All Roads Lead to Rome as Gigabyte Motherboard Spy Photos Surface [Rumor]

cageymaru

Fully [H]
Joined
Apr 10, 2003
Messages
22,061
A spy photo of what might be the first high performance computing (HPC) AMD EPYC Rome motherboard has surfaced. This Gigabyte motherboard is designed for the upcoming 7nm Zen 2 AMD EPYC Rome 64-core processors. It sports PCIe Gen 4 connections in a x16, x16, x8, x16, x16 configuration and PCIe Gen 3 x8 and x16. 8-Channel RAM, four U.2 interfaces, two SATA, two M.2 on the board, dual BIOS, and two BGA spots unpopulated at the IO side of the board, so there is possibly logic going there that GBT did not want to show off. The server board also has a couple of mezzanine slots in what looks to possibly be PCIe x16 and x8 configurations which allow for the installation of all sorts of other cards. We are not sure but we think we spy four high bandwidth Ethernet connections as well. Time for all our server guys to chime in and tell us what we got wrong in the HardForum.
 
Last edited:
I know it's a server board but gotta give Giga credit where it counts. They've pretty much covered all possible card/port needs with this. Pretty much a tradition at this point for them. I know I still love my UD3X79 board for the same reasons.
 
MFW Eight channel
Ooooooooooooooooochimpanzeee2.jpg

Very sexy board.
Anyone know if PCie4 is backwards compatible with 3? Looks same format. Or I guess that's why some are 3 and 4.. they are not.

P.s. cageymaru your link goes to aggregate [H] news articles, not the specific article. E.G. https://www.hardocp.com/news/2018/1...gigabyte_motherboard_spy_photos_surface_rumor
 
MFW Eight channel
View attachment 119918
Very sexy board.
Anyone know if PCie4 is backwards compatible with 3? Looks same format. Or I guess that's why some are 3 and 4.. they are not.

P.s. cageymaru your link goes to aggregate [H] news articles, not the specific article. E.G. https://www.hardocp.com/news/2018/1...gigabyte_motherboard_spy_photos_surface_rumor

they are all backwards compatible.. same reason 3.0 cards work in 2.0 slots, etc.. all it means is that it has the bandwidth/power requirements for that certification.
 
While doing my normal surfing I noticed some other site(s) only have a partial pic of this mobo.

Way to score the centerfold pic Cagey!
 
If you guys find 80 year old grannies to be sexy, then yes, this mobo is sexy.
It looks like something from the 70's.
 
Only a single processor socket? meh!



;-)

there are 2P options as well.. they had one sitting in the demo room at the New Horizon event but wasn't being demo'd :(

256 threads on a single system will be redonkelous.
 
and two BGA spots unpopulated at the IO side of the board, so there is possibly logic going there that GBT did not want to show off.
I like how one of the PCI-E slots is burnt. Nice touch.
I had noticed how lots of stuff on this board had been desoldered. It's clearly been through QA and failed, so my assumption as to the missing components is due to doing postmortem. I'd say it's obvious that the PCIe dying spectaularly is the reasoning for the dissection :p By the burn marks I suspect it was those 4 PCIe Switch (or are they multiplexors?) that are also missing.

Seems like every major IC that was connected to PCIe on the slots-half of the board have been removed (which based on silkscreen text orientation, would be the "top half"). So waaay more than just "2 BGA spots unpopulated" lol I count 10 or 11 total that were removed (there's one near the U2 that looks like it was never populated).



That being said, the two things I'm curious about are: If those are U.2 ports (which I have no reason to doubt it), why are they so vastly different than the one on my MSI X370 Titanium? It's not the fact that mine is plastic vs that one's being metal, but it's the total size and shape. Those just look like the SATA ports, only rotated 180º. For that matter, they look just like the 4 SLINK ports above the PCIe slot.

The other thing is how on earth can that thing have the power delivering capability for an EPYC with only 6 power phases?!?! o_0 Or does the back of the board have further MOSFETs?
I know my Titanium and A88X-G45 have a couple power related chips on the back, though I'm pretty sure they are simply the FET Driver (I think that's the correct term?), so that's why they're not cooled beyond the transfer of the solder pad and dissipation by the board. My other thinking was those +12V CPU plugs weren't delivering 12V and that they were directly connected to the CPU, thus the phases were off-board (inside the PSU perhaps).



So... does NVMe via M.2 automatically get faster over PCIe 4.0?
I would say only marginally, if at all. The only reason it'd get faster would be due to the increased bandwidth, BUT only because that now is helping offset any overhead that on PCIe 3.0 was eating into a drive's overall potential.

For example, I remember that happening when USB3.0 first came out, as some of the higher end USB2 thumb drives plugged into a 3.0 port would see an uplift in their transfer performance by a couple MB/s. It wasn't like the fact it was 2.0 was passing off the signaling to an only-2.0 controller... it was still being handled by the chip that was capable of 3.0, so it was definitely capable of handling speeds in excess of what anything 2.0 could require.

I also seem to recall that it occurred with SATA-III, but that ended up being more a case of the new channeling which meant the cables (wiring) was laid out differently to handle the higher transfer speeds and not produce crosstalk. Side benefit was SATA-II drives using just the cables performed better :D
 
Last edited:
Mid-late 70s is still milf territory IMO lol.
I'm only late 80s myself ;)
PoAAILF, perhaps?? Because remember what Shephard Book said: "I never married.... I'm not a grandpa." :p

(Person of Advanced Age ... ILF)
 
That being said, the two things I'm curious about are: If those are U.2 ports (which I have no reason to doubt it), why are they so vastly different than the one on my MSI X370 Titanium? It's not the fact that mine is plastic vs that one's being metal, but it's the total size and shape.
Those connectors don't look like SFF-8643 at all. Could be iPass/SFF-8087 or something custom.
The other thing is how on earth can that thing have the power delivering capability for an EPYC with only 6 power phases?!?! o_0 Or does the back of the board have further MOSFETs?
A 6-phase with 70A powerstages should sit at ~16W total FET losses at 1V/200A out.
Entirely manageable even with that dinky heatsink as long as it's getting forced air from the cpu shroud.
 
Back
Top