News Gigabyte's B550 Motherboard Is PCIe 4.0 Heaven

COLGeek

Cybernaut
Moderator
Courtesy and respect are "must haves" folks. Civility is an absolute requirement.

Either be courteous, or remain silent.

Attack ideas, with facts and sources. Do not attack people.

Now, lets try this again.

Oh yeah, NEVER feed trolls. Just report them. 'Nuf said.
 

InvalidError

Titan
Moderator
That may be true, but we've never yet seen an affordable board with 3 x m.2, so better keep expectations in check.
16 phases aren't going to be cheap either. This is likely going to be one of the top-10 most expensive B550 boards.

I'd be more interested in finding out where $100-ish B550 boards are going for VRM and how they will split the chipset's limited PCIe/SATA budget.
 
I'd like to see the table of mutually exclusive functions on those boards.
Me too....I'm sitting here wondering how you'll get 3 PCIe gen 4 NVME's functioning since the B550 chipset furnishes only Gen 3 lanes to devices.

I'm thinking the GPU socket will be lowered to 8x Gen 4 lanes. That's effectively the same bandwidth as 16x Gen 3 lanes so a Navi GPU on the board will work just the same as it would on any 300 or 400 series board.

But more concerning to me is let's say you have three Gen 3 NVME's. Does this mean you'll HAVE to give up the 16x lanes to your GPU since they're going to be steered to the NVME's even if to be initialized at a mere Gen3? Can the CPU initialize 1/2 the GPU lanes at gen 3 and 1/2 at gen 4...will your Navi be gimped to 8x Gen3 like it or not?

Or will the board have smart enough PCIe steering to pull lanes from the chipset in that case to spare gimping the GPU? Is that even possible? I can't imagine the PWA layout challenges that would entail while keeping costs in line.
 
Last edited:

InvalidError

Titan
Moderator
Or will the board have smart enough PCIe steering to pull lanes from the chipset in that case to spare gimping the GPU? Is that even possible?
Possible, yes. Practical, no.

The two extra NVMe 4.0 slots necessarily come from splitting the x16 GPU slot into x8x4x4 and each PCIe host controller has independent clock, different speeds for each slice shouldn't be an issue. For boards that don't rob lanes from x16, an extra NVMe x4 slot would eat four out of 10 possible chipset PCIe lanes and four of those are shared with SATA, doesn't leave much PCIe budget for PCIe slots on an ATX-size board. I would have liked B550 to have two extra flex IO lanes so having a 3rd SATA device didn't mean having to give up on an x4 secondary NVMe or PCIe slot.
 
Possible, yes. Practical, no.

The two extra NVMe 4.0 slots necessarily come from splitting the x16 GPU slot into x8x4x4 and each PCIe host controller has independent clock, different speeds for each slice shouldn't be an issue. For boards that don't rob lanes from x16, an extra NVMe x4 slot would eat four out of 10 possible chipset PCIe lanes and four of those are shared with SATA, doesn't leave much PCIe budget for PCIe slots on an ATX-size board. I would have liked B550 to have two extra flex IO lanes so having a 3rd SATA device didn't mean having to give up on an x4 secondary NVMe or PCIe slot.
Those sound like similar issues with B350 and B450...and why I felt they're more suitable for mATX applications. But mATX is in disfavor, for some reason, in AM4. So they do strange things, like leaving large swaths of open board space with no PCIe sockets or gimped-but-showy 16x4 PCIe sockets that disable anyway, becoming completely useless, as you add other devices just to make a board that fills up an ATX case.

I really think they should leave the 3 by gen 4 NVME solutions to the X570 line and keep B550 in the somewhat more budget realm. I totally can see people buying this and being surprised to see the compromises when they start populating all the NVME's. I don't know why they have to screw up the obvious product segmentation opportunities that the two chipsets offer.
 

tlapier

Honorable
Jan 26, 2015
6
2
10,515
I wish they wouldn't push PCI 4.0 boards yet - better they be ignored. PCI 5.0 is right around the corner and PCI 4.0 popularity would just delay its release.
 

InvalidError

Titan
Moderator
Those sound like similar issues with B350 and B450...and why I felt they're more suitable for mATX applications.
True, they are all IO-anemic for ATX-sized boards. Intel's H-series has 30 total HSIO lanes split between USB3.x, SATA, PCIe and LAN with enough flexibility in-between for most people to find a no-compromise ATX board configuration. AMD should really add at least four in B650. The CPU already provides a decent complement of core IO, the USB10G chipset lanes are kind of superfluous and could be replaced with PCIe/USB3.x flex IOs. The two fixed SATA and two fixed PCIe lanes could be made SATA/PCIe flex. Add two of those PCIe/SATA lanes on top, remove SATA functionality from two lanes in the x4 PCIe host and that would already be a whole lot better with as little as two extra total IOs by adding more flex.
 

alextheblue

Distinguished
I'd like to see the table of mutually exclusive functions on those boards.
Does reduced lanes on that first slot count? I don't see any reason any of that HAS to be mutually exclusive. If those M.2 slots are all truly 4.0 you're looking at x8, x4, x4, x4 all off the CPU. The other PCIe slots could very well be 3.0 fed off the chipset. Would you want to USE multiple GPUs with it? Not really (then again mGPU is all but dead now regardless). But that doesn't necessarily mean they wouldn't all function at the same time. We'll find out later on, I guess.
True, they are all IO-anemic for ATX-sized boards. Intel's H-series has 30 total HSIO lanes split between USB3.x, SATA, PCIe and LAN with enough flexibility in-between for most people to find a no-compromise ATX board configuration.
The flexibility is nice for sure, but how many lanes are feeding the whole shooting match? These consumer platforms are all pretty limited in overall I/O bandwidth. It has really become one of the largest factors separating the top-end consumer and true HEDT platforms. Speaking of B650, the first thing they need is a PCIe 4.0 uplink! That was the biggest disappointment with B550, although I guess that would put it too close to X570 for comfort. Then again, for what I need B550 is plenty. All I need is a board with 4.0 on the first x16 slot and first x4 M.2. The second M.2 and everything else can be 3.0 off the chipset.
 

InvalidError

Titan
Moderator
Does reduced lanes on that first slot count? I don't see any reason any of that HAS to be mutually exclusive. If those M.2 slots are all truly 4.0 you're looking at x8, x4, x4, x4 all off the CPU.
Depending on how games and GPUs move forward, 4.0x8 might not cut it for much longer and people will end up wanting the full x16 for GPU.

For budget buyers, splitting the x16 may not be an option since they will need it to keep GPU performance from imploding when it runs out of VRAM. Of course, this point may be moot since few budget buyers are going to bother with fancy boards like the Aorus Master offering any other option in the first place and most people who do aren't going to buy 6GB-under GPUs either.

The flexibility is nice for sure, but how many lanes are feeding the whole shooting match? These consumer platforms are all pretty limited in overall I/O bandwidth.
It is likely safe to say that both Intel and AMD realize that as with any other network, it is bad news when a single end-point (3.0x4 SSD) can nuke the node's uplink. I'd be very disappointed if mid-range chipsets don't get a 4.0 uplink in 2021 if not earlier.

If intensive asset streaming becomes a commonplace thing, I suspect we'll see many people who dump games on a secondary SSD connected to the chipset complain about stutter caused by uplink contentiion with other chipset IO on 4GB/s platforms.
 
I wish they wouldn't push PCI 4.0 boards yet - better they be ignored. PCI 5.0 is right around the corner and PCI 4.0 popularity would just delay its release.
I honestly wonder if PCIe gen 5 is even necessary. I mean Gen 4 has been out for a year and nobody's hurting if they don't have it. IMO that's the real reason AMD won't support pre-Zen2 CPU's on X570 and B550 as they Gen 4 attraction alone just wasn't compelling enough.

Gen 3 isn't even fully utilized...you can trim GPU's to 8 lanes of Gen 3 and few users would notice it without running an appropriate benchmark. I have a Gen3 NVME in a Gen2-limited M.2 socket on my B450m motherboard for my data drive and I do not notice the difference during game launch or level loads of the same game running from the system NVME in the Gen 3 M.2. Possibly that's because my GPU is still limited to Gen3 transfers...but when I initialize the system M.2 at Gen2 in BIOS it also makes no perceptible difference in boot-up times for Windows.

That's not to say there aren't applications that will benefit, but for the broader market it doesn't seem any more compelling than Gen 4 is at the moment. Maybe PCIe Gen 5 has other benefits that will be more compelling, but data transfer rates alone doesn't seem to be the one just yet.
 
Last edited:

InvalidError

Titan
Moderator
I honestly wonder if PCIe gen 5 is even necessary. I mean Gen 4 has been out for a year and nobody's hurting if they don't have it.
You don't know what you are missing until you don't have it anymore. Nobody is missing it now because relatively few people have had it so far and even fewer have lost it.

Gen 3 isn't even fully utilized...you can trim GPU's to 8 lanes of Gen 3 and few users would notice it without running an appropriate benchmark.
4GB RX5500 owners are hurting pretty bad from 3.0x8 vs 4.0x8 when the GPU runs out of VRAM. If next-gen entry-level GPUs get 50+% more powerful but are still stuck at 4GB, 4.0x16 will be sorely needed to offset the VRAM deficit.
 
....
when the GPU runs out of VRAM.
...
I really don't think even gen 4 helps much it when it runs out of VRAM. I certainly have experienced it when I had an older 4Gb card; it was slow enough as it was but running it out of VRAM brought frames to a genuine slide show. But then I've not seen any benches doing the same with a modern card, especially on an x570 motherboard with gen 4 GPU.

Agreed, gen 5 could only help...but 8Gb GPU's seems to be pretty much standard for high(er) end GPU's so even then it's probably not going to be all that much of an issue. It's a safe assumption that a customer buying a gen4 (or gen 5 when it comes) capable system will also be getting just such a top-end card. It just seems pointless to me to fix the problem at both ends.

Pretty sure there's some GPU compute applications that will benefit though.
 
Last edited:

InvalidError

Titan
Moderator
It's a safe assumption that a customer buying a gen4 (or gen 5 when it comes) capable system will also be getting just such a top-end card.
That is a dumb assumption to make when B550 is staring you right in the face with PCIe4 for entry-level and Intel likely doing the same when 11th-gen rolls around. PCIe4 on the x16 and main NVMe slot are about to become baseline and will get paired with baseline stuff as more of it becomes available.
 
That is a dumb assumption to make when B550 is staring you right in the face with PCIe4 for entry-level and Intel likely doing the same when 11th-gen rolls around. PCIe4 on the x16 and main NVMe slot are about to become baseline and will get paired with baseline stuff as more of it becomes available.
X570 is not and never was aimed to be budget and the B550 boards we've seen so far aren't looking like budget choices, notwithstanding comments otherwise. I'm not going to assume AMD's making these market moves because they want to drop prices on hardware if they don't have to.

In fact, it's just as dumb to assume AMD doesn't see an opportunity to move up-market and start tapping into the margins Intel's historically enjoyed on both processor and chipsets. I'll wait to see just how many B550 boards come out in the under $100 segment before I change my assumption
 
Last edited:

InvalidError

Titan
Moderator
X570 is not and never was aimed to be budget and the B550 boards we've seen so far aren't looking like budget choices, notwithstanding comments otherwise.
The B550 boards announced "so far" are merely the flagship products for people who don't want X570 but do want premium VRM and heaps of bells and whistles that motherboard manufacturers haven't bothered with on B450.

Normal B550 motherboards will do the same thing that all plain mid-range boards have always done: expose available CPU+chipset connectivity in a hard-wired fashion. Most will cost within $10 of equivalent B450 boards' launch MSRP.
 

alextheblue

Distinguished
Depending on how games and GPUs move forward, 4.0x8 might not cut it for much longer and people will end up wanting the full x16 for GPU.

For budget buyers, splitting the x16 may not be an option since they will need it to keep GPU performance from imploding when it runs out of VRAM. Of course, this point may be moot since few budget buyers are going to bother with fancy boards like the Aorus Master offering any other option in the first place and most people who do aren't going to buy 6GB-under GPUs either.
I wasn't really getting into whether it was a good or bad thing. I was only disagreeing with the notion that things had to be disabled for all of these slots to function. In fact my stated preference was for a single 4.0 x16 slot and x4 M.2, the rest being fed off the chipset (speed isn't a big issue for anything else I'd attach). With that being said, if your GPU is also 4.0 it would be fine for the mid-range, for quite some time.

I'd like to see a future I/O chip from AMD integrate flexible switching capabilities, but again if it is too good it would compete directly with their own HEDT platform.
 
  • Like
Reactions: panathas