Intel Introduces the Max Series Product Family: Ponte Vecchio and Sapphire Rapids

[Wirko]

Please stop acting as if servers, compute clusters and ultra-high-end workstations are built in ways that resemble consumer PCs whatsoever.

... also those are built by people that tend to read manuals *before* anything goes the way of the Etna Krakatoa.

 

[Patriot]

It seems like Intel went with the new connector.

Intel adopts 12VHPWR connector for its Data Center GPU Max 1100 PCIe series - VideoCardz.com

12VHPWR is no longer an exclusive to NVIDIA The company behind the ATX 3.0 specs adopts the new power connector. Data Center GPU Max 1100 GPU with 12VHPWR connector, Source: Intel The so-called PCIe Gen5 or 16-pin or 12VHPWR power connector is not technically a requirement for ATX 3.0 power...

videocardz.com

lol, I tried to find it in their specs or a picture of the end of the card. Odd they would limit to 300w then.
Thanks, good to know.

No they won't. These accelerators will go into built-for-purpose servers (and a very few workstations), all of which will have built-for-purpose PSUs with native 12VHPWR cabling. If you're buying a $5-10 000 accelerator card, you're also buying a PSU that natively supports what that system needs.

Please stop acting as if servers, compute clusters and ultra-high-end workstations are built in ways that resemble consumer PCs whatsoever.

I mean, they will be adapted to go in current gen servers, but not the 4-1 adapters now.... it will be a 1:1 cable as its a 300w capped card. Probably why the first gen is only 300w to make it compatible with current systems. Which typically have a power plug or two on the riser for the card(s), and a short cable to the card. Or a breakout board on the PSU hotswap backplane.

 

[Valantar]

I mean, they will be adapted to go in current gen servers, but not the 4-1 adapters now.... it will be a 1:1 cable as its a 300w capped card. Probably why the first gen is only 300w to make it compatible with current systems. Which typically have a power plug or two on the riser for the card(s), and a short cable to the card. Or a breakout board on the PSU hotswap backplane.

Yeah, probably either 1:1 adapters or new risers/breakout boards with native outputs, depending on what is available and supported by the server vendor (and gets approved by acquisitions). Either way, nothing resembling the octopus-like adapters we've seen for the 4090. And as you say, with 300W required they don't even need any of the new sense pins, so I'm sure native 1:1 cables will be available for pretty much any server where this might be used.

 

[Patriot]

Yeah, probably either 1:1 adapters or new risers/breakout boards with native outputs, depending on what is available and supported by the server vendor (and gets approved by acquisitions). Either way, nothing resembling the octopus-like adapters we've seen for the 4090. And as you say, with 300W required they don't even need any of the new sense pins, so I'm sure native 1:1 cables will be available for pretty much any server where this might be used.

Probably just new cables, keep old risers. Fun fact, Single slot 150w V100s appear to have used 8pin PEG power. huh. DL380 with 5 baby V100s.


Noticed the Nvidia (not a tesla tesla) L40 is using the new 12v HPWR whereas the H100 is still using the 12v EPS. (scratch that, A100 is still on EPS, and some people reused it's picture and dubbed it H100)

 

[Dirt Chip]

No they won't. These accelerators will go into built-for-purpose servers (and a very few workstations), all of which will have built-for-purpose PSUs with native 12VHPWR cabling. If you're buying a $5-10 000 accelerator card, you're also buying a PSU that natively supports what that system needs.

Please stop acting as if servers, compute clusters and ultra-high-end workstations are built in ways that resemble consumer PCs whatsoever.

They don't generally, but I won't be surprised to see those MAX GPU's with adapters from time to time. No point refute this very much possible combinations in not well though builds.

 

[Valantar]

They don't generally, but I won't be surprised to see those MAX GPU's with adapters from time to time. No point refute this very much possible combinations in not well though builds.

Again, these are HPC accelerators, most likely in the $5-10000 price range. How many "not well thought out builds" do you think these will be going into? Like, globally? One? Two? Five? These won't be sold through retail channels, likely not even most OEM/SI channels. To even get your hands on these you need the right connections and access, and tons and tons of money - and even then supplies will likely be sufficiently constrained that high volume customers get most if not all stock for a long, long time.


And if some doofus finds one of these on Ebay in five years and uses it with an adapter - so what? PEBCAK.

It's also entirely likely that some server tech will be running these in weird ways for various testing purposes - but that would then either be in a relevant server chassis or on an open bench (unlikely due to lack of airflow, would require some sort of custom shroud). If in a chassis, it won't be using anything resembling a consumer adapter. If on an open bench for testing, who cares?

You're positing a problem so astronomically small and irrelevant that it really isn't worth discussing.

 

[Dirt Chip]

How many "not well thought out builds" do you think these will be going into?
Very Very few, I hope.

You're positing a problem so astronomically small and irrelevant that it really isn't worth discussing.

Agree mostly, but I haven't refer to it as a problem- more like a funny situation that might happen with a 'bonus'* to it.
I didn't try to argue that MAX server and data-centers will cach fire from now on because of the connector, if that`s what so bothering you mate. Also, I don`t think that there is a fundamental problem by using adaptors (you just want to avoid them if possible).
What interesting imo is to see Intel version of adaptors of 12VHPWR connector (if any will be made) and compare them to NV adaptors. All of that come from the assumption that Intel will use 12VHPWR based on the new MAX family and the indicated 300, 450, 600W power jumps that correspond (to my understanding) to 12VHPWR utilizing 4,5,6 "hot" wires. Nothing more, Nothing less.
*IF Intel made any, comparing both on consumers system (RTX 4090) will be very interesting and might further help understand the (very rare) melting incident you saw recently.

I also agree to end the discussion on the topic