Yes, I had to do a double-take. Rialto was AMD's (ATI's) PCIe-to-AGP bridge chip that they released back in 2005, first used with the Radeon X800/X600/X300 series of cards. This was before ATI was acquired by AMD.
Still don’t understand how it’s legal for Intel to buy wafers from a competing foundry.
‘That it makes sense for Intel to do’ is obvious - but it benefits literally no one but Intel (and to a lesser extent TSMC) - and hurts everyone but Intel, especially the consumer.
Are we just saying antitrust laws don’t apply anymore?
There’s an easy solution: spin off the fabs. Solved.
ACHKTUALLY. By going to TSCM. Intel outright confirmed to everyone and their mother that their process nodes are garbage and cannot compete to the Samsung or TSCM. They fell from their throne.
Intel has been a long standing client of TSMC and for decades has been purchasing as much or more silicon from them than AMD does. This is not a recent thing in the slightest.
No, this is absolutely a recent thing. Just cause Intel got simple logic chips from tsmc before doesn't mean it's not game changing that they will now produce GPUs and CPUs at s different fab. But you already knew this.
I take it you don't recall that Intel made those purchases to decrease capacity for it's competitors so as to kill off the other CPU manufacturers at the time?
Did they? Have a link? Interesting claim - also because it might have backfired, giving TSMC the scale and capacity to grow faster and catch up/overtake intel…
The only advantage? Toxic intel fanboys are certainly interesting and delusional. AMD is technologically years ahead, accept the reality and stop coping. :) If you want highest power and efficiency you buy AMD.
AMD had the edge in efficiency even when they were on GloFo 14/12, which is certainly not a more advanced node in any way than what Intel has been on at the same time.
TSMC plans capacity based on customer demand. with the record $100bn capex over 3 years, it's likely they've built additional capacity for Intel without jeopardizing AMD's access to wafer
lemurbutton, the same can be said to you. post your sources, show us your math. show us that your high praise of apple is justified and mot just blind fanboy posts.
800W per socket - either you need a category 5 strength hurricane blowing through your cooler or you watercool this baby.
We just saved 60W / server by adjusting fans from "full" to "optimized" setting. I don't want to think how much power you need to feed into the fans on a 1U or 2U box that runs couple of these things...
"this implies that Intel has opted to integrate the Rambo cache on-die with the compute tiles"
The already have 408MB of L2, with over half embedded in the base tile. It would make sense to me if they also embedded the L3 Rambo cache in the base tile.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
37 Comments
Back to Article
SydneyBlue120d - Tuesday, May 31, 2022 - link
If the words "Rialto Bridge" seems familiar to you, well, you're officially old.Jawadali - Wednesday, June 1, 2022 - link
Yes, I had to do a double-take. Rialto was AMD's (ATI's) PCIe-to-AGP bridge chip that they released back in 2005, first used with the Radeon X800/X600/X300 series of cards. This was before ATI was acquired by AMD.Stuka87 - Thursday, June 2, 2022 - link
Yeah, I thought the same thing. I had an ATI X800 which used a Rialto Bridge chip.Sahrin - Tuesday, May 31, 2022 - link
Still don’t understand how it’s legal for Intel to buy wafers from a competing foundry.‘That it makes sense for Intel to do’ is obvious - but it benefits literally no one but Intel (and to a lesser extent TSMC) - and hurts everyone but Intel, especially the consumer.
Are we just saying antitrust laws don’t apply anymore?
There’s an easy solution: spin off the fabs. Solved.
lemurbutton - Wednesday, June 1, 2022 - link
Cause money.Don't get hurt that Intel is about to nullify the only advantage AMD has which is a more advanced node from TSMC.
tamalero - Wednesday, June 1, 2022 - link
ACHKTUALLY.By going to TSCM.
Intel outright confirmed to everyone and their mother that their process nodes are garbage and cannot compete to the Samsung or TSCM.
They fell from their throne.
Lakados - Wednesday, June 1, 2022 - link
Intel has been a long standing client of TSMC and for decades has been purchasing as much or more silicon from them than AMD does. This is not a recent thing in the slightest.monkeboi - Wednesday, June 1, 2022 - link
No, this is absolutely a recent thing. Just cause Intel got simple logic chips from tsmc before doesn't mean it's not game changing that they will now produce GPUs and CPUs at s different fab.But you already knew this.
ballsystemlord - Thursday, June 2, 2022 - link
I take it you don't recall that Intel made those purchases to decrease capacity for it's competitors so as to kill off the other CPU manufacturers at the time?jospoortvliet - Thursday, June 9, 2022 - link
Did they? Have a link? Interesting claim - also because it might have backfired, giving TSMC the scale and capacity to grow faster and catch up/overtake intel…Khanan - Wednesday, June 1, 2022 - link
The only advantage? Toxic intel fanboys are certainly interesting and delusional. AMD is technologically years ahead, accept the reality and stop coping. :) If you want highest power and efficiency you buy AMD.Dolda2000 - Friday, June 3, 2022 - link
AMD had the edge in efficiency even when they were on GloFo 14/12, which is certainly not a more advanced node in any way than what Intel has been on at the same time.Unashamed_unoriginal_username_x86 - Wednesday, June 1, 2022 - link
I guess AMD is screwed since they can't license Arm cores (they compete against Arm in CPU IP)If dogfooding was compulsory we wouldn't call it dogfooding
Y250 - Wednesday, June 1, 2022 - link
AMD already uses ARM designs in their Zen architecture as secure processors. So they already own a ARM licence.lmcd - Wednesday, June 1, 2022 - link
This is /s and is meant to demonstrate how off-base the original poster is.lmcd - Wednesday, June 1, 2022 - link
AMD owned GloFo still while purchasing GPU wafers for ATi.Please, don't talk about things you don't understand.
TheinsanegamerN - Friday, June 3, 2022 - link
TSMC's factories use hardware fromt he US, that use chips often made by *gasp* INTEL,or even SAMSUNG!THAT SHOULD BE ILLEGAL!!!11!!!!
It's not illegal for a company to provide a product. That's just all sorts of silly.
ammaterasu - Friday, June 3, 2022 - link
TSMC plans capacity based on customer demand. with the record $100bn capex over 3 years, it's likely they've built additional capacity for Intel without jeopardizing AMD's access to wafermonkeboi - Tuesday, May 31, 2022 - link
This is actually the first product on TSMC 3nm, but Intel would rather keep that on the down low.lemurbutton - Wednesday, June 1, 2022 - link
No it isn't. This is sampling in in mid 2023 and probably won't be released until late 2023. Apple is rumored to be making the M2 on 3nm.monkeboi - Wednesday, June 1, 2022 - link
Wrong, Intel won the main contract on 3nm at TSMC, do they'll be first out.lemurbutton - Wednesday, June 1, 2022 - link
Wrong again. Apple always eats first.jospoortvliet - Thursday, June 9, 2022 - link
Yet the M2 is (sadly) on N5.Khanan - Tuesday, May 31, 2022 - link
What a ridiculous product, the manufacturing cost will be gigantic and not worth the price.lemurbutton - Wednesday, June 1, 2022 - link
It's not used to run your puny games. It's used exclusively in the enterprise or government contracts.Khanan - Wednesday, June 1, 2022 - link
Maybe your puny brain didn’t comprehend it, so I’m spelling it out for you: I’m not talking about my PC, I’m talking about server usage.lemurbutton - Wednesday, June 1, 2022 - link
Ok then what’s your source? Tell us more. Show us your math and benchmarks that led to your conclusion.Khanan - Thursday, June 2, 2022 - link
I don’t have to tell a toxic troll anything, stay dumb.Qasar - Thursday, June 2, 2022 - link
lemurbutton, the same can be said to you. post your sources, show us your math. show us that your high praise of apple is justified and mot just blind fanboy posts.zepi - Wednesday, June 1, 2022 - link
800W per socket - either you need a category 5 strength hurricane blowing through your cooler or you watercool this baby.We just saved 60W / server by adjusting fans from "full" to "optimized" setting. I don't want to think how much power you need to feed into the fans on a 1U or 2U box that runs couple of these things...
Khanan - Wednesday, June 1, 2022 - link
Whole power plants will be needed to power super computers with this product.Ryan Smith - Wednesday, June 1, 2022 - link
"800W per socket - either you need a category 5 strength hurricane blowing through your cooler or you watercool this baby."The 800W configuration is meant to be used with liquid cooling.
Lakados - Wednesday, June 1, 2022 - link
In the server space anything 600w or greater is almost always cooled via liquid, it’s not feasible on air.flgt - Thursday, June 2, 2022 - link
STH just ran some interesting articles on the 85F liquid cooling solutions used in theses systems.Sam lebon - Thursday, June 2, 2022 - link
A cat-5 hurricane could work. Another option is to hire the Iceman.Hosip - Sunday, June 5, 2022 - link
GreatJayNor - Friday, June 24, 2022 - link
"this implies that Intel has opted to integrate the Rambo cache on-die with the compute tiles"The already have 408MB of L2, with over half embedded in the base tile.
It would make sense to me if they also embedded the L3 Rambo cache in the base tile.