News Intel Engineer Outs Panther Lake Architecture on LinkedIn

[bit_user]

Pretty soon Broadcom will have priced out small companies from VMware.

What will they use, then? KVM or Hyper-V? What are the main benefits of using VMware over cheaper/free options?

 

[jeremyj_83]

What will they use, then? KVM or Hyper-V? What are the main benefits of using VMware over cheaper/free options?

Hyper-V isn't cheap either. MS went to a per core license model starting in 2016 and they give you 16 total cores/host "free" and then every core after that is licensed in 2 core packs. For a dual socket 32c system it comes out to like $60k/host for licensing.

I have heard of companies using KVM and Proxmox instead of Hyper-V or VMware. The biggest benefits in using Hyper-V or VMware are support and training. AFAIK for KVM and Proxmox your support is basically forums. With Hyper-V and VMware you get actual support from company trained professionals. The other thing is you have tons of classes for training on Hyper-V and VMware. Both MS and VMware offer their own HCI storage as well which is nice. KVM requires you to use Ceph which is another open source thing so training and support is more lacking.

 

[bit_user]

for KVM and Proxmox your support is basically forums.

I think enterprise Linux vendors, like Redhat, probably will provide support for licensees of RHEL.

KVM requires you to use Ceph

You mean, if you want shared storage that's accessible by VMs running on different physical machines? Does it have to be Ceph, or what about GlusterFS or others?

https://en.wikipedia.org/wiki/Comparison_of_distributed_file_systems

 

[jeremyj_83]

You mean, if you want shared storage that's accessible by VMs running on different physical machines? Does it have to be Ceph, or what about GlusterFS or others?

https://en.wikipedia.org/wiki/Comparison_of_distributed_file_systems

Ceph, VMware vSAN, Hyper-V Storage Spaces Direct, etc... are all forms of Hyperconverged Infrastructure (HCI). Basically you have a server with say 24x 2.5" drive bays. You can use a cluster of hosts, VMware best practice is 4 minimum, and have an amount of storage in each bay and you pool it together to be accessible by VMs. You use this instead of a physical SAN for your shared, usually block, storage. Doing HCI does utilize a bit of your hosts CPU cycles and RAM but you save rack space compared to a SAN. Also a lot of HCI solutions are based on Ethernet so that is cheaper than Fibre Channel and Ethernet speeds have increased MUCH faster than FC as well.

I'm not sure if KVM supports GlusterFS. I know that it supports Ceph as AWS is KVM with Ceph storage.

 

[SiliconFly]

Well I just got updates from sources in Intel and they say that meteorlake desktop has been cancelled and some ML laptop dies will fill the lower ranks of the raptor lake refresh that they will be rushing to market because arrow lake on Intel 3 (it will not be on 20A as you say) is being delayed to 2024 due to yield/defect issues. Definitely not a rocking client side either.

But I do agree with you on AMD, if their completely new ground up core design for Zen5 isn’t impressive, AMD will need to abandon their high price points.

ARL is a 20A product according to multiple Intel press releases itself. And it's bit too late to simply change nodes from 20A to 3 at this point in time, if not impossible. If some successor to MTL is cooking in Intel 3, it'll have to be a MTL refresh and not ARL.

Kindly double check your source. Seems to be a bit off track.

 

[SiliconFly]

That's not how it works. Moving to a smaller node makes CPUs faster in two main ways:

1. Gives you additional transistor budget that you can spend on architectural improvements or things like more cache.
2. Enables higher frequencies, as shrinking distances within the same design naturally results in tighter timing.

The new wrinkle that's recently come to light is that SRAM scaling has broken down. So, increased density applies mostly just to logic.

TSMC's 3nm Node: No SRAM Scaling Implies More Expensive CPUs and GPUs

Big problems from tiny memory cells.

www.tomshardware.com

Say goodbye to free cache enlargements. This is one huge reason AMD's strategy of using chiplets for cache (on both CPUs and GPUs) is so key. Intel will have to follow. In recent generations, Intel has been increasing cache quite a lot. That can't continue, without significantly affecting price.

That's why I'm saying node improvements ain't what they used to be.


They're rumored to have cancelled Meteor Lake for the desktop. If so, that must be counted as a miss. If the desktop spends a 3rd year on Golden Cove, a lot of people are going to start having that Skylake de ja vu.

You're right. Both TSMC & Intel are having trouble with SRAM scaling. 😪

 

[The Historical Fidelity]

ARL is a 20A product according to multiple Intel press releases itself. And it's bit too late to simply change nodes from 20A to 3 at this point in time, if not impossible. If some successor to MTL is cooking in Intel 3, it'll have to be a MTL refresh and not ARL.

Kindly double check your source. Seems to be a bit off track.

Good catch, sorry, I assumed they were talking of Intel 3, but it’s actually TSMC N3 which is being used to manufacture Arrowlakes GPU tile. Intel revised their TSMC purchase order’s start date to Q4 2024 which means realistically we won’t see Arrowlake until Q2 2025. However, I believe it is the 20A process which is giving them yield/defect issues as the source states that TSMC N3 is surpringly better than Intel anticipated.

 

[SiliconFly]

Good catch, sorry, I assumed they were talking of Intel 3, but it’s actually TSMC N3 which is being used to manufacture Arrowlakes GPU tile. Intel revised their TSMC purchase order’s start date to Q4 2024 which means realistically we won’t see Arrowlake until Q2 2025. However, I believe it is the 20A process which is giving them yield/defect issues as the source states that TSMC N3 is surpringly better than Intel anticipated.

Agree. Intel 20A uses multi-patterning like before. And 18A even more!!! Possibly a repeat of 10++++++ fiasco again (in spite of EUV). I see only two scenarios:

(1) The new team is unbelievably stupid, they never learns from it's past mistakes and are bound to repeat them again & again.
(2) The new team is unbelievably awesome, they managed to crack the multi-pattering monster to get the yields up in both 20A & 18A.

To be honest, both don't sound right.

I suspect, 20A will have mediocre yields and performance issues to begin with. And Intel will have low production, high cost & supply issues with both 20A & 18A at start. And it's gonna take a while for them to ramp up fully.

It's a very difficult road ahead for them. Only time can give us answers!

 

[edzieba]

The Rankine cycle is still a very relevant and accepted thermodynamics process…not an obsolete unit of measure no longer supported by all authorities in measures.

Not the Rankine cycle, the absolute-zero-referenced-Fahrenheit unit-of-temperature Rankine. You will still find it in active use to this day in rocketry.

 

[The Historical Fidelity]

Not the Rankine cycle, the absolute-zero-referenced-Fahrenheit unit-of-temperature Rankine. You will still find it in active use to this day in rocketry.

Oh okay I get you. Yeah, unfortunately to maintain math compatibility with historical aerospace programs and tools, they continue to create new versions that continue to use the Rankine. However, at least there is a reason to continue use the Rankine, whereas there is no reason, historical trend, etc. to use the angstrom in the semiconductor industry.