Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Nvidia to Acquire Mellanox for $6.9B

nvidianews.nvidia.com

Post reply on HN
Print view View original

191–200 of 229 posts

Re: Nvidia to Acquire Mellanox for $6.9B

#191

Post by shaklee3 »

dogma1138 wrote: post #163
shaklee3 wrote: post #121

Earlier quoted context omitted.

Infiniband hasn't increased in speed in a while, while Ethernet has. IB is all but done since Ethernet 200 and 400Gbps will be out soon, and in fact, are already supported by Mellanox switches.

From what I understand the reason it hasn't increased is because primarily of PCIe, it is set to double in speed with PCIe 4.0 and again with 5.0 once those are made more available with maximum speeds of 1.6 and 4.0 TB for IB x16. IB still has lower latency than Ethernet at least on paper especially when it comes to RDMA but I don't know how much of an issue that is for these applications. But overall I'm not sure ho…

Yup, but that's exactly why IB isn't too relevant anymore. Ethernet had plenty of time to catch up because of PCIe-SIG and Intel dragging their feet.

Re: Nvidia to Acquire Mellanox for $6.9B

#192

Post by drewg123 »

nerpderp82 wrote: post #166
drewg123 wrote: post #111

Earlier quoted context omitted.

A lot of IB technology is based on precursors like Myrinet, Quadrics, etc. Those companies were driven from the HPC market and the top500 by Mellanox in the last 15 years. It is arguable that IB won simply because it had better marketing, and a lot more venture capitol behind it. At one point, 6 of the top 10 supercomputers in the top 500 were interconnected with Quadrics, and over 1/3 with Myrinet. However, both Qua…

IB has been doubling in bandwidth with every release, 10,20,50,100 and now 200G. Those other interconnect technologies did not do that.

Honestly, they competed on price & marketing at first. The handwriting was already on the wall for these companies long before IB had any real advantage on speed. We were all limited by PCIe bandwidth, after all. So no matter what the link speed is, you're not going to get more bandwidth than what PCIe can provide.

I worked for Myricom when Mellanox was starting out with IB. I recall stories from that time of customers that would try IB for something like 1/2 to 1/3 of what we charged. But they could never get it to work, and ripped it out and installed Myrinet. Sadly, this made our management smug. Mellanox eventually ate their lunch because we never responded to their marketing and pricing war.

Re: Nvidia to Acquire Mellanox for $6.9B

#193

Post by chx »

bsder wrote: post #182
chx wrote: post #138

Earlier quoted context omitted.

Xilinx is badly overpriced, no matter how you look at it. You could compare its P/E to similar semiconductor companies, you could compare it to all US companies, you could even look at the predicted P/E growth and it's still overpriced. Every analyst I am aware of currently advises you stay away from XLNX.

> Xilinx is badly overpriced, no matter how you look at it. Is it, really? FPGA's are now all over HPC, the 5G rollout is going to put a LOT of really expensive FPGA's in new towers, and high-end networking gear is about to get a bump due to PCI going up. I don't know how much damage the Intel acquisition did to Altera, but if Altera has been AWOL with big customers due to the Intel acquisition, Xilinx may have a mas…

Well, let's look at numbers, shall we? XLNX P/E is 37, INTC P/E is 12, MRVL P/E is 22.8, AVGO is 23.19, the current S&P 500 PE is 21.34. Of course P/E is just one number but it's certainly one important number. Do you have a better comparison that makes XLNX at this price an alluring stock?

My original comment reacted to nVidia acquiring Xilinx and I am trying to explain why that won't happen: it's too expensive.

Re: Nvidia to Acquire Mellanox for $6.9B

#194

Post by shaklee3 »

jabl wrote: post #178
shaklee3 wrote: post #119

Earlier quoted context omitted.

Understand, manage, and buy networking equipment for. Infiniband is a thing of the past, especially with the Mellanox VPI adapters that support both Ethernet and Infiniband with a single bit flipped on the adapter.

IDK, IME IB is pretty much plug and play in an HPC setting. Plain ethernet is too, sure, but if you want to do HPC type workloads you'll have to do a lot of configuration and testing to setup DCB, RoCEv2, EVPN+VXLAN+BGP or such. But I think this is the way the market is going in the longer term.

It is, but if you're a large enough company to be buying millions of dollars in adapters and switches, reading a guide from Mellanox to turn on DCB should be fairly seamless. RoCEv2 is API-compatible with IB for the most part, so there is really no configuration on that layer. The other pieces -- not really sure what you're getting at. Most of those are for going across data centers, which IB won't do anyways. At least Ethernet would give you the option to run RDMA from the east coast to the west coast.

Re: Nvidia to Acquire Mellanox for $6.9B

#195

Post by angry_octet »

bewo001 wrote: post #115
jabl wrote: post #23

This is interesting. Mellanox has apparently been under activist investor pressure to reduce their R&D expenses and pay more dividends. And then there was the rumors that Intel were interested, but apparently Nvidia in the end offered more. From a HPC perspective I think it's good Nvidia got the deal, Intel is already a quite dominating force in that market, and if they'd have gotten the deal it wouldn't have surpris…

While I agree about Intel's market force, they have a much better open source software story than nvidia. Nvidia is as closed as the other big player, Broadcom.

There is a lot of heavily patent/trade secret encumbered IP in both graphics and compute drivers, making open source extremely difficult. Above that layer I've found Nvidia extremely open.

Intel is far worse to deal with, and additionally engages in anti-competitive architectural wars, preventing other vendors from interfacing with the CPU bus.

As a result we have NVLink, and now we will soon have official IB cards with NVLink ports, and probably ARM cores too.

Re: Nvidia to Acquire Mellanox for $6.9B

#196

Post by baruch »

patrickg_zill wrote: post #147
sacheendra wrote: post #99

Earlier quoted context omitted.

infiniband - The high performance interconnect by Mellanox

Infiniband is actually a standard and Mellanox became the go to supplier. However in the early stages there were many more suppliers.

Intel bought one of the IB players and for all I know killed it.

Re: Nvidia to Acquire Mellanox for $6.9B

#197

Post by angry_octet »

abcdef123xyz123 wrote: post #77
jabl wrote: post #23

This is interesting. Mellanox has apparently been under activist investor pressure to reduce their R&D expenses and pay more dividends. And then there was the rumors that Intel were interested, but apparently Nvidia in the end offered more. From a HPC perspective I think it's good Nvidia got the deal, Intel is already a quite dominating force in that market, and if they'd have gotten the deal it wouldn't have surpris…

We have been moving away from IB for our platform (algorithmic trading) since Ethernet now has almost comparable latency and is a lot easier to understand and manage.

Latency really isn't comparable, but now probably isn't an issue for algorithmic trading. Still many other closely coupled codes for which it dominates cluster performance.

IB is actually easier to reason about and debug than DCE, but obviously a different community of practice.

Re: Nvidia to Acquire Mellanox for $6.9B

#198

Post by dogma1138 »

shaklee3 wrote: post #191
dogma1138 wrote: post #163

Earlier quoted context omitted.

From what I understand the reason it hasn't increased is because primarily of PCIe, it is set to double in speed with PCIe 4.0 and again with 5.0 once those are made more available with maximum speeds of 1.6 and 4.0 TB for IB x16. IB still has lower latency than Ethernet at least on paper especially when it comes to RDMA but I don't know how much of an issue that is for these applications. But overall I'm not sure ho…

Yup, but that's exactly why IB isn't too relevant anymore. Ethernet had plenty of time to catch up because of PCIe-SIG and Intel dragging their feet.

It doesn’t matter nearly all Mellanox ASICs support both IB and Ethernet they are the ones who are driving the speed of Ethernet to this level.

You also need to use their cables and transceivers (or a similar alternative) for these speeds doesn’t matter if you are using Ethernet or IB.

Re: Nvidia to Acquire Mellanox for $6.9B

#199

Post by jabl »

shaklee3 wrote: post #194
jabl wrote: post #178

Earlier quoted context omitted.

IDK, IME IB is pretty much plug and play in an HPC setting. Plain ethernet is too, sure, but if you want to do HPC type workloads you'll have to do a lot of configuration and testing to setup DCB, RoCEv2, EVPN+VXLAN+BGP or such. But I think this is the way the market is going in the longer term.

It is, but if you're a large enough company to be buying millions of dollars in adapters and switches, reading a guide from Mellanox to turn on DCB should be fairly seamless. RoCEv2 is API-compatible with IB for the most part, so there is really no configuration on that layer. The other pieces -- not really sure what you're getting at. Most of those are for going across data centers, which IB won't do anyways. At lea…

> The other pieces -- not really sure what you're getting at.

What I'm getting at is setting up clusters larger than what you can fit behind a single switch. So you'll want e.g. a CLOS fabric with multipathing (the typical IB setup, FWIW). As Trill and SPB seem pretty dead, it seems the momentum is to do the multipathing at the L3 level, using the aforementioned EPVN+VXLAN+BGP, or something similar.

Re: Nvidia to Acquire Mellanox for $6.9B

#200

Post by baybal2 »

jabl wrote: post #186
SEJeff wrote: post #112

Earlier quoted context omitted.

And yet an ethernet frame, by design, is larger than an infiniband frame (think layer 2). When it comes down to node to node latency, given perfectly equal silicon, infiniband will still be faster.

I think the minimum size of an IB packet with no payload is 26 octets, vs. 64 octets for an eth packet. So sure, a difference of 38 octets, but at, say, 100 Gbit/s, that's less than a nanosecond difference, much much less than the IB vs. ethernet latency difference. So I think you'll have to look somewhere else for information. I have no idea what it is, actually. Some ideas that may or may not matter (or might not e…

Sure, IB is simply is a superior fabric for its niche.

For everything else, RDMA on Ethernet buys you with ability to reuse your L2, and this matters way way more to people running DC businesses than anything else.

Post reply on HN