AMD Execution Thread [2025]

[Albuquerque]

I find it difficult to develop something so low level remotely. What if the kernel crashes all hardware?

This is where IPMI comes in. It's baseboard level, even below the firmware. If a Linux kernel crashes so hard as to hard-lock the box, you can still get in and power-cycle it. This is a very common technology on datacenter floors for decades because, so long as the box has literal power and the IPMI device has a network connection, you can do anything to the box as if you were physically sitting in front of it.

 

[arandomguy]

You misunderstood. I am saying George Hotz does little more than whine on twitter. Hardware reviewers review hardware, that is their 'product' and it ends up being marketing for the manufacturer anyways.

I'm being facetious with that "whine on twitter" framing but really at the core it's the same thing in that they're basically broadcasting to the public as a marketing exchange proposition. None of those groups inherently deserves considerations but ultimately it's the same cost/benefit (future benefit) analysis for the company to make.

I'll admit I'm really just poking at the idea that hardware reviewers derserve hardware samples and that they are a public service when in reality they're a business like any other and anything they do should just be viewed from a business angle.

 

[Florin]

I find it difficult to develop something so low level remotely. What if the kernel crashes all hardware?

I'm not sure how that would ever be a problem. You can reset or power cycle any kind of instance using the cloud vendor GUI or API.
If the actual hardware underlying the VM had somehow crashed they'd just spawn your instance on another server from their pool.
And if you somehow managed to get the kernel on your instance in a state that is borked right upon boot you could always revert to a previous snapshot of the storage beforehand, or even reinstall the instance's OS altogether.

IPMI and BMC access take this to another level, still. Even if the server's actual hardware crashes IPMI will still allow you to login from remote and power cycle the whole unit, or use screen/keyboard/mouse to fix anything, right up to going into the BIOS/firmware to change any settings.

IPMI/BMC works like a remote KVM, or a sort of separate server within your server if you will. There literally isn't anything your software can do to your server that would prevent IPMI/BMC from still being able to provide you control.

 

[IQandHDR]

For people never doing OOB(Out of Band) mgmt before:
HPE ILO:

Dell iDrac:

Cisco CIMC:

Every single server I have worked on have had this.
You can sit in another country and do work like if you were located at the server with keyboard/mouse og monitor attached.

Everything from installing a OS to doing a power cycle.
It is at the foundation for running a datacenter/serverpark.

It also house the server-logs, so you can pull supportlogs for the vendor in case of RMA of parts.

 

[Sega_Model_4]

You do realize there's a difference between couple consumer or even pro cards and two full datacenter servers + $1M, right?
Had Holz requested the former from AMD he no doubt would have gotten what he wanted and maybe more.

At that time, Nvidia's datacenter GPUs were inferior to consumer GPUs, which evolved faster. Nvidia Teslas ran at lower clock rates and had ECC memory. They were more for parallelism work and not something super fast.
They changed the focus around there from Volta architecture.

About Geohot, if it can fix consumer GPUs, the next step would be datacenter GPUs.

The point is that AMD could have just lent, made some kind of contract to avoid any problems. The way they handled this is bizarre. Even though the colleagues above mentioned that it is possible to develop in the cloud, for direct development on the metal, I still think that having the physical product works better.

 

[TopSpoiler]

Somebody didn’t get the memo: AMD ad claims you can enjoy gaming on Radeon RX 9070 XT already - VideoCardz

They forgot to change the advertising contract, which proves that the cancellation of the January release was a last-minute decision..

 

[Scott_Arm]

For people never doing OOB(Out of Band) mgmt before:
HPE ILO:
..
Dell iDrac:
...
Cisco CIMC:
...

Every single server I have worked on have had this.
You can sit in another country and do work like if you were located at the server with keyboard/mouse og monitor attached.

Everything from installing a OS to doing a power cycle.
It is at the foundation for running a datacenter/serverpark.

It also house the server-logs, so you can pull supportlogs for the vendor in case of RMA of parts.

Yep. Works great. I've been able to do my job fully remotely. Fully virtualized environment built on top of servers with OOB mgmt.

 

[digitalwanderer]

Somebody didn’t get the memo: AMD ad claims you can enjoy gaming on Radeon RX 9070 XT already - VideoCardz

They forgot to change the advertising contract, which proves that the cancellation of the January release was a last-minute decision..

The ineptitude in which they're handling this launch is insanely terrible, it's embarrassing to watch.

 

[IQandHDR]

Yep. Works great. I've been able to do my job fully remotely. Fully virtualized environment built on top of servers with OOB mgmt.

Same, same...

 

[IQandHDR]

The ineptitude in which they're handling this launch is insanely terrible, it's embarrassing to watch.

You sound surprised?

 

[Seanspeed]

Not an official source and cost translations should be taken with some salt as well, but this paints a pretty ugly picture of what AMD's intentions were with pricing:

https://videocardz.com/newz/retailer-confirms-powercolor-radeon-rx-9070-xt-red-devil-limited-edition-is-in-stock-talks-amd-pricing-strategy

Sounds like nobody should be getting their hopes up that AMD will offer any kind of actually appealing pricing for RDNA4. AMD keeps wanting to prevent themselves from being seen as a budget brand, but seems to have no clue whatsoever that they cannot compete with Nvidia with worse technology and only a small little discount for it. smh

 

[Kaotik]

Not an official source and cost translations should be taken with some salt as well, but this paints a pretty ugly picture of what AMD's intentions were with pricing:

https://videocardz.com/newz/retailer-confirms-powercolor-radeon-rx-9070-xt-red-devil-limited-edition-is-in-stock-talks-amd-pricing-strategy

Sounds like nobody should be getting their hopes up that AMD will offer any kind of actually appealing pricing for RDNA4. AMD keeps wanting to prevent themselves from being seen as a budget brand, but seems to have no clue whatsoever that they cannot compete with Nvidia with worse technology and only a small little discount for it. smh

Quite surely not talking about AMD MSRPs if anything real, just like the supposed RTX 50 prices from australia weren't accurate either.
If company is saying to investors it's aiming for mainstream to gain marketshare those prices simply couldn't be an option.

 

[digitalwanderer]

Not an official source and cost translations should be taken with some salt as well, but this paints a pretty ugly picture of what AMD's intentions were with pricing:

https://videocardz.com/newz/retailer-confirms-powercolor-radeon-rx-9070-xt-red-devil-limited-edition-is-in-stock-talks-amd-pricing-strategy

Sounds like nobody should be getting their hopes up that AMD will offer any kind of actually appealing pricing for RDNA4. AMD keeps wanting to prevent themselves from being seen as a budget brand, but seems to have no clue whatsoever that they cannot compete with Nvidia with worse technology and only a small little discount for it. smh

So they were going to charge more for a less powerful card, brilliant!

/s

 

[Seanspeed]

Quite surely not talking about AMD MSRPs if anything real, just like the supposed RTX 50 prices from australia weren't accurate either.
If company is saying to investors it's aiming for mainstream to gain marketshare those prices simply couldn't be an option.

I doubt they were gonna be quite as high as the translated costs suggest, but if the general idea of higher prices at all was true, it's still a really bad sign for anybody hoping AMD will provide some good value and competitive price pressures for the GPU market as a whole.

 

[IQandHDR]

Again, transistor costs are rising, but a lot of people stil live in denial

 

[Seanspeed]

Again, transistor costs are rising, but a lot of people stil live in denial

These GPU chips do not cost nearly as much as you think they do.

 

[Albuquerque]

And again, we revisit the age-old teachings of pricing of consumer goods not being linked to material costs, at the very least not in a capitalist society. There's functionally nothing that you purchase in the US (and most other countries) that isn't price-adjusted to what the market will bear.

Some countries realize how dangerous this can be, and leverage government controls to put limits on said pricing for "human needs" items like food and water, medicine and healthcare, etc. Dunno about you, but I don't live in one of those countries. Even then, discrete GPUs are a "luxury item" and will always and forever be priced accordingly.

Prices aren't going down, and to quote @trinibwoy : can we move on now?

 

[Kaotik]

Already posted in other thread, but Azor confirmed $899 claimed has never been in plans

 

[Albuquerque]

Yup, saw that note too. However, his comment also doesn't say whether the target price was higher or lower... /s

I (speaking for myself, hence the "I" in this phrase) think it's safe to assume the price was always planned to be lower. Just how much lower is an interesting question, and again one we will likely never know unless someone in the AMD marketing or presales departments decides to let loose with some pricing sheets...

 

[Seanspeed]

And again, we revisit the age-old teachings of pricing of consumer goods not being linked to material costs, at the very least not in a capitalist society. There's functionally nothing that you purchase in the US (and most other countries) that isn't price-adjusted to what the market will bear.

Some countries realize how dangerous this can be, and leverage government controls to put limits on said pricing for "human needs" items like food and water, medicine and healthcare, etc. Dunno about you, but I don't live in one of those countries. Even then, discrete GPUs are a "luxury item" and will always and forever be priced accordingly.

Prices aren't going down, and to quote @trinibwoy : can we move on now?

Prices aren't going down, but nobody is asking them to take a loss on the hardware. And unless their margins are super slim(they aren't), then there is absolutely scope for value for consumers to improve from where it is today, at least on-paper.

Point is, transistor costs rising is not an excuse for GPU's to keep costing more. You yourself make the supposed inarguable claim that manufacturing costs have NOTHING to do with end pricing whatsoever, after all.