Apology rejected: you, Sir, have to take full responsibility for this fine joke!
BTW: Gary and archae, definitely choose Haswell over Ivy unless there's a significant markup on the newer parts (which shouldn't be there according to official prices). The performance per clock went up by almost 10% on average - which is surely not yet noticeable, but viewed the other way around this lets you achieve the same performance at 300 - 400 MHz less. Which is certainly a nice deal if you're not just looking at 4.5+ GHz clock speeds!
And I vote for using that iGPU, especially the slightly faster models in Haswell. I've got 2 Hd4000's running, both slightly overclocked and at significantly reduced voltage. In fact, the lower limit to my load voltage is actually what the GPU gets under idle conditions. So if I had more BIOS options I could run it even more efficiently.
As luck would have it, Nvidia today introduced their first Maxwell parts, and, unusually, their first offerings are centered directly on my interest area for this machine (i.e. 55 to 60 W, $120 to $170 price), rather than on the high end.
Despite the impediment that these are introduced without benefit from a TSMC process generation jump, the claim is that design changes have very greatly improved the power efficiency (compute/watt) and appreciably improved the area efficiency (compute/square centimeter) and thus compute/purchase cost. If these claims hold true for Einstein, I definitely want one of these cards. Power matters for me in this particular box more than usual, because the computer desk it will go into is a bit under strength for cooling, and directly warms the user (good in winter--not good in summer).
Of course, Einstein computation is nowhere near the center of their user scenarios, so it remains to be seen whether some of the structural changes which give them these advantages in their primary target markets happen by bad luck to cripple Einstein computation running the current code.
But I'd score it a plus if I helped uncover that fact, as lots more Maxwells are coming and not everybody is going to buy AMD (most particularly not the people with an interest in power consumption).
As it happens, I had ordered nearly all the parts needed for my last XP replacement, including a GTX 650 card, a couple of weeks ago and they are sitting in their boxes in my living room waiting for me to finish terminating the service of the XP predecessor and start assembling and commissioning the new machine.
After reading announcement day material, my current intention is to return the unopened 650 card for refund, and buy either a GTX750 or a GTX750 Ti. In this particular case, the Ti cards come at a very small increment in power consumption, and a pretty small increment in both performance and price compared to some previous 'TI' offsets.
If anyone here has any new Maxwell (or 750, or 750 Ti) insights, I'd be happy to hear them before I place my order.
I am currently running an AMD Radeon R9 270X (showing as a: CAL AMD Radeon HD 7850/7870 series (Pitcairn) (2048MB) driver: 1.4.1848) on a VM with Win 7. I crunch BRP 5 WUs in just under 5 hours. This card at a Best Buy is about $249 which violates your price point but I mention it because of my positive experience with this card.
I have usually always gone the NVIDIA route but the VM effort had me looking at Radeon because of their support for GPU pass-through in a virtual environment. I have a good opinion of the Radeon card that I have and would probably consider going Radeon on future purchases because of their support towards the open source community. NVIDIA seems resistant in this area.
That happens to be a review I saw yesterday. The higher power on the competitive Radeon cards is a dealbreaker for this particular build for me. Plenty of other users will be in a different situation.
The R7 265 uses almost a full Pitcairn GPU, which is quite a different beast than the much smaller GM107. The AMD is faster, for sure, but consumes almost double the power.
Generally Maxwell looks like an excellent improvement over Kepler for GP-GPU. How much this manifests at Einstein or other BOINC projects remains to be seen. Especially the high memory bandwidth requirement of Einstein could make things very interesting: the significantly larger L2 cache could help Maxwell enormously, or the 128 bit but could hamper GM107 performance. And how does it compare to current AMD offerings?
After this strong initial showing of Maxwell I wouldn't currently buy any other card until we know how good it really is for our favorite BOINC projects.
I picked up EVGA's high end 750 Ti (which uncharacteristically requires an extra 6-pin power connector) and installed it last night.
Here are some quick impressions with just a few results. I'm running 3 CPU jobs and 2 x BRP5 jobs right now. Time to complete 2 jobs ~14,777 seconds, so approximately 39K credits / day. It is drawing about 40..45 watts at the wall (via kill-a-watt meter) to run the 2 GPU jobs above and beyond the 110 watts to run the 3 CPU jobs.
MSI's Afterburner indicates 95% GPU utilization and a GPU load temp of 43 C degrees. The heat blowing out the GPU exhaust is very low compared to the usual superheated jet exhaust from my 7970's.
I picked up EVGA's high end 750 Ti (which uncharacteristically requires an extra 6-pin power connector) and installed it last night.
Here are some quick impressions with just a few results. I'm running 3 CPU jobs and 2 x BRP5 jobs right now. Time to complete 2 jobs ~14,777 seconds, so approximately 39K credits / day. It is drawing about 40..45 watts at the wall (via kill-a-watt meter) to run the 2 GPU jobs above and beyond the 110 watts to run the 3 CPU jobs.
MSI's Afterburner indicates 95% GPU utilization and a GPU load temp of 43 C degrees. The heat blowing out the GPU exhaust is very low compared to the usual superheated jet exhaust from my 7970's.
For a power and purchase cost conscious person, those sound like quite good results. It will be interesting to see by how much my dead stock EVGA 750 card underperforms it. I'm slightly concerned by the mere 1 GByte of memory, especially since Einstein just started up a new test with a 2 GB minimum.
But your results greatly reduce the chance that the current Einstein Perseus code is somehow gravely incompatible with Maxwell units.
Thanks for those numbers, Robert. How is the memory controller load? Have you already tried to push it a little further? The chip should easily take the maximum clock you can set via software (without BIOS mods), according to the launch reviews. The memory should also take about 6.0 GHz, give or take 200 MHz.
There is no punishment where
)
There is no punishment where no pun is meant.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: Apology rejected:
)
Apology rejected: you, Sir, have to take full responsibility for this fine joke!
BTW: Gary and archae, definitely choose Haswell over Ivy unless there's a significant markup on the newer parts (which shouldn't be there according to official prices). The performance per clock went up by almost 10% on average - which is surely not yet noticeable, but viewed the other way around this lets you achieve the same performance at 300 - 400 MHz less. Which is certainly a nice deal if you're not just looking at 4.5+ GHz clock speeds!
And I vote for using that iGPU, especially the slightly faster models in Haswell. I've got 2 Hd4000's running, both slightly overclocked and at significantly reduced voltage. In fact, the lower limit to my load voltage is actually what the GPU gets under idle conditions. So if I had more BIOS options I could run it even more efficiently.
MrS
Scanning for our furry friends since Jan 2002
As luck would have it, Nvidia
)
As luck would have it, Nvidia today introduced their first Maxwell parts, and, unusually, their first offerings are centered directly on my interest area for this machine (i.e. 55 to 60 W, $120 to $170 price), rather than on the high end.
Despite the impediment that these are introduced without benefit from a TSMC process generation jump, the claim is that design changes have very greatly improved the power efficiency (compute/watt) and appreciably improved the area efficiency (compute/square centimeter) and thus compute/purchase cost. If these claims hold true for Einstein, I definitely want one of these cards. Power matters for me in this particular box more than usual, because the computer desk it will go into is a bit under strength for cooling, and directly warms the user (good in winter--not good in summer).
Of course, Einstein computation is nowhere near the center of their user scenarios, so it remains to be seen whether some of the structural changes which give them these advantages in their primary target markets happen by bad luck to cripple Einstein computation running the current code.
But I'd score it a plus if I helped uncover that fact, as lots more Maxwells are coming and not everybody is going to buy AMD (most particularly not the people with an interest in power consumption).
As it happens, I had ordered nearly all the parts needed for my last XP replacement, including a GTX 650 card, a couple of weeks ago and they are sitting in their boxes in my living room waiting for me to finish terminating the service of the XP predecessor and start assembling and commissioning the new machine.
After reading announcement day material, my current intention is to return the unopened 650 card for refund, and buy either a GTX750 or a GTX750 Ti. In this particular case, the Ti cards come at a very small increment in power consumption, and a pretty small increment in both performance and price compared to some previous 'TI' offsets.
If anyone here has any new Maxwell (or 750, or 750 Ti) insights, I'd be happy to hear them before I place my order.
You might take a look at this
)
You might take a look at this comparison.
I am currently running an AMD Radeon R9 270X (showing as a: CAL AMD Radeon HD 7850/7870 series (Pitcairn) (2048MB) driver: 1.4.1848) on a VM with Win 7. I crunch BRP 5 WUs in just under 5 hours. This card at a Best Buy is about $249 which violates your price point but I mention it because of my positive experience with this card.
I have usually always gone the NVIDIA route but the VM effort had me looking at Radeon because of their support for GPU pass-through in a virtual environment. I have a good opinion of the Radeon card that I have and would probably consider going Radeon on future purchases because of their support towards the open source community. NVIDIA seems resistant in this area.
Just my two cents.
RE: You might take a look
)
That happens to be a review I saw yesterday. The higher power on the competitive Radeon cards is a dealbreaker for this particular build for me. Plenty of other users will be in a different situation.
The R7 265 uses almost a full
)
The R7 265 uses almost a full Pitcairn GPU, which is quite a different beast than the much smaller GM107. The AMD is faster, for sure, but consumes almost double the power.
Generally Maxwell looks like an excellent improvement over Kepler for GP-GPU. How much this manifests at Einstein or other BOINC projects remains to be seen. Especially the high memory bandwidth requirement of Einstein could make things very interesting: the significantly larger L2 cache could help Maxwell enormously, or the 128 bit but could hamper GM107 performance. And how does it compare to current AMD offerings?
After this strong initial showing of Maxwell I wouldn't currently buy any other card until we know how good it really is for our favorite BOINC projects.
MrS
Scanning for our furry friends since Jan 2002
I picked up EVGA's high end
)
I picked up EVGA's high end 750 Ti (which uncharacteristically requires an extra 6-pin power connector) and installed it last night.
Here are some quick impressions with just a few results. I'm running 3 CPU jobs and 2 x BRP5 jobs right now. Time to complete 2 jobs ~14,777 seconds, so approximately 39K credits / day. It is drawing about 40..45 watts at the wall (via kill-a-watt meter) to run the 2 GPU jobs above and beyond the 110 watts to run the 3 CPU jobs.
MSI's Afterburner indicates 95% GPU utilization and a GPU load temp of 43 C degrees. The heat blowing out the GPU exhaust is very low compared to the usual superheated jet exhaust from my 7970's.
Robert wrote:I picked up
)
For a power and purchase cost conscious person, those sound like quite good results. It will be interesting to see by how much my dead stock EVGA 750 card underperforms it. I'm slightly concerned by the mere 1 GByte of memory, especially since Einstein just started up a new test with a 2 GB minimum.
But your results greatly reduce the chance that the current Einstein Perseus code is somehow gravely incompatible with Maxwell units.
Thanks.
Thanks for those numbers,
)
Thanks for those numbers, Robert. How is the memory controller load? Have you already tried to push it a little further? The chip should easily take the maximum clock you can set via software (without BIOS mods), according to the launch reviews. The memory should also take about 6.0 GHz, give or take 200 MHz.
MrS
Scanning for our furry friends since Jan 2002
Gamespot has a nice review of
)
Gamespot has a nice review of the GTX 750 Ti.