[NewsPoster]

HP is targeting Mac power users by allowing them to tap into its high-end Z Workstations remotely using its new Remote Graphics Software for OS X. Previously only available for PC and Linux users, the new version 7.1 has been ported to the Mac platform. The host HP Z workstation uses a "sender" plugin that connects with Receiver for OS X on a Mac, allowing both machines to share screen and apps. The announcement has been timed to coincide with the launch of the new third-gen HP Z workstations, including a new AIO model, that can feature up to 44 CPU cores and Nvidia's latest high-end Quadro GPUs.


The new HP Receiver software for Mac allows users to connect remotely to a Windows or Linux workstation for joint development, interactive edits and design reviews. In particular, it allows users to work with high-performance 2D, 3D video and other media rich applications. Mac users can also view rich content on a single display or on multi-monitor set ups. The system supports up to 4K Ultra HD transmissions at up to 60fps by sending pixels, not data, using HP's proprietary HP3 codec. However, depending on bandwidth, Mac users may not see the footage in native 4K, with the software adjusting resolution accordingly. With Apple failing to update the Mac Pro since its introduction in 2013, HP and other PC makers are looking to encourage film makers and other creative professionals across to its platform. HP has even set up a From Mac to Z page on its site highlighting case studies where users have transitioned from the Mac Pro to HP Z workstations. The latest HP Z840 is fitted with the latest sixth-gen Intel E5-2600 v4 Xeon chips and Nvidia Quadro graphics solutions that far outstrip the performance of the best Mac Pro configuration currently on offer. The RGS receiver is available for free to Mac users, while RGS sender plug-in is pre-installed on HP Z workstations. The RGS software can also work on other, non-HP Windows or Linux machines, but this comes with a software licensing fee.

[prl99]

I went to HP's store page and tried to configure a 44c HP Z840 and I could only configure the first processor at 22cores. There wasn't a matching second 22c processor. Base Z840 configuration starts at $2967, E5-2699v4 2.2 2400 22C 1st CPU adds $6,660 and you need faster RAM (128GB DDR4-2400 extra for $3200) and I haven't even added the second CPU or internal storage. I'm already at $12,577.

Yes, you can configure a monster desktop if you're willing to spend north of $30K but you're still running Windows (or maybe linux but custom order didn't list this feature) and I have to wonder how many Windows apps can make use of this many cores.

As for using it as a secondary computing system with a Mac front-end, I'd rather look at combining Mac Pros into a cluster computing system than mess with a PC. I'd also look for applications that make use of the GPU instead of only the CPU. Having 22 cores could be nice but they aren't as fast or powerful as dedicated GPU computing solutions. HP offers these but they are also very expensive.

[Steve Wilkinson]

I'm looking forward to externalizing GPUs at some point once TB gets fast enough. At first, I thought this was maybe that kind of thing. I think I read an article about a gaming laptop that does something like that.

[Mike Wuerthele]

This is fancy remote access. TB was fast enough at Thunderbolt 2 for GPU use.

[Steve Wilkinson]

re: fancy remote access - yea

I'm not sure TB2 was fast enough from what I read... possibly to put one GPU external for some uses. But, certainly not fast enough to toss 2+ high-performance GPUs in a case. (i.e.: I think TB2 is 20 gigabits/sec, while PCIe is 16 gigaBYTES/sec = 128 gigabits/sec)

I don't know how much bandwidth is actually used/necessary, but even TB3 is only a fraction of even a fairly old internal bus standard.

[Makosuke]

Thing with eGPU is, depending on what you're doing with it the vast amount of bandwidth a 16x PCIe slot has may not provide a big advantage.

BareFeats took an NVIDIA Titan X (mid-2015's top of the line, with 12GB RAM) and compared its performance in a 16x PCIe slot in a 2010 MP (64Gbit throughput, I believe) vs. as an eGPU in a 2013 MP (20Gbit throughput via TB2). Results varied considerably depending on the benchmark, but using OctaneBench (a CUDA GPU rendering test), the eGPU on TB2 only lost about 8% of its performance vs one on a full 16x bus. Games tended to suffer more, with the worst result being a 25% decrease in performance (though still quite fast).

http://barefeats.com/tube21.html

Not to say that more bandwidth isn't better (TB3 @ 40Gbit will be welcome, and a hypothetical TB4 @ 80Gbit will make eGPUs even more practical), but GPU speed hit isn't necessarily anywhere near proportional to raw bandwidth, and depending on your use case an eGPU might be quite practical.

[Steve Wilkinson]

@Makosuke - Yes, that's the kind of thing I'm thinking about... but isn't it kind of proof-of-concept at this point? Can you buy one. I also saw something like that for a higher-end gaming laptop, I think.

But, yea, I suppose it depends on how thoroughly the bus is saturated for a given application. That's good to see, for sure! I guess that's a bit typical for Apple, sometimes making some moves years ahead of it actually being real-world practical. Maybe with TB3, we're now close to reality.