• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Discussion Intel current and future Lakes & Rapids thread

Page 937 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Notably, Intel CEO Lip-Bu Tan is currently the chairman of SambaNova Systems. Intel Capital, which Intel is in the process of spinning off into a standalone fund, has also invested in SambaNova Systems. Another investor in SambaNova, Japan’s SoftBank Group, made a major investment in Intel earlier this year.
Click to expand...
Hmmm....
 
There's the question of conflict of interest. Putting that aside the question is it really a deal that's going to happen. How much will the deal cost? Can Sambanova's hardware be integrated to OneAPI in a timely manner ?
Would they be able to execute GPU based AI roadmap well enough along with it without getting affected ?
 
regen1 said:
There's the question of conflict of interest. Putting that aside the question is it really a deal that's going to happen. How much will the deal cost? Can Sambanova's hardware be integrated to OneAPI in a timely manner ?
Would they be able to execute GPU based AI roadmap well enough along with it without getting affected ?
Click to expand...
It's now HW deal but talent/Scaling IP
 
regen1 said:
There's the question of conflict of interest. Putting that aside the question is it really a deal that's going to happen. How much will the deal cost? Can Sambanova's hardware be integrated to OneAPI in a timely manner ?
Would they be able to execute GPU based AI roadmap well enough along with it without getting affected ?
Click to expand...

I wonder what it says about Jaguar Shores, if Intel is shopping for yet another start-up.
 
Last edited:
511 said:
And this would mark another mess of a deal they need to focus on GPGPU not shop for startups.
Click to expand...
There was another thread where @adroc_thurston said that Intel should have shipped, whatever they had (good / bad / evil / mediocre).

With iterations + feedback, with use of chiplets that Intel is getting pretty good at, they could have found a combination, niche, use case and a customer.

In 2022/2023, Intel and AMD were seen as being in similar (un)competitive position vs. Nvidia. But since that time, AMD shipped 4 products and Intel cancelled everything.

N+2 was always described to be solving all the shortcoming of N+1, and by cancelling N+1, N+2 will get to market so much faster. Except it never happened.

Some exuberant Intel fans on Twitter say that this will be a game changer with Intel Foundry, filling Intel fabs. So out of curiosity, I asked Grok for a ballpark estimate, to estimate from known SambaNova shipments, how many TSMC wafers per month it needs.

Grok answer was 2 wafers per month.
 
Last edited:
Joe NYC said:
Some exuberant Intel fans on Twitter say that this will be a game changer with Intel Foundry, filling Intel fabs. So out of curiosity, I asked Grok for a ballpark estimate, to estimate from known SambaNova shipments, how many TSMC wafers per month it needs.
Click to expand...
Twitter Fans RIP How is this different from Habana/Nervana Intel has always been on the right track but they simply fumble at the most important moment.
Co Packaged Optics Intel had years of R&D before Nvidia and they have working prototype but no one talks about them they are pioneer in Glass Substrate and are ahead of TSMC/Samsung and yet we only heard it was cancelled like how is it not dumb.
AVX-512 they were on the correct path fumbled with ADL.
Optane crazy tech missed opportunity.
I can go on things they keep fumbling.
Joe NYC said:
Grok answer was 2 wafers per month.
Click to expand...
Grok needs work cause 2 Wafer is totally wrong
 
511 said:
Grok needs work cause 2 Wafer is totally wrong
Click to expand...

Of course, nobody would process 2 Wafers per month, but Grok took the total shipments (certain number racks), how many dies went into them and over what period. Using die calculator of how many good dies per wafer.

The key input into this is how many (or more precisely how few) racks SambaNoa actually sold to customers. They could find info on about 10-15 racks.

1765452182702.png
 
rollingout.com

Intel closing in on SambaNova for $1.6 billion

Intel is in advanced talks to acquire AI chip startup SambaNova for $1.6 billion, a steep discount from its $5 billion valuation in 2021.
rollingout.com

1.6Bn $ is too much to waste at this stage if it turns out to be another bad acquisition.

511 said:
It's now HW deal but talent/Scaling IP
Click to expand...
That remains to be seen though Intel definitely has issues to be sorted with their GPU-AI team.
Joe NYC said:
I wonder what it says about Jaguar Shores, if Intel is shopping for yet another start-up.
Click to expand...
Whether JSH is competitive or not but it should at least solve(if not canceled) rack-scale limitations of Falcon Shores. Ideally they should focus on s/w stack with the help of Battlematrix, Crescent Island, etc. and get it to a decent state before JSH release. Sambanova doesn't have a good reputation regarding software.

Joe NYC said:
Intel should have shipped, whatever they had (good / bad / evil / mediocre).
Click to expand...
Joe NYC said:
In 2022/2023, Intel and AMD were seen as being in similar (un)competitive position vs. Nvidia. But since that time, AMD shipped 4 products and Intel cancelled everything.
Click to expand...
They tried with PVC but there weren't many takers, neither much from the likes of Dell, SMCI, etc. It is getting discontinued next month. Inventories were/are shredded, donated to Unis, etc. RTB(Rialto) was incremental/~30%? over PVC, even worse wrt timing. Could have tried to get some deals with Falcon Shores or projected it for inference.

511 said:
How is this different from Habana/Nervana Intel has always been on the right track but they simply fumble at the most important moment.
Click to expand...
At least Movidius IP gets used in their NPUs. Habana and Nervana were not managed well, more so with Nervana.
 
www.crn.com

Intel Hires Qualcomm Executive To Lead GPU Engineering For Data Centers

Intel has hired a Qualcomm executive who led the rival’s GPU engineering efforts to aid with the semiconductor giant’s latest attempt at building AI chips for data centers.
www.crn.com www.crn.com

www.crn.com

Exclusive: Intel Taps Ex-Arm, HPE Exec For Data Center Systems Post Amid AI Reorg

Intel has hired a former Arm and HPE executive to take on a new leadership role focused on data center systems and solutions as its CEO made another AI-related reorganization, CRN has learned.
www.crn.com www.crn.com
In detailing Demers’ responsibilities, Kechichian said that the executive, who previously led Qualcomm’s GPU efforts, will lead GPU IP engineering and oversee Intel’s data center GPU solutions. This will involve “orchestrating the work across architecture, compilers and drivers to successfully develop integrated hardware and software,” he added
Click to expand...
With the appointment, Kechichian said that Keith Rowe’s GPU IP and architecture team, which previously reported to Allegrucci, will now move to Demers’ team.
Click to expand...

As a result of this reorganization, Kechichian said that Jean-Didier Allegrucci, vice president of AI system-on-chip engineering, will report directly to him. Dubé and Demers are also becoming direct reports for the data center leader.
Click to expand...
Eric Demers will be responsible for GPU IP and would report to Kevork(btw the same guy who seemingly caused tremors about Intel's future Open Source contributions at ITT 2025).

The reorganization represents a reversal by Tan, who moved Intel’s AI accelerator chip group out of what was previously called the Data Center and AI Group last April, roughly a month after he became the chipmaker’s CEO.

The move also shakes up Intel’s organizational chart once again with Kechichian taking over responsibilities for the company’s AI accelerator chip efforts from Tan.
Click to expand...
 
regen1 said:
Eric Demers will be responsible for GPU IP and would report to Kevork(btw the same guy who seemingly caused tremors about Intel's future Open Source contributions at ITT 2025).
Click to expand...
I just want to say F Kevork for destroying your software expertise.
 
Someone asked about Clearwater Forest during Intel earnings call, and LBT did seem to even bother answering. Just listed other things that are a priority.
 
While they mention Coral Rapids time and again they don't make any mention of
DMR-HD(RRF) or their E-core server roadmap.
 
adroc_thurston said:
Those have no customers.
Click to expand...
Relatively low for SRF but the likes of Alibaba, Tencent and Telco had there share apart from other CSPs. CWF could've been better if it wasn't that delayed.
Internal target for them were the likes of Graviton, Ampere besides other markets.

RRF theoretically would be(have been?) lot better and versatile than its predecessors.
 
regen1 said:
Relatively low for SRF but the likes of Alibaba, Tencent and Telco had there share apart from other CSPs
Click to expand...
Microscopic volume footprint.
regen1 said:
CWF could've been better if it wasn't that delayed.
Click to expand...
It was optimistically Turin-D perf at hella money.
regen1 said:
Internal target for them were the likes of Graviton, Ampere besides other markets.
Click to expand...
See, that's the thing, Graviton and friends are not really core spam platforms per se; they focus very much on scale up perf.
Ampere is dead for the same reasons Atom Xeons are dead.
regen1 said:
RRF theoretically would be(have been?) lot better and versatile than its predecessors.
Click to expand...
It gets vivisected by the SP7 Venice.
Never had any chances.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top