. Forbes reported the story on that originally stating that this is what AMD told them. It wasn't just Destiny 2 that didn't work, evidently there were issues with some Linux distros as well.
 
Thanks for the follow up write up. A shame there wasn't a more drastic change for the better but ultimately it just sounds like the platform needs more time to mature and hopefully it does. Still seems like a great CPU but it'd be nice to see it go further.
 
Thanks for the write-up. Great quality as always.

I have to say, I'm a little bit peeved at AMD, changing the definition of advertised boost clocks to essentially be a max, not a guarantee.

This is different than how these things have ever worked before to my knowledge. Certainly not what I expected.
 
Thanks for the write-up. Great quality as always.

I have to say, I'm a little bit peeved at AMD, changing the definition of advertised boost clocks to essentially be a max, not a guarantee.

This is different than how these things have ever worked before to my knowledge. Certainly not what I expected.
Yeah it's weird as it's almost marketing speak. They could advertise any boost clock if it isn't guaranteed because you can put the CPU under LN2 and it may hit it once launching a single threaded JS task for a website.

I do like how the CPU basically pushes itself to the limits so we get maximum performance out of box, but I can't run my CPU on LN2 all day and water cooling doesn't seem to be giving the uplifts it once did.

Gamer's Nexus did a nice video on PBO explaining why its basically pointless and the boost clocks will never be hit in real world scenarios.
 
Yeah it's weird as it's almost marketing speak. They could advertise any boost clock if it isn't guaranteed because you can put the CPU under LN2 and it may hit it once launching a single threaded JS task for a website.

I do like how the CPU basically pushes itself to the limits so we get maximum performance out of box, but I can't run my CPU on LN2 all day and water cooling doesn't seem to be giving the uplifts it once did.

Gamer's Nexus did a nice video on PBO explaining why its basically pointless and the boost clocks will never be hit in real world scenarios.

The thing is, you can actually hit those boost clocks. The reason why we didn't see them in the initial article was due to the AGESA code. So technically, they aren't wrong. Intel never expressly guaranteed boost clocks either as far as I know. Granted, they are pretty good at achieving them because those CPU's can generally overclock well in excess of the boost clock values.
 
That is a well written piece. Also seems like all generations of AM4 motherboards are having similar issues which must put a strain on the bios making teams dealing with not only the new mother boards but all A4 models in the last 3 years.

I have a 3900x on order, will try it on the ASUS Croshair 6 Hero (Dan's favorite motherboard ;), probably not). If that does not work out too well then looking at a Gigabyte Aorus Master, X570 - maybe I should order that now . . .
 
That is a well written piece. Also seems like all generations of AM4 motherboards are having similar issues which must put a strain on the bios making teams dealing with not only the new mother boards but all A4 models in the last 3 years.

I have a 3900x on order, will try it on the ASUS Croshair 6 Hero (Dan's favorite motherboard ;), probably not). If that does not work out too well then looking at a Gigabyte Aorus Master, X570 - maybe I should order that now . . .

The Crosshair VII's are fine. I went back and looked, it was the Crosshair VI Hero (X370) that I hated. It is one of two motherboards I've reviewed in 14+ years that failed to complete the review testing.
 
The Crosshair VII's are fine. I went back and looked, it was the Crosshair VI Hero (X370) that I hated. It is one of two motherboards I've reviewed in 14+ years that failed to complete the review testing.

I get the impression (I have no personal experience) that many of the X370, B350 and B320 motherboards were pretty bad. Something to do with the board makers not being sure AMD had a serious contender with Ryzen, and thus being unwilling to invest a lot of resources in developing solid boards.

Actually. I lied. I built my stepson a system with a Ryzen 5 1600x and an MSI B350 Tomahawk (Microcenter Bundle Special) in December 2017, so I do have a little bit of experience.

The Tomahawk doesn't seem like a bad little board.
 
I get the impression (I have no personal experience) that many of the X370, B350 and B320 motherboards were pretty bad. Something to do with the board makers not being sure AMD had a serious contender with Ryzen, and thus being unwilling to invest a lot of resources in developing solid boards.

Actually. I lied. I built my stepson a system with a Ryzen 5 1600x and an MSI B350 Tomahawk (Microcenter Bundle Special) in December 2017, so I do have a little bit of experience.

The Tomahawk doesn't seem like a bad little board.

So, here is the truth behind that. There are a few reasons why so many AM4 motherboards are kind of "bad". It comes down to a wide range of things. First, Intel moved up Z270's launch date to interfere with AMD's Ryzen 1000 series and AM4 (X370) launch. This strapped design teams and motherboard vendors for time. Naturally, they concentrated on the Z270 launch which was first due to the new time table.

Secondly, expanding on the earlier point, your right. Motherboard manufacturers had to choose where to hedge their bets and they went with Intel. Resources and money were not spent on developing as many AM4 models, or building them to higher standards. If they didn't sell and they cost allot, the motherboard makers would be out more money. The boards aren't necessarily ****ty, but motherboard partners built less expensive motherboards as they can always sell budget offerings and had successfully sold AM3/AM3+ in the past even though Bulldozer was basically a huge steaming turd. Sometimes you need to build a computer and have no money, so you compromise. Sometimes your building for friends or relatives and either they have no money, or you don't want to spend a ton of money on a computer for someone else. Whatever the reason, loss leaders can always sell.

That's why you see so many bottom dollar B350 boards with questionable VRM's. Also, AMD doesn't have the control over the motherboard partners that Intel does. There was no reason at the time to build AM4 boards at that price point that were capable of delivering 200 amps to a 16c/32t CPU. The only reason why those boards will work at all, is because AMD managed to keep the Ryzen 3000 series within that same 105w TDP as the earlier chips. So at stock speeds, you should be OK on most boards. Its up to the motherboard makers to update the legacy boards with a Ryzen 3000 series compatible BIOS. They may choose not to do so on borderline designs. We'll have to see.

The same thinking led a different direction with Threadripper. Motherboard makers weren't sure that Threadripper / X399 would succeed, and that platform couldn't be built cheaply. Motherboard makers decided to build good designs, but restrict their design teams to one or two high end models and take a wait and see approach. Obviously, AM4 and TR4 were far more successful than initially anticipated and we have a ton of B350, X470, B450 and X399 options out there now.

X570 is a different animal for a multitude of reasons. AMD now has more clout with the manufacturers to say: "We want you to build X570 motherboards as premium option." The technology behind it also has more strict requirements and for enthusiasts, better VRM's are required. Basically the VRM's are beefy on all X570 motherboards and the PCB's are far more expensive. AMD opted to go with newer more advanced materials than increase PCB thickness using conventional materials. So there is added cost there that will come down eventually. The actual X570 chipset itself is more expensive as well. It's derived from the I/O die of the Ryzen 3000 series CPU itself. It's a power hungry 15w part that needs more power itself. I'd bet most if not all X570's use 32MB BIOS chips too. So the costs keep going up.

Ryzen is obviously selling well and AMD has serious momentum right now. Hence the shake up with X570 over earlier and more budget conscious X470 designs.
 
Last edited:
The Crosshair VII's are fine. I went back and looked, it was the Crosshair VI Hero (X370) that I hated. It is one of two motherboards I've reviewed in 14+ years that failed to complete the review testing.
Lol, definitely a board to love to hate or is it hate to love? Anyways have a 3900x to go into her now. If it does not work out then an update will be needed. Now please update your x570 reviews pace sir.
 
3900x with all cores loaded using Aida64 stress test, all cores clock at 4.1ghz with some variations at 1.36v. Bios is using PB only. Doing single thread on Cinebench 20, up to 4.525 ghz depending upon core.

So far no issue using D.O.C.P for ram at 3200mhz 4 sticks 32gb

PCIE4 works with the 5700XT.

Anyways this was almost too easy for this older generation board. Basically take out the old CPU, pop in new with current bios. Turn on. In the bios I set pcie for all items pcie 3 to pcie 3 and pcie 4 for the 5700XT in the 16x slot.

Much more testing and experimenting needed.
 
Nice. I haven't had the time to try the Ryzen 3000's on an older motherboard, but that will happen soon.
 
The newest bios 7302 for the C6H, what a night and day difference! 3900x, 4 dims GSkill 3200 CAS 14 (BDie) running at 3733, passed MemTest Pro for over 500% test run, cold boot problem is also solved for higher ram speeds. A very happy camper so to speak. Looks like AMD 4 dim configuration problems are much better with Zen 2/Ryzen 3.

3733memTest.jpg

Been gaming with PCIe 4 for a Radeon 5700 XT Lisa Sue flawlessly. 3dMark PCI Express benchmark - 25.19GB/s
http://www.3dmark.com/pcie/26830

I am tempted to get a PCIe 4 SSD now and try that out and replace the ill conceived Intel 600p which regrettably I am using as a boot drive.
 
I'm really surprised by your results. I wouldn't think they are typical given that AMD flat out says in its reviewer's guide that you can only expect 2933MHz or so with 4x DIMMs.
 
They lied! :) Thank God. ASUS just release a kick *** bios is probably the answer and really all the memory stuff is on the CPU which determines capability as long as signalling to the dims are good - why not? On the 2700 My top stable was 3200mhz, not much higher than 2933 in what AMD said, which was probably conservative. Being a T-Topology board probably helps but the new C8H is also doing 4 sticks at 3733 and that board is daisy chain - so for these speeds the topology probably does not make much difference if the traces are well done to the memory sticks.

Which begs the real question - what advantage if any are the much higher ram speeds, 3800mhz plus when you are no longer at 1:1 ratio between the ram speed and infinity fabric? It actually looks like performance goes down going over the 1:1 ratio. Sweet spot for Ryzen 3 looks to be the 3600-3733 memory speeds. So those just going after ram speed numbers, ignoring any real benefit, a daisy chain motherboard will probably get you there - most of the X570 boards except for ASRock.
 
Last edited:
Something else unique to the ASUS C8H is it is a 8 layer board which could potentially give better signalling to the ram slots as well as PCIe slots. It will be interesting how the 6 layer boards like the Gigabyte Master does with ram speeds not only two slots but all four. I may get a new motherboard for the 3900x (maybe 3950x if TR is way extreme in cost this time around) and put the 2700 back into the C6H and retire the FX 9590 rig.
 
Become a Patron!
Back
Top