News Reviewer reports RTX 5080 FE instability — PCIe 5.0 signal integrity likely the culprit

[Admin]

Suspected signal integrity losses have forced reviewers to switch to PCIe Gen 4.0.

Reviewer reports RTX 5080 FE instability — PCIe 5.0 signal integrity likely the culprit : Read more

 

[bit_user]

So, PCIe 5.0 ended up being worse than a pointless waste of money - it's downright harmful!

 

[hotaru251]

So, PCIe 5.0 ended up being worse than a pointless waste of money - it's downright harmful!

more like design flaw same as the connector with 40 series.

concept was fine just done in a way that failed.

 

[Gururu]

Yikes.

 

[alceryes]

So, PCIe 5.0 ended up being worse than a pointless waste of money - it's downright harmful!

No. The PCIe 5.0 standard and design is fine.
Current speculation is that NVIDIA's multi-PCB design introduces too much noise, messing with signal integrity.

 

[alceryes]

more like design flaw same as the connector with 40 series.

concept was fine just done in a way that failed.

The 40 series doesn't use a multi-PCB design for the PCIe connector.

With the 50 series, the slot component (the part that fits into the motherboard) is not on the same PCB that houses the GPU core and VRAM. They are two different pieces with a type of FFC connecting them. Currently, it is theorized that this FFC allows for more noise introduction in the signaling. Too much to handle at times, apparently.

If this turns out to be true, this is a VERY big deal as it is part of the core design for the Founder's Edition card.

 

[edzieba]

No. The PCIe 5.0 standard and design is fine.
Current speculation is that NVIDIA's multi-PCB design introduces too much noise, messing with signal integrity.

The question is: is the 5080 within spec, or actually out of spec? If it's within spec but marginal when combined with motherboard traces designed assuming PCIe 4.0 operation, then the finger needs to be pointed elsewhere. Time for someone with the big grunty extreme bandiwdth 'scope that costs more than your house to check the eye patterns in order to find out.

The good news is performance delta between PCe 5.0 16x and PCIe 4.0 16x is basically nil, so no impact to actual users beyond inconvenience.

 

[JarredWaltonGPU]

Note that this is an issue with one particular card and not endemic to all 5080 Founders Edition cards. Also, I'm not saying this will be the only card with an issue, just that it's probably more of a QA and testing thing rather than bad design. I guess we wait and see.

My cards have been working fine (knock on wood), and there's certainly more potential for problems with three PCBs. Well, really it's just the two PCBs and the ribbon cable between them: the PCIe 5.0 slot connector, ribbon to the main PCB, and the GPU PCB. A crimped or damaged cable would obviously be one potential culprit.

And naturally, the melting 16-pin 12VHPWR connectors on the 4090 started with just one instance. LOL. Would be very interesting if, over time, there are a bunch of failures or issues with the Founders Edition cards and PCIe 5.0 that don't crop up on the custom AIB designs!

 

[A Stoner]

Hopefully it is just a one off on that specific card. Is it widespread or just this one? If it is a design flaw, hopefully it can be fixed with some hardened cables replacing the current cables in the design.

 

[Eximo]

I'm curious if there are going to be waterblocks for the FE cards. Not likely to buy one, just interested how that problem would be solved.

If I convince myself to get a 5070 Ti or something, likely slap a block on that.

Been holding out for a big Intel card, just for fun.

 

[bit_user]

No. The PCIe 5.0 standard and design is fine.

I didn't say the standard was bad. I meant including it in a consumer desktop machine was not only a pointless waste of money, but now it's causing actual problems.

Current speculation is that NVIDIA's multi-PCB design introduces too much noise, messing with signal integrity.

Yeah, which isn't an issue with at PCIe 4.0 speeds. So, the fact that Intel decided to reach for PCIe 5.0 (and AMD followed) just created a pitfall and Nvidia walked right into it.

 

[alceryes]

The question is: is the 5080 within spec, or actually out of spec? If it's within spec but marginal when combined with motherboard traces designed assuming PCIe 4.0 operation, then the finger needs to be pointed elsewhere. Time for someone with the big grunty extreme bandiwdth 'scope that costs more than your house to check the eye patterns in order to find out.

The good news is performance delta between PCe 5.0 16x and PCIe 4.0 16x is basically nil, so no impact to actual users beyond inconvenience.

True. We don't yet know enough to point the finger.

In mechanical engineering, if a design is bad enough to lend itself to being used incorrectly, on a regular basis, then the design itself is at fault. We saw this with the 12VHPWR.

Like the 12VHPWR, it could be that the safety margin for SNR in PCIe 5.0 is just too tight and is unable to account for production imperfections. However, the FFC is currently a suspect as it will DEFINITELY introduce more noise than a solid PCB design.

 

[alceryes]

I didn't say the standard was bad. I meant including it in a consumer desktop machine was not only a pointless waste of money, but now it's causing actual problems.


Yeah, which isn't an issue with at PCIe 4.0 speeds. So, the fact that Intel decided to reach for PCIe 5.0 (and AMD followed) just created a pitfall and Nvidia walked right into it.

We'll have to wait and see whether or not it's NVIDIA's actual design causing the issue. If all new single PCB PCIe 5.0 GPUs coming out in the next couple years are fine then, well, it's 100% NVIDIA's fault. 😉

 

[alceryes]

Note that this is an issue with one particular card and not endemic to all 5080 Founders Edition cards. Also, I'm not saying this will be the only card with an issue, just that it's probably more of a QA and testing thing rather than bad design. I guess we wait and see.

My cards have been working fine (knock on wood), and there's certainly more potential for problems with three PCBs. Well, really it's just the two PCBs and the ribbon cable between them: the PCIe 5.0 slot connector, ribbon to the main PCB, and the GPU PCB. A crimped or damaged cable would obviously be one potential culprit.

And naturally, the melting 16-pin 12VHPWR connectors on the 4090 started with just one instance. LOL. Would be very interesting if, over time, there are a bunch of failures or issues with the Founders Edition cards and PCIe 5.0 that don't crop up on the custom AIB designs!

Dang it, Jarred! Here I was already reaching for my tinfoil hat too. 😀

Paul over at Paul's Hardware (great tech channel, BTW) also had similar issues with his RTX 5090 review. Not sure of his test bench config though.

 

[spongiemaster]

Cramming 575W of power inside a dual-slot package for the RTX 5090 required some creative engineering solutions. For starters, the RTX 5090 FE features three PCBs rather than one large PCB, one for the PCIe 5.0 x16 connector, one for the video ports, and the main PCB hosting the GB202 package, GDDR7 memory, and power delivery circuitry. We suspect these modular boards have been connected via ribbon cables to not interfere with cooling.

That's all well and good, except he had a problem with his 5080, not 5090.

 

[uwhusky1991]

Note that this is an issue with one particular card and not endemic to all 5080 Founders Edition cards. Also, I'm not saying this will be the only card with an issue, just that it's probably more of a QA and testing thing rather than bad design. I guess we wait and see.

My cards have been working fine (knock on wood), and there's certainly more potential for problems with three PCBs. Well, really it's just the two PCBs and the ribbon cable between them: the PCIe 5.0 slot connector, ribbon to the main PCB, and the GPU PCB. A crimped or damaged cable would obviously be one potential culprit.

And naturally, the melting 16-pin 12VHPWR connectors on the 4090 started with just one instance. LOL. Would be very interesting if, over time, there are a bunch of failures or issues with the Founders Edition cards and PCIe 5.0 that don't crop up on the custom AIB designs!

I wish it was more clear if the multi-pcb design of the FE card was a common design or if none or all of the 3rd party designs also use this.

 

[spongiemaster]

I wish it was more clear if the multi-pcb design of the FE card was a common design or if none or all of the 3rd party designs also use this.

It's exclusive to Nvidia's FE cards.

 

[abufrejoval]

So, PCIe 5.0 ended up being worse than a pointless waste of money - it's downright harmful!

I wouldn't say that PCIe v5 is pointless.

Yes, using 16 lanes of PCIe v5 on an RTX 5080 for gaming may well be pointless, but that's not how I'd use it, if I got one of them.

Instead I'd do what I do already on my Minisforum BD790i, which features a Ryzen 9 7945HX with 24 PCIe v5 lanes, 16 in one x16 slot and dual x4 in M.2.

Since I use that machine currently as a µ-server, it's iGPU-only at the moment. And for that use what it lacks is fast network and SATA ports for big chunks of spinning rust.

For initial testing I put one of my Aquantia AQC107 10Gbit NICs into the x16 slot, because that's the only one it has.

With that the x16 PCIe v5 slot is blocked via a card that runs on PCIe v2 x4 or PCIe v3 x2. It's younger brother, the AQC113 could actually run 10Gbit Ethernet on a single PCIe v4 lane, so that's 15 or even 15.5 PCIe v5 bandwidth wasted. Rather sad, when you'd really want them elsewhere.

But I bought it with an eye on bifurcation and knew I'd need SATA ports, might add further storage, or even faster networking, or even eventually a dGPU, should it move to a different use case: I try not to get boxed in on my boxes!

So I split the lanes, got myself a bifurcation board which splits it into 8+4+4 lanes via two extra M.2 slots below a low profile x8 above, which fits the Mini-ITX form factor of the mainboard just fine, one on each side.

It gets me six SATA ports off 2 PCIe v3 lanes via an M.2 form factor ASMedia controller on the outward facing M.2 (where I need to run the cables), another M.2 slot for NVMe storage facing inward (enough for NVMe and cooled by the CPU fan), while the x8 slot on top is currently filled by the AQC107 NIC, wasting only 4 PCIe v5 lanes.

I could still use 4+4+4+4 bifurcation to make that better, or put someting x8 there...

Like an RTX 5080, which at PCIe v5 x8 should in theory operate just like with PCIe v4 at x16.

So PCIe v5 in that case makes tons of sense, because it openes up that flexibility.
But only if it actually were to actually work as you'd expect.

As PCIe releases are moving forward, what makes ever less sense is the fixed allocation of PCIe lanes to slots. Not only are mainboard traces at increased speeds much harder to do than on cables, it's getting ever more wasteful and inflexible.

(Unfortunately PCI v5 capable cables and connectors may not be cheap either)

Instead those GPUs should really just get M.2 connectors themselves, so they don't get connected via slots but use a variable number of M.2 cables to their mainboard. After all that so called PCIe connector already is actually using a ribbon cable to the central mainboard on the GPU.

Of course switches would be nicer yet, but nobody would want to pay for those, not at PCIe v5, because those make EPYCs look cheaper.

 

[abufrejoval]

It's exclusive to Nvidia's FE cards.

So far, no reason why others couldn't do this, or better (multi M.2 cables instead).

 

[Eximo]

So far, no reason why others couldn't do this, or better (multi M.2 cables instead).

Higher labor assembly costs. Re-tooling. Nvidia did this to make the card smaller, but people are already used to 3+ slot cards, so no reason to replace what they've got.

People doing AI work can cram an extra card in their systems, but consumer boards don't have the layout to benefit from that. So workstation boards are the only way to get, say triple 5090, going. That pushes the limits of what an off the shelf PSU can do though, so dual cards is likely what we will see most.

 

[abufrejoval]

I'm curious if there are going to be waterblocks for the FE cards. Not likely to buy one, just interested how that problem would be solved.

You'd be paying double, for the vapor chamber design and the water block. One of the main motivations previously for using a water block might have been to free up some slots the 4+ slot previous cards blocked. If you run EPYCs with tons of single width x16 slots, water might still be attractive for filling them all.

Or getting peak performance, even if they seem to do just fine here with air.

And if noise isn't the remaining motive, there doesn't seem to be an incentive left.

The problem doesn't seem related to the basic physical architecture but the 5080 PCB, the ribbon cable between the dGPU mainboard and the PCIe connector, or the PCIe connector itself: nothing a waterblock could change.

If I convince myself to get a 5070 Ti or something, likely slap a block on that.

If you like water in your computer, that's your privilege. I try to avoid it.

Been holding out for a big Intel card, just for fun.

Not sure they'll have one. But holding out can be a pleasure on its own... in measure, I'd say.

 

[abufrejoval]

Higher labor assembly costs. Re-tooling. Nvidia did this to make the card smaller, but people are already used to 3+ slot cards, so no reason to replace what they've got.

People doing AI work can cram an extra card in their systems, but consumer boards don't have the layout to benefit from that. So workstation boards are the only way to get, say triple 5090, going. That pushes the limits of what an off the shelf PSU can do though, so dual cards is likely what we will see most.

Using M.2 connectors instead of PCIe connectors won't change the assembly cost for Nvidia. Nor OEMs, nor mainboard makers. System integrators or end-users would have to compensate, but also gain the most.

For the longest time I thought that Nvidia was making these cards so super wide (and actively seemed to boycott vendors who tried to offer dual-slot blower cards), on purpose to disable multi-GPU setups and enforce market segmentation.

These days I'm less sure and more ready to accept that their main interest was in quiet-enough operation even under the heaviest of loads, while multi-GPU machine learning with PCIe or the early NVlink just doesn't work.

To say that people are used to 3+ cards is stretching it, at least for me. I've loathed these more than dual slot designs, because I bought workstation boards with 44 PCIe lanes, all of which I wanted to actually use, too. Mostly for storage (PCIe or hardware RAID), but also for 10Gbit nics.

I went with PNY dGPUs specifically to retain the ability to use those slots: their cheaper RTX 4090 made do with "only" 3 slots, while the RTX 4070 "only" used two slots.. On the latter sytem I absolutely need hardware RAID and 10Gbit Ethernet, so I was very glad to have that dual slot option. Otherwise I would have had to stick with the GTX 1060, since an ARC 770 (nice form factor!) didn't work with my KVM. (Actually, the step from a [Zotac]GTX 1080ti mini (short, low, dual slot) to a [Palit] RTX 2080ti (long, high, 2.5 slot) seemed way out of line, when I just needed to explore those tensor cores for CUDA: the difference in gaming performance was nearly trivial at the time.)

I could even run both, the 4090 and the 4070 in a Ryzen 9 "desktop" board for a while, when I was evaluating multi-dGPU performance for LLMs without the benefit of NVlink (spoiler: it's really, really bad!).

Then there is the smaller form factors people are getting attracted to, after HDDs have disappeared from desk computers. Who doesn't want to get rid of those clunky, heavy cards that used to also break just from their sheer weight? Small is truly beautiful, if it doesn't compromise on performance or cost a kidney.

I'm actually quite sure that tons of profanity is still echoing among OEM hallways, both because those official NVidia MSRP are nothing anyone can match (seems to include NVidia itself, judging by their Intel Battlemage-like MSRP availability) and the FE cooler design just has their stuff looking really, really sad as well.

If NVidia were to put formal exclusivity bricks on OEMs copying their design, I'm pretty sure AMD might see a sharp rise in popularity: there is only so much OEMs can take and NVidia can less afford enemies than a mere week ago.

 

[bit_user]

Like an RTX 5080, which at PCIe v5 x8 should in theory operate just like with PCIe v4 at x16.

Yeah, but if they upgraded the chipset connection, then you could get plenty of PCIe slots via that route and wouldn't even need to bifurcate the x16 slot.

As for people running dual GPUs, that's not really a mainstream thing to do. Games don't support it, so AI is the main reason one might do it. That's sufficiently niche that I think mainstream desktops don't need to cater for it.

Of course switches would be nicer yet, but nobody would want to pay for those, not at PCIe v5, because those make EPYCs look cheaper.

They're already integrated into motherboard chipsets.

 

[edzieba]

In mechanical engineering, if a design is bad enough to lend itself to being used incorrectly, on a regular basis, then the design itself is at fault. We saw this with the 12VHPWR.

Counterpoint: nothing is foolproof to a sufficiently talented fool.

 

[TJ Hooker]

This article doesn't really make sense to me. It speculates that there could be a systemic issue related to PCIe 5.0 and multi PCB cards. But also says the RTX 5090, which uses the same multi PCB design and PCIe 5.0, doesn't have an issue. Maybe the debauer video explains things better, but I kinda hate watching techtuber videos.

The linked igorslab article is kinda confusing too. It makes some vague, passing mention of RTX 5090 boot issues, but doesn't seem to actually describe what those issues are. Then Igor says he fixed the issue with some soldering. What issues was he having, and what did he solder? Admittedly I haven't read the entire review, maybe he put those details in some otherwise unrelated page of the article.