This topic is ment to discuss all about X399 motherboards,
and their vrm designs.
Edit: Users also feel free to share your user experiances on X399 boards with us.
Things like pci-e passtrough and iommu groups, features, UEFI´s, overclocking experiances etc etc.
Yeah that board looks pretty beefy yeah.
I believe something like 19 phase (16+3 phase design) or so?
IR35201 straight 8 phases doubled to 16 with IR3599´s,
and a Second IR35201 or IR35204 providing 3 phases for Soc.
Still need to figure out which powerstages they used this time arround.
But if they used some decent powerstages (which they most likelly did).
Then its an awesome vrm.
MEG has no 10GbE and GB Xtreme has 10 phase (10 true or 5 doubled??) but 50A IR… not 60A.
In practice it should be enough and their cooling is significantly better on paper than the Zenith, but I would have liked to see MSI put their extra BOM into an Aquantia chip rather than the 4xNVME add-in card.
4.0GHz “all core” (x32) looks to require 1.35v? or higher which doesn’t leave a lot of room with 10x50A on the GB Xtreme and the Zenith is a hot mess (literally).
5 phases doubled of course.
There are no pwm´s used on motherboards today that can do more then 8 true phases.
There has been a true 10 phase pwm used on certain x58 boards in the past,
the Voltera VT1165 / VT1185, pwm.
But that pwm cannot be used on modern motherboards today, because its not VRD12 certified.
Yeah, I eventually got to the bottom of that as well. Curiosity got the better of me, so the MEG Creation is on the way for some play-time.
I wasn’t the only one who pointed out the stupidity of trading an on-board Aquantia for the 4xNvme AIB. That just makes zero sense if you are targeting an HEDT work-station market at this point. All of your competitors are putting 10G on-board.
Per Anand the MEG scored poorly on DPC (system/driver latency), but to be honest, at this point, the 2990WX appears to be a chip that you want to avoid running under windows for now. It looks like Windows is really hurting this chip, but I’ll know next week whether that gamble is true.
Well kinda depends really, a 10Gbe nic can be added as an addin card of course.
Since you basically have more then enough pci-e lanes to saturate.
But i kinda agree on a top end board like the MEG, 10Gbe should have been available onboard.
Also dpc latencies can often be fixed with driver updates.
However its kinda annoying to deal with.
The 2990WX also suffers from the fact that only two of the four ccx units have direct memory acces.
So yeah that will allways going to be some kind of a bottleneck.
And yeah as far as Gigabyte is concerned they often use IR3556 50A powerstages on their topend boards.
This has been my assumption since the announcement of 4 dies (necessarily on TR4 gimping 2 dies with no physical memory access)… Looking at windows benchmarks around the 'net I had decided that it was as bad or worse than I had feared with the 7980xe (and my 2x18 core 2696v3 system) not just performing better, but better by leaps and bounds.
What changed my mind was the linux benchmarks seem to be showing MUCH better scaling to all 32 cores for applications well beyond just rendering.
One of the things I’m thinking of trying is setting up an 8-core windows gaming (GPU pass-through) VM and pinning it to the “far” dies and see just how bad it really is (or how much is windows terrible scheduling). Of course, I am going to compare my other engineering apps that scale reasonably well to 16-18 cores. Should have parts by next weekend…
So after having test this the gb Xtreme is awesome. The default fan curve doesn’t even kick in till the vrms are past 110 and it was able to sustain the 2990wx with an extreme overclock for hours on end.
The biggest thing holding the Xtreme back is that it needs dynamic voltage offset. only a few workloads need the volts. The retail 2990wx was stable @ 4.0 on all 32 cores at 1.2750 for most real workloads and a lot of synthetic workloads … 1.3250 for almost every case covered and 1.35 for a few edge cases.
2933 was most stable tho. 3200 was really pushing it on the 2990 … too far imho
Yeah kinda depending on how he exally did his said tests.
And what kind of settings he used for the said overclocks on the different boards.
I kinda have to place some question marks with some of that video´s outcomes aswell.
The Aorus Extreme performing that bad in his tests,
makes me think that he might have messed something up regarding the OC settings.
So I have discovered the default vrm fan profile is supposed to kick in sooner but doesn’t for whatever reason. But if you fiddle with it then save it does apply. It’d probably breathe better without the shroud. But it’s been solid on prolonged indigo tests
Yeah all those blingy shrouds are mainlly for looks,
i would rather remove them.
According to the fan profile, if you mean the fan curve you set in the bios,
then yeah its one of those tiny little things that Gigabyte still needs to work on regarding their bios.