The ARM server market is tough, seems its been on life support for ~5 years and non-existent. Hopefully with more players entering the market (QCOM? AMD coming back??), SW devs and users will port more apps and expand the ecosystem.
Not sure why the last slide has a Nvidia logo and a GPU card in it, those obviously aren't powered by X-Gene.
These things have PCIe lanes and I assume PCIe slots. Why wouldn't you be able to shove an Nvidia cards into one of their mainboards and use it for the mentioned HPC stuff? If you aren't interested in Intels or AMDs big cores and just rely on GPU computing, this might be more cost efficient.
I doubt AMD will reopen their ARM efforts unless the market takes off. They don't have a lot of financial room for more projects. They'd probably be better off spending the money on trying to get their GPUs and x86 CPUs in data centers.
The companies that seem to have the best chance to successfully put ARM chips into servers are Qualcomm, Cavium, and Project Denver. It's encouraging because they seem to be getting at least some support from various parts of the software ecosystem.
Agree with all that. As far as I can see the main aim of the early ARMv8 server chips was to fight Intel on networking and storage with the server market as a bight future target. That's what Broadcom and AppliedMicro looked to be doing but they have both been taken over by companies who aren't interested in that angle, Cavium is the only one left standing of the original lot and has taken over Broadcom's effort. It is interesting to see that X-Gene is still alive as well but I'd have thought they'd need to be joined up with some other company. Qualcomm is interesting in that it seems to be taking direct aim at the sever market despite having such a huge interest in wireless communication.
Handset growth has stagnated, so I think Qualcomm is looking for more growth opportunities. Of course, they are well-positioned for 5G, but right now data center is where the growth is. I have read that hyperscalers have actually been encouraging Qualcomm to develop ARM server chips because they want more competition with Intel. Whether that's true or not, Qualcomm has deep pockets and experience designing ARM chips, so an ARM server chip is an obvious avenue for them to try to get into the data center. The fact that Intel has moved into baseband perhaps gives Qualcomm even more impetus.
I think that trying to break into networking and storage servers seems like a good strategy. I also think that Cavium's "workload-optimized" strategy seems to make sense for data centers. But perhaps Intel is fighting hard on the sales side to prevent it from happening. Mostly what hyperscalers want is better bargaining power with Intel, so maybe they are happy just buying the Intel processors at cheaper prices than they would pay otherwise. Cavium also has relationships with data centers, though, and might be able to work their way in if they can develop a competitive processor. Obviously, through buying Vulcan IP from Broadcom they have indicated that they intend to expand their ARM server efforts.
Oh, one other note. Both Vulcan and X-Gene were on the market at the same time and Cavium chose to buy Vulcan. Vulcan is reportedly a high power and high performance design, so the fact that Cavium chose Vulcan further suggests they want to expand into server CPUs aggressively.
Oh, one more little note. :D I think maybe part of the reason Cavium bought Q-Logic was to improve their data center relationships so that they could more easily place their server chips. To me, all signs point to Cavium wanting to aggressively pursue server chips, even if they are being low key about it.
Project Denver was NVIDIA's code name for their custom 64-bit ARMv8 CPU. I believe NVIDIA originally intended to develop it into a server chip, but significantly scaled back their efforts when the first push of ARM processors into servers didn't take hold. I wonder if this new company has been formed by former NVIDIA employees.
It's now a year old already and not sold commercial. So what else than an utter failure can this be called? ARM server companies are failing left and right. No one will buy this X-Gene now simply because it has a very unknown future and could be canceled at moments notice. 0% guarantee you get a future upgrade and even if you have a contract on that, if they have like 5 customers it's still cheaper to fuck them over and pay than develop a successor.
What ARM server chip makers always got wrong is that they didn't use the latest cutting edge process that was available to them. Sure, ARM has its advantages in power efficiency, but the advantage is not that great to allow them to be 1.5 generations behind in process node technology compared to Intel. And that's disregarding software support and such.
So ARM chip makers will need to not only beat Intel on perf/W and perf/price, but they also need to beat AMD now, because anyone who is going to consider switching to something other than Intel, is going to take AMD's EPYC chip into account, too.
So far only Qualcomm has been serious about using a cutting edge process, and it also has a little more more than all the other ARM competitors so far, so we'll see how it goes.
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.
13 Comments
Back to Article
webdoctors - Tuesday, October 31, 2017 - link
The ARM server market is tough, seems its been on life support for ~5 years and non-existent. Hopefully with more players entering the market (QCOM? AMD coming back??), SW devs and users will port more apps and expand the ecosystem.Not sure why the last slide has a Nvidia logo and a GPU card in it, those obviously aren't powered by X-Gene.
Death666Angel - Tuesday, October 31, 2017 - link
These things have PCIe lanes and I assume PCIe slots. Why wouldn't you be able to shove an Nvidia cards into one of their mainboards and use it for the mentioned HPC stuff? If you aren't interested in Intels or AMDs big cores and just rely on GPU computing, this might be more cost efficient.Yojimbo - Tuesday, October 31, 2017 - link
I doubt AMD will reopen their ARM efforts unless the market takes off. They don't have a lot of financial room for more projects. They'd probably be better off spending the money on trying to get their GPUs and x86 CPUs in data centers.The companies that seem to have the best chance to successfully put ARM chips into servers are Qualcomm, Cavium, and Project Denver. It's encouraging because they seem to be getting at least some support from various parts of the software ecosystem.
Dmcq - Wednesday, November 1, 2017 - link
Agree with all that. As far as I can see the main aim of the early ARMv8 server chips was to fight Intel on networking and storage with the server market as a bight future target. That's what Broadcom and AppliedMicro looked to be doing but they have both been taken over by companies who aren't interested in that angle, Cavium is the only one left standing of the original lot and has taken over Broadcom's effort. It is interesting to see that X-Gene is still alive as well but I'd have thought they'd need to be joined up with some other company. Qualcomm is interesting in that it seems to be taking direct aim at the sever market despite having such a huge interest in wireless communication.Yojimbo - Thursday, November 2, 2017 - link
Handset growth has stagnated, so I think Qualcomm is looking for more growth opportunities. Of course, they are well-positioned for 5G, but right now data center is where the growth is. I have read that hyperscalers have actually been encouraging Qualcomm to develop ARM server chips because they want more competition with Intel. Whether that's true or not, Qualcomm has deep pockets and experience designing ARM chips, so an ARM server chip is an obvious avenue for them to try to get into the data center. The fact that Intel has moved into baseband perhaps gives Qualcomm even more impetus.I think that trying to break into networking and storage servers seems like a good strategy. I also think that Cavium's "workload-optimized" strategy seems to make sense for data centers. But perhaps Intel is fighting hard on the sales side to prevent it from happening. Mostly what hyperscalers want is better bargaining power with Intel, so maybe they are happy just buying the Intel processors at cheaper prices than they would pay otherwise. Cavium also has relationships with data centers, though, and might be able to work their way in if they can develop a competitive processor. Obviously, through buying Vulcan IP from Broadcom they have indicated that they intend to expand their ARM server efforts.
Yojimbo - Thursday, November 2, 2017 - link
Oh, one other note. Both Vulcan and X-Gene were on the market at the same time and Cavium chose to buy Vulcan. Vulcan is reportedly a high power and high performance design, so the fact that Cavium chose Vulcan further suggests they want to expand into server CPUs aggressively.Yojimbo - Thursday, November 2, 2017 - link
I think this Project Denver team might have to do a lot of work to turn X-Gene into something successful.Yojimbo - Thursday, November 2, 2017 - link
Oh, one more little note. :D I think maybe part of the reason Cavium bought Q-Logic was to improve their data center relationships so that they could more easily place their server chips. To me, all signs point to Cavium wanting to aggressively pursue server chips, even if they are being low key about it.mpbello - Tuesday, October 31, 2017 - link
This is very significant. The resulting company will be the only company in the world that will have ARM server processors as its core business.Yojimbo - Tuesday, October 31, 2017 - link
Project Denver was NVIDIA's code name for their custom 64-bit ARMv8 CPU. I believe NVIDIA originally intended to develop it into a server chip, but significantly scaled back their efforts when the first push of ARM processors into servers didn't take hold. I wonder if this new company has been formed by former NVIDIA employees.Yojimbo - Tuesday, October 31, 2017 - link
Here's an NVIDIA blog post from January, 2011: https://blogs.nvidia.com/blog/2011/01/05/project-d...beginner99 - Wednesday, November 1, 2017 - link
It's now a year old already and not sold commercial. So what else than an utter failure can this be called? ARM server companies are failing left and right. No one will buy this X-Gene now simply because it has a very unknown future and could be canceled at moments notice. 0% guarantee you get a future upgrade and even if you have a contract on that, if they have like 5 customers it's still cheaper to fuck them over and pay than develop a successor.Krysto - Wednesday, November 1, 2017 - link
What ARM server chip makers always got wrong is that they didn't use the latest cutting edge process that was available to them. Sure, ARM has its advantages in power efficiency, but the advantage is not that great to allow them to be 1.5 generations behind in process node technology compared to Intel. And that's disregarding software support and such.So ARM chip makers will need to not only beat Intel on perf/W and perf/price, but they also need to beat AMD now, because anyone who is going to consider switching to something other than Intel, is going to take AMD's EPYC chip into account, too.
So far only Qualcomm has been serious about using a cutting edge process, and it also has a little more more than all the other ARM competitors so far, so we'll see how it goes.