Advanced search

Message boards : Number crunching : HELP: x16/x16/x16/x16 vs. x16/x8/x8/x8?

Author Message
Profile Tuna Ertemalp
Send message
Joined: 28 Mar 15
Posts: 46
Credit: 1,547,496,701
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 42266 - Posted: 1 Dec 2015 | 17:56:20 UTC
Last modified: 1 Dec 2015 | 17:56:54 UTC

I am building a PC just for BOINC projects. I already ordered all the components, but just got the news that the motherboard I bought ran out of stock before they shipped mine, and it is now on backorder. So, before I buy something else, thought I'd ask here in case some knowledgeable hardware geek has some insights. :)

The CPU is Intel® Core™ i7-5960X Processor Extreme Edition (20M Cache, up to 3.50 GHz). It does support 40 PCI express lanes.

And, four EVGA GeForce GTX TITAN X SC GPUs for a 4-way SLI.

The motherboard was supposed to be ASUS X99-E WS. It does support 4-way SLI, all x16, thus achieving 40 lane 4-SLI via x16/x16/x16/x16. The only other motherboard I found doing that was ASRock X99 Extreme11, but that is out of stock, too.

Any other X99 motherboards that will accept this CPU and these GPUs are "less than" x16/x16/x16/x16, like x16/x8/x8/x8 (e.g. ASUS RAMPAGE V EXTREME).

I hear that this might cause for latency in gaming, but I have zero interest in gaming. Then the question becomes if x16/x8/x8/x8 would hinder this PC from achieving nirvana in terms of BOINC computations (I support 50+ BOINC projects, and a fair number of them, including GPUGRID, use GPU).

Does anyone know for sure, as opposed to having a gut feelings?

Thanks
Tuna
____________

AyalaZero
Send message
Joined: 22 Nov 15
Posts: 8
Credit: 19,131,150
RAC: 0
Level
Pro
Scientific publications
watwat
Message 42268 - Posted: 1 Dec 2015 | 18:46:35 UTC - in response to Message 42266.
Last modified: 1 Dec 2015 | 18:46:47 UTC

I will hold back and not tell you my gut feeling. However, here you go:

https://pcpartpicker.com/search/?cc=us&q=+ASUS+X99-E+WS

Profile Tuna Ertemalp
Send message
Joined: 28 Mar 15
Posts: 46
Credit: 1,547,496,701
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 42269 - Posted: 1 Dec 2015 | 18:51:38 UTC - in response to Message 42268.

I will hold back and not tell you my gut feeling.


LOL. Thx. :)

Profile Tuna Ertemalp
Send message
Joined: 28 Mar 15
Posts: 46
Credit: 1,547,496,701
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 42270 - Posted: 1 Dec 2015 | 19:00:22 UTC

And, just found this on Newegg; must have just come in to stock. The same board with the USB 3.1 built-in, as opposed to missing it or needing a PCI card for it. ASUS X99-E WS/USB 3.1 LGA 2011-v3 Intel X99 SATA 6Gb/s USB 3.1 USB 3.0 CEB Intel Motherboard. So, I am going to order that.

Still, if anybody has actual technical info about my original question, I'd love the education.

Thanks
Tuna

Dayle Diamond
Send message
Joined: 5 Dec 12
Posts: 84
Credit: 1,663,883,415
RAC: 1
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 42287 - Posted: 3 Dec 2015 | 17:12:52 UTC
Last modified: 3 Dec 2015 | 17:13:27 UTC

I also have no technical info.

But it's really cool that you're making a big investment into BOINC!

Have you considered using this opportunity to return your Titan Xs and replace them with 980 Tis?

They are a smidgen more powerful and will save you in the range of $1,200.

You could make real progress on a second computer with that sort of cash.

And if you've started with one Titan X and are looking for SLI, remember that BOINC projects run individually, so you can pair different cards in the same system.

All my best wishes!!

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 0
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 42289 - Posted: 3 Dec 2015 | 19:55:54 UTC
Last modified: 3 Dec 2015 | 19:56:53 UTC

There's no real difference in the performance between PCIe3.0 x8 and x16.
The WDDM overhead of modern Windows OSes have much severe impact on the performance of high end GPUs.
For example my GTX980 under Windows XP is as fast as a GTX980Ti under Windows 7~10 (you can check it under the performance tab).
I'm sorry to see the large amount of computing power wasted this way.
Linux is as fast as Windows XP.
If you put 4 of those Titan X OC's in a host, it will be hot and loud.
You should use water cooling to overcome this. It will also let you to put 7 Titan X OC's in that MB.
After 5 years of GPUGrid crunching I would not build a multi-GPU system based on a very expensive MB and CPU for GPUGrid crunching. Instead I would go for more inexpensive Socket 1150 (or 1151) MB (for example Gigabyte H81M-S2V) with an inexpensive Core i3 CPU (41x0) and 4-8GB RAM, and a small SSD (60GB would be fine).
These cheap MBs usually have RealTek Network Interface Controllers, which have drivers for Windows XP.
The high-end cards don't have drivers for Windows XP, but the driver's inf file can be easily "hacked" to make these cards work under XP.
You will certainly get to the 1st place on the host's toplist with this PC, until someone will use Windows XP (or linux) on the same hardware. :)

Profile Tuna Ertemalp
Send message
Joined: 28 Mar 15
Posts: 46
Credit: 1,547,496,701
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 42290 - Posted: 3 Dec 2015 | 20:41:12 UTC - in response to Message 42289.

There's no real difference in the performance between PCIe3.0 x8 and x16.
The WDDM overhead of modern Windows OSes have much severe impact on the performance of high end GPUs.
For example my GTX980 under Windows XP is as fast as a GTX980Ti under Windows 7~10 (you can check it under the performance tab).
I'm sorry to see the large amount of computing power wasted this way.
Linux is as fast as Windows XP.
If you put 4 of those Titan X OC's in a host, it will be hot and loud.
You should use water cooling to overcome this. It will also let you to put 7 Titan X OC's in that MB.
After 5 years of GPUGrid crunching I would not build a multi-GPU system based on a very expensive MB and CPU for GPUGrid crunching. Instead I would go for more inexpensive Socket 1150 (or 1151) MB (for example Gigabyte H81M-S2V) with an inexpensive Core i3 CPU (41x0) and 4-8GB RAM, and a small SSD (60GB would be fine).
These cheap MBs usually have RealTek Network Interface Controllers, which have drivers for Windows XP.
The high-end cards don't have drivers for Windows XP, but the driver's inf file can be easily "hacked" to make these cards work under XP.
You will certainly get to the 1st place on the host's toplist with this PC, until someone will use Windows XP (or linux) on the same hardware. :)


Thanks for exactly the kind of info I was looking for! :-)

Just for the record: I don't build my machines for BOINC or for GPUGrid. They are machines on which I would do things that range from mundane web browsing to photo/video editing to running 50+ BOINC projects (many of which fully utilize maximum cores/threads at maximum cycles). Thus, when I build something new/replacement for a spot at my home, I first look at the physical space I have at that spot to decide on the size & cooling specs of a case, then try to maximize the computational specs within that case to give both myself and BOINC all it can while the computer is not in use (in this case, it is my home office, under the desk, no size limit, so went all out for a single-CPU system). Anything old kicked out of the rooms at the house either moves to a corner in the garage until it physically dies, or (as I will start doing soon) gets cannibalized and consolidated into one of the bigger old boxes that still has space to accept another GPU or RAM or better CPU etc.

That said, to break all those BOINC records in all projects, I am looking forward to the upcoming 22-core Skylake XEONs and Pascal-based Nvidia GPUs. A dual CPU (88 threads!), 4-way SLI of Titan P(ascal), with plenty of RAM per thread (enabling "keep stuff in memory when suspended" without running out of RAM), using maintenance-free sealed hybrid liquid coolers with dedicated fans per CPU & GPU along with RAM fans and using only SSDs... Now, there is something to look forward to saving money for to replace my new home office PC with... :-)



____________

AyalaZero
Send message
Joined: 22 Nov 15
Posts: 8
Credit: 19,131,150
RAC: 0
Level
Pro
Scientific publications
watwat
Message 42291 - Posted: 3 Dec 2015 | 21:30:59 UTC - in response to Message 42290.

Tuna Ertemalp, take you're time... wait til the prices go down. We are in a recession.

Note: Take this advise with a grain of salt, I just enjoy being #1 for the time being. hahaha

Profile Tuna Ertemalp
Send message
Joined: 28 Mar 15
Posts: 46
Credit: 1,547,496,701
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 42296 - Posted: 5 Dec 2015 | 0:40:00 UTC - in response to Message 42291.

Tuna Ertemalp, take you're time...


LOL!

Well, truth be told, I already upgraded a machine with an 8800GT to a Titan X, made it churn only for GPUGrid for a week or two, about three months ago, and it got into top 10 within a week, I believe. Now it is churning for all 50 projects, so it slipped down.

In a day or two, another machine with GTX 480 will receive its Titan X Hybrid (I found it "open box" on Amazon with some screws missing, and ended up paying only $835 for it!), and I will let it churn exclusively for GPUGrid for a week or two, just to see what it does. Then, it'll go back to serving all projects.

Within the next 10 days, this 4-way Titan X will be completed, will be targeted at GPUGrid for a while, out of curiosity, before going back into the pool.

Lastly, the leftover 8800GT and GTX480 will join a GTX580 inside the machine replaced by this new 4-way, and head to the garage. I think I'll skip targeting it to GPUGrid.

This Skylake+Pascal dream machine is probably later part of 2016.

But you get the idea: I don't permanently target any machine to any individual project trying to get to the top of scoreboards. I used to believe in supporting just one project, SETI, when I started this 16 years ago. Then I added 4-5 more projects to my list around 2010. Earlier this summer I totally changed my philosophy; I now believe in spreading the wealth across all projects. So, anyone choosing to support only one or two projects will always be above me on scoreboards, no matter what I build.

So, you are safe... :-)

Tuna

Profile Tuna Ertemalp
Send message
Joined: 28 Mar 15
Posts: 46
Credit: 1,547,496,701
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 42297 - Posted: 5 Dec 2015 | 6:05:09 UTC - in response to Message 42296.
Last modified: 5 Dec 2015 | 6:08:06 UTC

In a day or two, another machine with GTX 480 will receive its Titan X Hybrid (I found it "open box" on Amazon with some screws missing, and ended up paying only $835 for it!), and I will let it churn exclusively for GPUGrid for a week or two, just to see what it does. Then, it'll go back to serving all projects.

This single eVGA Titan X Hybrid machine (1152/1241 Mhz) just got its first two tasks. But it is also doing CPU-only tasks from Collatz, Einstein, Poem, Rosetta and SETI to keep all CPU cores & GPU at 100% for 1.5 days while it gets tested for stability and the heating/cooling characteristics of the fans & radiators. Let's see what it does in the meantime although I expect some perf will be lost to task switching (is the BOINC manager smart enough to not switch away from the only GPU project to keep the GPU always utilized as opposed to suspend the GPU task to give the associated CPU core to another project?). It should be a bit faster than my other current non-hybrid single eVGA Titan X SC machine (1127/1216 Mhz) that had managed to get into top 10 when concentrating on GPUGrid back in Aug/Sep/Oct. That was the machine I had already upgraded:
Well, truth be told, I already upgraded a machine with an 8800GT to a Titan X, made it churn only for GPUGrid for a week or two, about three months ago, and it got into top 10 within a week, I believe. Now it is churning for all 50 projects, so it slipped down.

Tuna

Post to thread

Message boards : Number crunching : HELP: x16/x16/x16/x16 vs. x16/x8/x8/x8?

//