Qualcomm Announces its First Socketed Enterprise CPU

[HumanSmoke]

for big supercomputers and data centers I don't think that arm can do anything because Oracle and IBM are fighting X86 for long time.

Just needs to gain traction. ARM supercomputing has been a thing for only a few years. Now that 64-bit ARM is available, you will see more implementations along the lines of BSC's Mont-Blanc

Nvidia Knew that they can't fight the big three so their processing unit doesn't replace the CPU and the platform but work with it that what make Tesla survive else way Tesla wouldn't have any chance.

Of course. Nvidia's own roadmap targets ARM+Tesla co-processor. That is why the company worked to integrate their hardware+CUDA with x86, and the RISC based OpenPOWER and ARM. When you are a peripheral player, you evolve or become irrelevant.

 

[Musaab]

Just needs to gain traction. ARM supercomputing has been a thing for only a few years. Now that 64-bit ARM is available, you will see more implementations along the lines of BSC's Mont-Blanc

Of course. Nvidia's own roadmap targets ARM+Tesla co-processor. That is why the company worked to integrate their hardware+CUDA with x86, and the RISC based OpenPOWER and ARM. When you are a peripheral player, you evolve or become irrelevant.

BSC's Project is impressive as a concept maybe if they kept with Tegra+CUAD thay may have done something that can be transformed into real world because Nvidia built a big ecosystem around their CODA but for general ARM with random ARM+GPU we are going back to Android's square one, fragmentation and a lot of it. what made wintel platform a hit is the unification. no company my bet their future on fragmented system, Business pay a lot of money for consistency and support and arm in it's present state can't offer either. Tegra+Tesla can be part of a big system contains core parts from the big three (X86, Power and Spark) but as a stand alone, I don't think that is possible in the next five years. Nvidia maybe the only candidate because they worked in servers for along time and their CODA and Tesla are parts of most of the modern supercomputer, but stand alone not even in five years. in big business you need to go smoothly not ramming the walls.

 

[truth teller]

BSC's Project is impressive as a concept maybe if they kept with Tegra+CUAD thay may have done something that can be transformed into real world because Nvidia built a big ecosystem around their CODA but for general ARM with random ARM+GPU we are going back to Android's square one, fragmentation and a lot of it. what made wintel platform a hit is the unification. no company my bet their future on fragmented system, Business pay a lot of money for consistency and support and arm in it's present state can't offer either. Tegra+Tesla can be part of a big system contains core parts from the big three (X86, Power and Spark) but as a stand alone, I don't think that is possible in the next five years. Nvidia maybe the only candidate because they worked in servers for along time and their CODA and Tesla are parts of most of the modern supercomputer, but stand alone not even in five years. in big business you need to go smoothly not ramming the walls.

sweet jesus, the "marketing" spam never stops
and then im the one that

have no grasp on the english language or were so high the keyboard was fighting you

anyway, you are all making a big fuss out of nothing, just because the product now also comes in a socket-able sku doesnt mean it will gain much more traction that it already has.
this is server grade material, for this to become part of a home grade pc developers need to target it, x86 / x86-64 will remain the goto platform for the coming years, that is unless developers start targeting linux+arm. and normally the people who do target that environment dont care about optimizing for it (as long as it works, android comes to mind with the dalvik userspace vm), while those who target x86* normally "care a bit more" about the release/production performance of the software. also x86* binaries tend to be smaller and easier to obfuscate.

 

[Uplink10]

this is server grade material, for this to become part of a home grade pc developers need to target it, x86 / x86-64 will remain the goto platform for the coming years, that is unless developers start targeting linux+arm. and normally the people who do target that environment dont care about optimizing for it (as long as it works, android comes to mind with the dalvik userspace vm), while those who target x86* normally "care a bit more" about the release/production performance of the software. also x86* binaries tend to be smaller and easier to obfuscate.

You are right about the binaries, RISC software does need more lines of code and that is why it is easier for programmers to develop software to be used on CISC where you have more instructions available.
I would just like to add that on Linux OSes the size of the programs is even smaller because the mostly open-source apps use the same libraries and that makes them small, no need to write the code again and reinvent the wheel or take up the space on a drive. Again all of this is possible because of open-source software.

Home PCs are intended for multi-porposes and that is where the CISC is better than RISC.

 

[Musaab]

sweet jesus, the "marketing" spam never stops
and then im the one that



anyway, you are all making a big fuss out of nothing, just because the product now also comes in a socket-able sku doesnt mean it will gain much more traction that it already has.
this is server grade material, for this to become part of a home grade pc developers need to target it, x86 / x86-64 will remain the goto platform for the coming years, that is unless developers start targeting linux+arm. and normally the people who do target that environment dont care about optimizing for it (as long as it works, android comes to mind with the dalvik userspace vm), while those who target x86* normally "care a bit more" about the release/production performance of the software. also x86* binaries tend to be smaller and easier to obfuscate.

I'm from Iraq so my English isn't that good. And yes I am fighting my hp 635 laptop keyboard.
I think you didn't read my two posts and Human Smoke's argue. Qualcomm is big but entering the servers market isn't easy and I talked about NVIDIA because they are the only company entered the servers market in the last ten years and survived. For a company to enter supercomputers and data centers processes market they need huge support from software companies. When I talked about CUDA as example. NVIDIA spent the last five years pushing it in every where, now they have comfortable position in the market and alot of support from software companies and programmers but still they can't make their own platform.

 

[remixedcat]

My first encounter with a Cavium CPU was in the Xirrus arrays I tested.

 

[Musaab]

My first encounter with a Cavium CPU was in the Xirrus arrays I tested.

Cavium in the market for sometime but the talk about it's results is so little that we can't make good judgment about the performance. Maybe you can share some of your results.

 

[truth teller]

I talked about NVIDIA because they are the only company entered the servers market in the last ten years and survived.

nope, there are plenty others, for instance nallatech (now isi) does exactly what nvidia is currently doing in the server market *u-rack-barebone + fpga acceleration

 

[Musaab]

nope, there are plenty others, for instance nallatech (now isi) does exactly what nvidia is currently doing in the server market *u-rack-barebone + fpga acceleration

Nallatech are building cards around Altera and Xilinx's chips. So it's not the same.
It's nice to have someone to keep the conversation on (thanks for that) but it's 1AM here.

 

[truth teller]

Nallatech are building cards around Altera and Xilinx's chips. So it's not the same.

in what way is it not the same? not having fpgas built in-house or having custom software for them? cause i can rebut both points:
1) one company "makes sure" users can only use their closed source software on their fpgas and vice-versa, another uses an opencl backed sdk, instead of vhdl, to program the fpgas
2) one company sources the fpgas from altera and makes the boards in-house, afaik the other company also does not own any fabrication, so it sources their fpgas from tsmc and also makes boards in-house, granted the gpu design came from in-house.
still, you can do a lot more and faster on a real fpga than on a gpu, and being bound to a walled garden that you cant control (nor exit after entering without losing money) is not the best option. and if you are going to bring rendering into play, note that one solution allows easy rendering and processing to be done on any opengl/vulkan enabled gpu, the other doesnt even support a proper opengl pairing (mem copy to host is required for processing), i will leave you to guess which is which

 

[anubis44]

Heyyyy qualcomm.
Sup. Uhm... listen. If you could kinda make an X86 CPU with like, 8 fast cores and not charge 999$ for it, that'd be cool and stuff.

Just sayin, if ur into it.

You just hit the nail on the head. And Qualcomm could indeed do exactly this. All they would need to do is buy AMD. As they have over $22 billion in cash on hand, and AMD would probably cost them about $8-$10 billion to buy, they could do this and still have over $12 billion in cash on hand.

In return, Qualcomm would gain:

-an excellent Jim Keller-designed Zen x86 core CPU, about to be released, that was designed to scale from the largest datacentre CPU clusters to the smallest low-power x86 devices. Oh, and the only other x86 license currently available.
-a world class GPU core in the form of the Radeon Graphics divison, which would allow them to include additional advancements into their SnapDragon ARM chips, as well as break into the HPC GPU market place
-a footprint in the gaming world, which could be parlayed into better gaming chops for portable (either x86 or ARM) chips in Android phones, etc.

This, or an Apple buyout, for similar reasons, would make the most sense. A Microsoft buyout would essentially turn Microsoft into Apple, but with an x86 license to boot, and possibly a monopoly over gaming hardware for some time.

I'm from Iraq so my English isn't that good. And yes I am fighting my hp 635 laptop keyboard.
I think you didn't read my two posts and Human Smoke's argue. Qualcomm is big but entering the servers market isn't easy and I talked about NVIDIA because they are the only company entered the servers market in the last ten years and survived. For a company to enter supercomputers and data centers processes market they need huge support from software companies. When I talked about CUDA as example. NVIDIA spent the last five years pushing it in every where, now they have comfortable position in the market and alot of support from software companies and programmers but still they can't make their own platform.

(bold italics mine)

Perhaps nVidia should merge with/buy out AMD. Then they'd have x86 as well as all the patents/technology from the Radeon division.

 

[Musaab]

You just hit the nail on the head. And Qualcomm could indeed do exactly this. All they would need to do is buy AMD. As they have over $22 billion in cash on hand, and AMD would probably cost them about $8-$10 billion to buy, they could do this and still have over $12 billion in cash on hand.

In return, Qualcomm would gain:

-an excellent Jim Keller-designed Zen x86 core CPU, about to be released, that was designed to scale from the largest datacentre CPU clusters to the smallest low-power x86 devices. Oh, and the only other x86 license currently available.
-a world class GPU core in the form of the Radeon Graphics divison, which would allow them to include additional advancements into their SnapDragon ARM chips, as well as break into the HPC GPU market place
-a footprint in the gaming world, which could be parlayed into better gaming chops for portable (either x86 or ARM) chips in Android phones, etc.

This, or an Apple buyout, for similar reasons, would make the most sense. A Microsoft buyout would essentially turn Microsoft into Apple, but with an x86 license to boot, and possibly a monopoly over gaming hardware for some time.

(bold italics mine)

Perhaps nVidia should merge with/buy out AMD. Then they'd have x86 as well as all the patents/technology from the Radeon division.

NVIDIA +AMD is something nice but almost impossible for two reasons
1- no one will allow the marriage of the only two discrete graphics chips makers in the world.
2- NVIDIA is strong but not strong enough to carry AMD's problems. And if they are smart they have learned the lesson from AMD +ATi.

 

[remixedcat]

Cavium in the market for sometime but the talk about it's results is so little that we can't make good judgment about the performance. Maybe you can share some of your results.

They were dual core 500mhz for the XR630 and the XR520 has a dual core 400Mhz Cavium. Nothing really even comparing to the new dual core 1.4Ghz SOCs in the newer APs. Both of those are MIPS based and NOT ARM.

Here's the benchmarks for those access points: XR630/XR520


In other news the thing is with this QCA CPU here in the article, for windows the server apps will need to be recompiled so this is linux only right now... MS is prolly gonna drag-ass on this, killing the potential for this to succeed. Windows servers are gaining marketshare due to hyper-v being included in the OS cost instead of paying a lot for vmware/commercial Xenserver/Commercial Virtuozzo instead of openvz. OpenVZ,free Xen, and KVM hypervizors do not have as many features as hyper-v does right now.*

Also OVZ and CVZ are both containers rather then full hypervisors. So true hypervizors you will be limited to FZ and KVM.*


* = Server hypervizors used for web hosting and NOT desktop hypervizors like VB parellels, etc..