If we’re after the entire ROM image, the most reliable way would be to desolder ROM chip off the board. Don’t know if there are other ways like debug header or something similar on Flex 170.
I hope @Forge sees your post and comes up with a better solution though, maybe there is no need to dump entire ROM, only a region of it.
I am not sure that reflashing whatever’s in here on an A770 would get you to where you want to be though, unless everything about the card is in NVRAM. I’m thinking of things like the PCI vendor/product code, which is probably fused in the ASIC.
But then again, I have no idea, which is why I was asking what you guys needed.
I think vendor and product code would not suffice. DeviceID seems to be 56C0 for Flex170 and 56A0 for A770, but I don’t think that’s the hard part.
I tried quick and dirty trick by text-searching for DEV_ID in SR-IOV-enabled drivers from Intel repository, repaced 56C0 for 56A0 in i915_pciids.h (and some other places I don’t remember) and tried to install that. It installs but device variables that control VFs did not appear, therefore commands to enable and manage VFs obviously did nothing. I did not have more time to investigate and stopped there.
So either I was doing something wrong, or there’s more to it and some part in firmware that accepts commands for enabling VFs in Flex FW is missing in A770 FW. That’s why I believe FW dump would be a good starting point, but not necessarily the only one.
This chip on your first photo does appear to be the flash chip with firmware, and trying reading it would be nice indeed. At the same time I don’t really trust the clips referenced in Mandaril’s post, had mixed results with those – e.g. borrowed a clip once, tried to dump BIOS from a motheboard and received 3 bit-different results in 3 reads. Maybe I had bad luck with mine.
Ill believe it when i see it. The key question is whether it relies on hardware only in B series or if they will open the door to consumer alchemist wih a new driver.
So pretty much a thing for 2026+ for server stuff.
Releasing SR-IOV more or less with the B50/B60 probably isn’t coincidental. I’m certainly following B50 release because I want one or two in 2026. It’s all coming together in late 2025 and 2026 \o/
MSRP in the US is mostly to get people excited by the low price. And I think Intel just doesn’t make money on them, so supply is limited, increasing margins in the process.
Competition is just great…3 is more than 2. Let’s keep the 3rd one and encourage a 4th one.
Eventually Intel will just do AMD+Nvidia things (remember quad-core decade-long stagnation?).
But we can all profit for the time being from Intel being desperate to get market share and get established
In the EU B580s are well available at €269. (close to a $250 MSRP if you the euro is about 1.17 dollars but the EU has around 20% sales tax). You have more of a taxation and trade issue going on, I suspect.
As far as I’ve experimented with SR-IOV usually you put the number of virtual functions you want in a kernel parameter and then you get the PCIe devices to pass through to a VM, so I’m not sure what kind of userland work would be required?