Raja Koduri Teases "Petaflops in Your Palm" Intel Xe-HPC Ponte Vecchio GPU

[Deleted member 205776]

This isn't a graphics GPU fellas, it's a purpose-built chip for datacenter workloads.

It's an impressive chip sporting 100 billion transistors. For reference an RTX 3090 is approximately 28 billion transistors. It's also valuable R&D in terms of modular design.

Don't ever underestimate the engineering powerhouse that Intel is. Intel may have fumbled a node and mis-judged the benefits of chiplet design, but they are still a force to be reckoned with.

Waiting to see this cooled on air.

 

[InVasMani]

Just how many magic tiles is it going to take for Intel to fix Valheim's 16 x 16 pixel texture detail? If you like Valheim's texture detail now just imagine playing it on a 85" display at 480p for extra stylistic quality.

 

[Vayra86]

to be honest, not sure if this is fault of Raja, seems like Intel Marketeering department is shit flinging again.
Like, BRUH, look at these magic gluing of words - "Alchemy of Technologies" "47 Magical Tiles" (with Exact Capitalization). Raja himself would probably threw up a little when he saw this and his photos presented together.

Strangely the silly marketing circus, including the 'what the F kind of nonsense is thát then?!' - moments are very keen to follow Raja wherever he goes.

There is a connection here. Poor Volta. Nuff Said

Hmm... the marketing material also includes "petaflops in your palm", which refers to mobile devices, which are used for gaming mostly by children. I guess that's it.

Don't you guys have phones?!

This isn't a graphics GPU fellas, it's a purpose-built chip for datacenter workloads.

It's an impressive chip sporting 100 billion transistors. For reference an RTX 3090 is approximately 28 billion transistors. It's also valuable R&D in terms of modular design.

Don't ever underestimate the engineering powerhouse that Intel is. Intel may have fumbled a node and mis-judged the benefits of chiplet design, but they are still a force to be reckoned with.

Its only impressive if it ever leaves the proof of concept stage.

Intel has built more impressive product... only to not do things with it because economically it was never feasible to begin with.

 

[Deleted member 205776]

Don't you guys have phones?!

don't remind me

 

[AusWolf]

Don't you guys have phones?!

There's a difference between owning a phone for phoney things, and attempting to game on a 5-7" touchscreen with no keyboard and mouse because daddy won't buy a gaming PC. No offense.

 

[Vayra86]

There's a difference between owning a phone for phoney things, and attempting to game on a 5-7" touchscreen with no keyboard and mouse because daddy won't buy a gaming PC. No offense.

None taken. But don't you have a phone?

Its coming whether you like it or not... I really do hope enough people keep seeing the value of real gaming. I will say the imaginary value of that retro gaming PC I'll make is increasing by the day. Right now I've got a win 10 rig but if that ever starts acting funny for 32 bit I'm going to get that done. Everything in local storage, pre installed and fully patched and modded, and the system remains offline. There's just so much content that'll never be released in a better iteration... no way I'm going to dive into remasters every five years that never really quite get there.

 

[AusWolf]

None taken. But don't you have a phone?

I fall under the "owning a phone for phoney things" category. My Galaxy A20e easily runs 2-2.5 days on one charge because I barely even touch it.

Its coming whether you like it or not... I really do hope enough people keep seeing the value of real gaming. I will say the imaginary value of that retro gaming PC I'll make is increasing by the day. Right now I've got a win 10 rig but if that ever starts acting funny for 32 bit I'm going to get that done. Everything in local storage, pre installed and fully patched and modded, and the system remains offline. There's just so much content that'll never be released in a better iteration... no way I'm going to dive into remasters every five years that never really quite get there.

That's awesome! Looking at retro hardware prices, it won't be cheap, but I wish you good luck!

 

[Valantar]

I really do hope enough people keep seeing the value of real gaming.

It's not going to change. Mobile gaming will grow, but it's going to complement, not replace other forms of gaming. It has the advantage of accessibility (as in: everyone has a phone) and portability, but has severe disadvantages in terms of the audiovisual-haptic experience. If that's all you've got it's better than nothing (unless you find it bad enough to be unusable, of course), but it's not going to replace experiences that materially improve tangible sensorial aspects crucial to the gameplay experience. There's a reason controllers don't consist of touch surfaces after all - it would be easy to make, and much cheaper than current controllers, but also vastly inferior in almost any way.

More on topic, this is no doubt a significant feat of engineering in the advanced packaging used, but I don't quite believe it won't be hot and extremely difficult to cool - I can't imagine flatness across that mass of dice is going to be good, so cooler contact will be iffy.

 

[TheoneandonlyMrK]

How about this dick stops waving the same prototype about once every three months, only to visually demonstrate how totally not ready it is via proto pics.
FFS it's a server chip that's been banded about a year now and it looks like the Dev board is still more breadboard then evaluation.
And just imagine how far out a consumer useful variant is.

2024 ( a year after tsmc makes them, FFS intel sigh)

Yep you can stick it top shelf on t right Raja.

 

[Caring1]

This isn't a graphics GPU fellas, it's a purpose-built chip for datacenter workloads.

It's an impressive chip sporting 100 billion transistors. For reference an RTX 3090 is approximately 28 billion transistors. It's also valuable R&D in terms of modular design.

Don't ever underestimate the engineering powerhouse that Intel is. Intel may have fumbled a node and mis-judged the benefits of chiplet design, but they are still a force to be reckoned with.

I can't read this without hearing Trump's voice.
I didn't know he was in their marketing department.

 

[windwhirl]

I can't read this without hearing Trump's voice.
I didn't know he was in their marketing department.

Oh, look what you did! Now I also hear his voice in my head!

 

[KainXS]

Raja can't be trusted, Raja is all talk, can anyone even verify if that was really the case at AMD on the development side. Unless you worked there, then you cannot verify that, i don't get it. Performance numbers would be nice but this is an enterprise level datacenter chip it seems, and a specialized one at that. The only gripe I have with this article is that they did not state how they achieved that petaflop.

 

[L'Eliminateur]

Raja has a history of being all words and no actions.

Dude, i was going to post exactly this!.

This guy is "chief smoke blower" wherever he lands, in AMD he was all teasing and all bullshit and zero execution, the things he released under AMD ended up being underperforming crap

and know he teases a product that has honestly ZERO relevance for consumers, and that no one gives a shit about practically.

 

[chstamos]

I think it is abundantly clear from this thread, as well as others on the internet, that people are growing very tired indeed of these "teaser" shenanigans. Can Raja keep a lid on it until he has a product to show, or is he putting in candidacy for hardware's peter molyneux?

 

[AusWolf]

Dude, i was going to post exactly this!.

This guy is "chief smoke blower" wherever he lands, in AMD he was all teasing and all bullshit and zero execution, the things he released under AMD ended up being underperforming crap

and know he teases a product that has honestly ZERO relevance for consumers, and that no one gives a shit about practically.

Maybe that's why Intel (and before that, AMD) advertise all these non-existent, non-functional pipe dreams with his name. I mean, there isn't a lot of people working for these tech companies that the world knows by name, but whenever we hear about a fancy new chip that's totally not going to work, we always know it's this guy's work. Even the article says "Raja Koduri teases..." and not "Intel teases...".

 

[Manoa]

more flops from intel, no one saw that coming

 

[Steevo]

All the marketing crap aside, I'm happy they succeeded in the multi chip (tile) approach and got another step closer to a working chiplet design with I/O hub and all. Large memory tiles closer to the compute side makes me really excited. Maybe we just got closer to the stacked chiplet designs.

A bit sad to see it all shushed and laughed at. This is some impressive stuff they got there, if it performs at this stage or not I don't care. Might take a few iterations. It's new stuff, exploring new ways to move forward. Exactly what Intel desperately needs.
I just agree that maybe posting these marketing slides and throwing a big number was a tad to early when the product appears to be still in R&D with only some prototypes in the lab.

It's all smoke and.mirrors until there are independent third party reviews and product for sale.

So far Xe is Exhausted and overhyped for anything other than a catch up to last gen AMD and Nvidia GPU tech that doesn't have half the bugs, and works on all platforms.

 

[Minus Infinity]

Well Intel has had so many flops over the last several years, so petaflops is no surprise.

 

[watzupken]

Looking at Raja's track record during his time at AMD, I think I rather wait and see what he said translates to actual performance. The more he beats around the bush, the more skeptical I am. The current Xe graphics on Tiger Lake looks decent, but the question remains as to how well it will scale up. For example, Vega works very well as an iGPU, but runs really hot and typically slower than competition at the high end. So far the dedicated Xe graphic solution seems pretty lacklustre to me.

 

[R0H1T]

I dunno, were they trying to open a portal into another multiverse

 

[Caring1]

I dunno, were they trying to open a portal into another multiverse

I don't think they are that smart, you can tell by the naming scheme on that machine, Dis 1, no doubt the other machine is called Dat 1.

 

[Valantar]

I can't read this without hearing Trump's voice.
I didn't know he was in their marketing department.

The sentences are way too coherent and there isn't enough repetition for that. Also no mention of CHY-NA.

I don't think they are that smart, you can tell by the naming scheme on that machine, Dis 1, no doubt the other machine is called Dat 1.

If there's a third one, is it Nother 1?

 

[Vayra86]

I can't read this without hearing Trump's voice.
I didn't know he was in their marketing department.

But these are coherent lines of text! How can you hear Trump saying this?

 

[Valantar]

But these are coherent lines of text! How can you hear Trump saying this?

I made an attempt at a Trumpslation:

This isn't a graphics GPU, guys. It's for companies. The most high-tech companies. Beautiful companies, we have the best companies. USA! We're bringing all the companies back here. All the jobs. Foxconn is opening a factory in Wisconsin - we paid them for it, so they're coming! USA!

They don't have this stuff in Chy-na. They don't even have GPUs!

[10-minute digression that passes by Hillary Clinton's emails, Benghazi, Covid being a hoax and/or Chinese, the election being rigged and Trump winning by the biggest landslide in the history of the universe, gun rights, Trump being appointed by God, and fast food]

This has the most transistors. Fantastic transistors, so pretty. So many transistors, you can't even count them! And they're the best. The best transistors from the best engineers. Covfefe.

 

[Houd.ini]

Well Intel has had so many flops over the last several years, so petaflops is no surprise.

If it really flops, they'll simply open a petting zoo called the intel Pet-a-flop - Petaflop in your palm®