• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."
  • Community Question: What makes a good motherboard?

Discussion Marvell exit the merchant ARM business

uzzi38

Golden Member
Oct 16, 2019
1,227
2,263
96

From the looks of it despite their presentation and honestly very interesting product portfolio, it seems Marvell are ducking out of the merchant ARM business. There'll probably be official articles soon, but this is direct from their earnings call, so it seems like it's seriously happening.

That leaves only two companies - Ampere and Nuvia now I believe. What's everyone's take on what this means for the ARM server ecosystem if Marvell - who were looking incredibly strong IMO - can't get a proper foothold?
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,153
1,674
136
The problem for big established businesses is they have A PnL to run. What everyone on the ARM side likes to ignore is that AMD/intel server development costs are offset by desktop/highend desktop/laptop. So its not surprising that big businesses that cant easily share a very large development cost around multiple products targeting multiple markets end up re-targeting.

Startups dont have this problem because at this stage their spending someone else's money so it allows them to take the risk.
 

LightningZ71

Senior member
Mar 10, 2017
583
521
106
Honestly, they are all in a very tough situation. They have to, essentially, engineer 90% of a new platform, to even begin to have a complete solution for their customers. Back in the bad old days, AMD, Cyril, IBM, and a few others were able to essentially just engineer a chip with the required pin out to fit into the existing socket on a PC motherboard. Now, the amount of IP licensing involved in that is both prohibitively expensive and likely not really supportable under current UEFI models. You have to start from the ground up, and only really have access to universal standards, which are still often not royalty free. Then, once you have something that works, you have to prove that it works, and not just that, it has to work that much better than any of the entrenched competitors that someone is willing to pay the cost to convert their operations to your platform on top of the cost of the actual metal itself.
Amazon, Apple and probably Alphabet/Google can handle that without breaking a sweat. Who else will be able to take that risk? How does Fujitsu do it? They have an established ecosystem and are selling essentially turn-key supercomputer solutions. How does IBM power do it? An established ecosystem with a VERY high cost of conversion and a highly risk averse client base is their game.

I personally see only space for established mobile SOC vendors, that have vast revenue streams to play with, to play in the low cost arena with the likes of Chromebooks or Android convertible tablets to establish critical mass. Marvel might have been able to do that as it’s cost would have been much lower and those sorts of chips fit into its existing volume portfolio better. Then, they could have leveraged the android App Store to do their software distribution with vendors working to do ports of important software. With Apple pushing the ARM message in their closed garden, it would at least have increased industry knowledge of ARM programming.

Maybe Qualcomm or Samsung can get somewhere some day.
 
  • Like
Reactions: uzzi38 and moinmoin

Doug S

Senior member
Feb 8, 2020
414
595
96
The problem for big established businesses is they have A PnL to run. What everyone on the ARM side likes to ignore is that AMD/intel server development costs are offset by desktop/highend desktop/laptop. So its not surprising that big businesses that cant easily share a very large development cost around multiple products targeting multiple markets end up re-targeting.

Startups dont have this problem because at this stage their spending someone else's money so it allows them to take the risk.
Yes this is overlooked by many, and is the main reason Intel was able to knock out the incumbent RISC players who not only had that (then even smaller, in the pre cloud days) market divided up several ways between them but their "volume" market was $10K workstations. Which sold in the thousands of units per month versus the millions per month that PCs sold.
 

DrMrLordX

Lifer
Apr 27, 2000
16,654
5,669
136
Using one of ARM's reference server platforms is supposed to mitigate a lot of the R&D side. No telling how much Amazon spent implementing Neoverse N1 with reduced L3 anyway (Graviton2). ThunderX3 was supposed to be a bit more exotic, with SMT4.

And Huawei seems to be using their establish HiSilicon unit to support their Kunpeng R&D.
 

Thala

Golden Member
Nov 12, 2014
1,128
441
136
Using one of ARM's reference server platforms is supposed to mitigate a lot of the R&D side. No telling how much Amazon spent implementing Neoverse N1 with reduced L3 anyway (Graviton2). ThunderX3 was supposed to be a bit more exotic, with SMT4.

And Huawei seems to be using their establish HiSilicon unit to support their Kunpeng R&D.
The nice thing about ARM IP is, that you do not have to just license Neoverse completely but its building blocks separately. It makes it possible for instance to license the mesh interconnect and memory controllers but not the cores - bringing in you own IP.
 

ASK THE COMMUNITY