Ivy Bridge Thread
 

I figure I'll get us started with some leaked Benchmark slides.

[url]http://lenzfire.com/wp-content/uploads/2011/11/26.jpg[/url]

^ That's 10-20% percent performance boost on top of an almost 20% power reduction :)

Full article, along with many more benchmark slides:
[url]http://www.zimbio.com/Laptop+Netbook/articles/Z_ypv4zT7eG/Official+Release+Date+Ivy+Bridge+vs+Sandy[/url]

Initial indications point to 2-3x increase of HD4000 graphics over HD3000.

[QUOTE=Dubslow;280602]
^ That's 10-20% percent performance boost on top of an almost 20% power reduction :)
[/QUOTE]

Is this a 10 to 20% boost in LL testing?

Hopefully; those numbers are fairly consistent across a variety of benchmark programs, so hopefully it means LL (FPU) improvement. I guess the only true answer is 'wait and see', but I would expect that we'll see that in LL testing. Also keep in mind that AVX, whenever George releases it, will be another ~10+% if memory serves.

[QUOTE=Dubslow;280624]Hopefully; those numbers are fairly consistent across a variety of benchmark programs, so hopefully it means LL (FPU) improvement. I guess the only true answer is 'wait and see', but I would expect that we'll see that in LL testing. Also keep in mind that AVX, whenever George releases it, will be another ~10+% if memory serves.[/QUOTE]

Sorry...AVX? I guess I am not very up to date!

AVX: Another instruction set extension on recent Intel hardware...I believe it stands for audio-video extensions...

[QUOTE=Christenson;280630]AVX: Another instruction set extension on recent Intel hardware...I believe it stands for audio-video extensions...[/QUOTE]

Nice!! I will look forward to this being released!

[url]http://en.wikipedia.org/wiki/Advanced_Vector_Extensions[/url]

Thanks, axn!

[QUOTE=kladner;280669]Thanks, axn![/QUOTE]

Ditto!

[QUOTE=axn;280633][url]http://en.wikipedia.org/wiki/Advanced_Vector_Extensions[/url][/QUOTE]

After reading this site I am still excited for this to come out; however, I am a little saddened by the fact that it will not apply to my computer. AVX2 looks like it will also provide a performance boost once it comes out in 2013.

[QUOTE=Primeinator;280704]After reading this site I am still excited for this to come out; however, I am a little saddened by the fact that it will not apply to my computer. [/QUOTE]
Ivy Bridge looks like the perfect upgrade for you!
[/fanboy]

[QUOTE=Dubslow;280738]Ivy Bridge looks like the perfect upgrade for you!
[/fanboy][/QUOTE]

Not on a college budget, it doesn't :jail:

It looks like Ivy Bridge will be up to 16% faster than Sandy Bridge at the same clock speed: [url]http://www.fudzilla.com/home/item/25988-ivy-bridge-core-i7-up-to-16-percent-faster-than-sandy-bridge[/url]

That appears to be more of a "higher turbo boost" than architectural.

Intel releases processors on a Tick-Tock schedule, where each new microarchitecture is followed by a new fabrication minimization and thus a reduction in die size for the same number of transistors.

SandyBridge was a new microarchitecture using the same 32nm process as the previous CPUs. Ivy Bridge will use a new 22nm process, so it won't be a new microarchitecture.

Where does the speed bump comes from if Ivy bridge is the same microarchitecture as SandyBridge?

By the way, I know the next microarchitecture, Haswell (on 22nm), will use 64KB L1 caches instead of 32KB. Will this help LL testing?

I would assume it's due to the new Tri-Gate transistors.

What do you mean speed bump? It is the same MA, yes, but they do add more fixes/optimizations/etc., and IPC can probably be expected to go up a few percent. The trigates at 22 nm means faster clock speeds for the same or less power, so increased speed there as well.

And yes, I imagine the increased cache size will help, though maybe not immensely. If it were data cache increase or L2/L3 increase, I think that'd help more. (I must disclaim that I am making an educated guess, and other more knowledgeable people should double check/correct me.)

[QUOTE=emily;289955]SandyBridge was a new microarchitecture using the same 32nm process as the previous CPUs. Ivy Bridge will use a new 22nm process, so it won't be a new microarchitecture.

Where does the speed bump comes from if Ivy bridge is the same microarchitecture as SandyBridge?[/QUOTE]
Even though the microarchitecture doesn't thoroughly change, Intel tunes a few things, hence the speedup.

[url]http://www.anandtech.com/show/5626/ivy-bridge-preview-core-i7-3770k/[/url]

It's a few weeks old, but still pretty cool.

[quote]- FP/integer divider delivers 2x throughput compared to Sandy Bridge
- MOV instructions no longer occupy an execution port, potential for improved ILP when MOVs are present[/quote]
^ -- George -- what sort of increase are we looking at there? (Are there a lot of divs in FFTs?)
[quote]Max supported DDR3 frequency is now 2800MHz (up from 2133MHz), memory speed can be moved in 200MHz increments[/quote]
I'm not sure if the "standard" clock of 1333 MHz will go up (e.g. less premiums on the faster mem, but that's more a mem-ufacture decision rather than Intel's), but it's still nice to see, especially with the memory bottlenecks introduced by AVX.

Release date appears to be late April to early June.
[url]http://hexus.net/tech/news/cpu/37129-new-intel-ivy-bridge-release-dates-revealed/[/url]
[url]http://www.cpu-world.com/news_2012/2012032702_Revised_Ivy_Bridge_launch_schedule_confirmed.html[/url]
[url]http://us.generation-nt.com/intel-ivy-bridge-processors-cpu-release-details-models-news-3465751.html[/url]
(^ largely duplicate)

[QUOTE=Dubslow;295024]
^ -- George -- what sort of increase are we looking at there?.[/QUOTE]

0%

[URL="http://www.anandtech.com/show/5728/intel-z77-panther-point-chipset-and-motherboard-preview-asrock-asus-gigabyte-msi-ecs-and-biostar"]Intel Z77 Panther Point Chipset and Motherboard Preview – ASRock, ASUS, Gigabyte, MSI, ECS and Biostar[/URL]
[quote]Well, today is a day that everyone was expecting but not the most exciting day of the month. Today, Intel officially releases the spate of 7-series chipsets for their 3rd generation Core microarchitecture processors (read—Ivy Bridge), despite the actual release of the processors being another three weeks away. This means that today we can actually look at, test, or purchase the range of motherboards that natively support Ivy Bridge. However, we can’t officially publish all the benchmarks for these products using Ivy Bridge until that date three weeks away (to be honest, we are still testing!). So today we are going to have a good in depth look into the Z77 chipset itself, and the review products we have received to let you know what we think looks good. All these boards today will be fully reviewed, warts and all, with Ivy Bridge, as close to official release as possible.[/quote]
[quote]For the high-end power user, the chipset is now geared to handle memory that is more powerful. Again, thanks to the improvement on the new processors, these combined systems should be able to handle (with ease) DDR3-2800 memory. The memory dividers work similar to the gear ratios in Sandy Bridge-E, with users able to call a larger array of memory dividers than before with up to 200 MHz increments. This is of great benefit to the integrated graphics on the new generation processors, which should benefit from the purchase of a higher end memory kit to provide enough bandwidth.[/quote]
[quote]If you happen to purchase ASUS for Ivy Bridge, there is also a little treat in store, as they have reworked the memory sub-system. [U]Their new method stunned Intel engineers[/U], but should provide distinct memory speed advantages. Simply put, instead of memory banks being read consecutively, the memory is read in parallel. We are awaiting more detail regarding how this feature works.[/quote](my own emphasis)

Page 2 is not really related to this forum, but is [i]incredibly[/i] interesting with regards to GPU fps vs. screen refresh rate, and synchronizing them.

From the end of the article, we have this expansion on ASUS/memory:
[quote]ASUS have outsmarted Intel and have decided to take their technology to another level. This is specifically in terms of memory, and how it is routed through the motherboard. Typically, routing through the memory would occur in a daisy chain type environment, whereby if data was in the furthest memory slot away from the board, it would take longer to get to the CPU, and perhaps cause synchronization issues and delays—all reads had to be done serially between sticks in the same channel.

With ASUS' new technology, they are essentially parallelizing memory reads that are commonly done serially between memory banks. This is part of their 'T-Topology' memory subsystem, which allows synchronization to be dealt with in hardware. This, according to ASUS, should allow for up to a 15% memory overclock beyond the previous methodology, where the motherboard is the limiting factor. In this circumstance, we could be seeing some new memory records being set in dual channel memory.

I have probed ASUS for specific details on how this works, and I am awaiting a response. I hope that by the time we are allowed to release our Ivy Bridge results on Z77 that I will have something more technical to pass on to you about how this works.[/quote]

Chirp... chirp...
 

Release for the mobile processors may be as early as next week (April 23rd). No word on the desktop parts.

[url]http://www.digitimes.com/news/a20120411PD216.html[/url]
[url]http://arstechnica.com/gadgets/news/2012/04/intel-may-be-pushing-up-ivy-bridge-launch-to-boost-next-gen-ultrabooks.ars[/url]

[quote=Ars Technica]Intel is believed to be pushing up the launch of its first wave of Ivy Bridge processors by one week to April 23. According to sources for DigiTimes, Intel will announce availability of its next-generation 22nm processors on that day in order to pump up demand for new ultrabook models from ASUS, Acer, Lenovo, and HP.

Intel suggested in February that the highly anticipated Ivy Bridge processors, which are built using 3D tri-gate transistors on a new 22nm process, would be delayed until June. However, leaked information about the launch that surfaced just two weeks ago suggested Intel would instead roll out various processors on a staggered release schedule, with some processors becoming available at the end of April. The rest would start shipping in quantity just over a month later in early June.[/quote]

Woohoo, my computer is slowly dying so it will need replacing in the near future. Hopefully the desktop parts will be out by then.

Some overclocking info about Ivy Bridge:
[url]http://www.anandtech.com/show/5763/undervolting-and-overclocking-on-ivy-bridge[/url]

A couple of reviews:
[url]http://www.anandtech.com/show/5771/the-intel-ivy-bridge-core-i7-3770k-review[/url]
and
[url]http://www.techspot.com/review/523-ivy-bridge-intel-core-i7-3770k/[/url]

Ivy Bridge has been officially announced today. Can't wait to see the benchmarks!

[QUOTE=ixfd64;297132]Ivy Bridge has been officially announced today. Can't wait to see the benchmarks![/QUOTE]

Umm... go see the reviews that Jeff posted?

Some more links:
[url]http://semiaccurate.com/2012/04/23/intel-launches-ivy-bridge-amid-crushing-marketing-buzzwords/[/url]
[quote]That last one is Intel Insider, and it is a horrendous step backward for Ivy Bridge, and Sandy Bridge before it. Basically it spends power and time to encrypt everything on the system buses. It is useless work, lessened battery life, and the only reason it is there is to placate the content MAFIAA.
To make matters worse, Intel Insider is not user controllable, it is only controllable by unnamed remote 3rd parties who can now do things to your system that Intel won’t list. Seriously, think about the security implications, you are giving an unknown list of entities that are proven to be hostile to users the right to silently deny you use of your computer. They can potentially put things on your PC, take things off, and do so in a way that you can’t control, avoid, or worst of all detect. This ‘technology’ is actively harmful to the owner, and enough of an issue that I suggest that you avoid Ivy Bridge until it is not just fully documented, but user controllable. Scary on a whole new level.[/quote]
One thing I haven't been able to find info about is "Intel OS Guard". Any ideas there? It sounds bad.
Edit: Here are two particularly useless quotes. No actual info yet.
[quote]New security features include Intel OS Guard, which prevents malware attacks on Windows 8 and some Linux distributions[/quote]
^ [url]http://www.zdnet.com/blog/computers/intel-announces-ivy-bridge-chips-for-high-end-laptops-and-desktops/7944[/url]
[QUOTE]Intel OS Guard helps defend against privilege escalation attacks where a hacker remotely takes over another person's system.[/QUOTE]
^ [url]http://newsroom.intel.com/community/intel_newsroom/blog/2012/04/23/3rd-generation-intel-core-processors-bring-exciting-new-experiences-and-fun-to-the-pc[/url]
(The second quote is clearly meant for the clueless masses.)

It seems that if you wanna go for "maximum kicks in the butt per ingested watt" the ivy-3770K would serve you better, but if you are targeting the performance (and don't mind about cost and killed watts) then sbe-3960X is still the king (I have yet to put my hand on one of those, never touched one, I tested a sbe-3820 some time ago but the performance was lousy compared to sb-2600k), and if you want to get the best price/performance, then sb-2600k is still the best by far (eventually a 2700k, but the only difference to 2600k is the clock a bit higher, for a higher price).

Per number of cores, like either buying 3 times ivy-3770k or buying 2 times sbe-3960X (to get the same number of 12 cores, as you can't really compare a 4-core with a 6-core), then you get a better deal with ivy, considering you would have 3 systems (three times the bus width, plus: if one crashed you still have 66% of the farm producing, contrary to 50% for sbe, etc) for about half of the price (the 3960x is still bloody expensive!) and about 70% of power consumption.

If you don't need a compute right now, in this very right moment, then I would recommend to wait for a while, few months. Some 6 cores or 8 cores ivy stuff should pop-up under (or around) 500 bucks. Then we could really compare, i.e. we would have the same number of cores as a 3960X, or more, maybe the same bus, for half of the price and 60% power (heating elements).

edit: and there is no hardware virtualization for 3770k???? Grrrrr....

[QUOTE=LaurV;297152]if you want to get the best price/performance, then sb-2600k is still the best by far (eventually a 2700k, but the only difference to 2600k is the clock a bit higher, for a higher price). [/QUOTE]

I'm not sure how you got there, but the 3770K is around the same price as the 2600K, and less than a 2700K, while getting slightly more performance. The 3770K is king in both performance/watt and price/performance.

Memory scales nicely -- frequency is directly correlated with bandwidth.
[url]http://www.tomshardware.com/reviews/ivy-bridge-benchmark-core-i7-3770k,3181-10.html[/url]
[quote]A synthetic like Sandra 2012 demonstrates sizable gains. Bandwidth literally doubles as we move from two channels of DDR3-1066 to DDR3-2133. <snip> Real-world performance improvements trail off a lot faster though, likely because memory isn’t the most debilitating bottleneck.[/quote]
Since of course for v27+/AVX memory [i]is[/i] the most debilitating bottleneck, we might see some returns even moving from 1866->2133. Has any one actually managed to acquire one, or are we all out of money? :razz:

For Bdot: [quote]HD Graphics 4000, on the other hand, supports FP32/FP64 under DirectCompute and FP32 in OpenCL. Intel currently lacks Khronos certification for ARB_gpu_shader_fp64, so it’s not enabled.[/quote][url]http://www.tomshardware.com/reviews/ivy-bridge-benchmark-core-i7-3770k,3181-6.html[/url]
However, in the AnandTech review that Jeff posted, I got the impression that OpenCL 1.1 wasn't initially supported.

[QUOTE=Dubslow;297137]One thing I haven't been able to find info about is "Intel OS Guard". Any ideas there? It sounds bad.[/QUOTE]
It's SMEP (Supervisor Mode Execution Protection) in disguise :smile:

[QUOTE=Ralf Recker;297330]It's SMEP (Supervisor Mode Execution Protection) in disguise :smile:[/QUOTE]

Cool! Thanks. A quick Google brings up [url]http://vulnfactory.org/blog/2011/06/05/smep-what-is-it-and-how-to-beat-it-on-linux/[/url], by way of [url]http://forums.overclockers.com.au/showthread.php?t=964305[/url].

Ok,
The integrated GPU improvements are less important on desktop system. (Add card instead)
The Ivy Bridge is the last LGA 1155 socket processor
The power improvements of Ivy Bridge are less important on the desktop.

It sounds to me like Ivy Bridge is better as a mobile upgrade and that a large desktop investment should wait for the LGA 1150 socket.

I just bought an i5-3570k on sale at Microcenter today. Put together the cheapest build I could (under $400--plus some parts I had sitting around) I don't know what numbers you guys actually want to see.

It's doing a double check @ .010 seconds per iteration.
and running a couple P1's
and an LL of a 56 million number @ .021 seconds per iteration.

temps are stable with intel's stock air cooler. I played with the Gigabyte auto-overclocker application a bit, but it didn't seem to improve times much.

[QUOTE=chappy;298584]I just bought an i5-3570k on sale at Microcenter today. Put together the cheapest build I could (under $400--plus some parts I had sitting around) I don't know what numbers you guys actually want to see.

It's doing a double check @ .010 seconds per iteration.
and running a couple P1's
and an LL of a 56 million number @ .021 seconds per iteration.

temps are stable with intel's stock air cooler. I played with the Gigabyte auto-overclocker application a bit, but it didn't seem to improve times much.[/QUOTE]

Those are pretty good timings.

[QUOTE=chappy;298584]I just bought an i5-3570k on sale at Microcenter today. Put together the cheapest build I could (under $400--plus some parts I had sitting around) I don't know what numbers you guys actually want to see.

It's doing a double check @ .010 seconds per iteration.
and running a couple P1's
and an LL of a 56 million number @ .021 seconds per iteration.

temps are stable with intel's stock air cooler. I played with the Gigabyte auto-overclocker application a bit, but it didn't seem to improve times much.[/QUOTE]
Those timings are about what I get with a i7-2600K slightly OCd to ~3.8-3.9GHz, so it does seem there is a slight IPC improvement, but not much.

[QUOTE=chappy;298584]I just bought an i5-3570k on sale at Microcenter today. Put together the cheapest build I could (under $400--plus some parts I had sitting around) I don't know what numbers you guys actually want to see.

It's doing a double check @ .010 seconds per iteration.
and running a couple P1's
and an LL of a 56 million number @ .021 seconds per iteration.

temps are stable with intel's stock air cooler. I played with the Gigabyte auto-overclocker application a bit, but it didn't seem to improve times much.[/QUOTE]
A full benchmark with 27.7 would be nice. It would be nice to compare it with a sandy bridge of the same clockspeed.

i5 3570K
 

I hope that tomorrow my i5 3570K will be delivered. I will use a Corsair H100 water cooling and will post my results here. Additionally, I bought new Corsair DDR3-2000 MHz CL9 RAM as Prime95 version 27 is memory bound.

some 3570K numbers
 

Intel(R) Core(TM) i5-3570K CPU @ 3.40GHz
CPU speed: 3888.62 MHz, 4 cores
CPU features: Prefetch, SSE, SSE2, SSE4, AVX
L1 cache size: 32 KB
L2 cache size: 256 KB, L3 cache size: 6 MB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
TLBS: 64
Prime95 64-bit version 27.7, RdtscTiming=1
Best time for 4K FFT length: 0.015 ms., avg: 0.015 ms.
Best time for 5K FFT length: 0.020 ms., avg: 0.020 ms.

Best time for 3200K FFT length: 18.648 ms., avg: 19.190 ms.
Best time for 3360K FFT length: 20.393 ms., avg: 20.476 ms.
Best time for 3456K FFT length: 20.879 ms., avg: 20.960 ms.
Best time for 3584K FFT length: 21.729 ms., avg: 21.912 ms.
Best time for 3840K FFT length: 22.849 ms., avg: 22.978 ms.
Best time for 4000K FFT length: 24.142 ms., avg: 24.167 ms.
Best time for 4096K FFT length: 24.493 ms., avg: 24.603 ms.
Best time for 4480K FFT length: 27.407 ms., avg: 27.575 ms.
Best time for 4608K FFT length: 28.201 ms., avg: 28.321 ms.
Best time for 4800K FFT length: 28.919 ms., avg: 28.993 ms.

Best time for 4608K all-complex FFT length: 28.458 ms., avg: 28.502 ms.
Best time for 4800K all-complex FFT length: 29.086 ms., avg: 29.145 ms.
Best time for 5120K all-complex FFT length: 31.960 ms., avg: 32.017 ms.
Best time for 5760K all-complex FFT length: 36.246 ms., avg: 36.445 ms.
Best time for 6144K all-complex FFT length: 38.291 ms., avg: 38.506 ms.
Best time for 6400K all-complex FFT length: 40.275 ms., avg: 40.331 ms.
Best time for 6912K all-complex FFT length: 45.131 ms., avg: 45.344 ms.
Best time for 7680K all-complex FFT length: 49.171 ms., avg: 49.396 ms.
Best time for 8000K all-complex FFT length: 52.381 ms., avg: 52.618 ms.

Best time for 23040K all-complex FFT length: 179.803 ms., avg: 180.391 ms.
Best time for 25600K all-complex FFT length: 184.759 ms., avg: 185.269 ms.
Best time for 30720K all-complex FFT length: 241.200 ms., avg: 241.839 ms.
Best time for 32000K all-complex FFT length: 243.634 ms., avg: 244.357 ms.
Best time for 32768K all-complex FFT length: 262.301 ms., avg: 263.185 ms.

Timing FFTs using 2 threads.
Best time for 4K FFT length: 0.015 ms., avg: 0.015 ms.
Best time for 5K FFT length: 0.020 ms., avg: 0.020 ms.
Best time for 6K FFT length: 0.025 ms., avg: 0.034 ms.
Best time for 8K FFT length: 0.028 ms., avg: 0.031 ms.
Best time for 3360K FFT length: 10.451 ms., avg: 11.189 ms.
Best time for 3456K FFT length: 10.756 ms., avg: 10.787 ms.
Best time for 3584K FFT length: 11.166 ms., avg: 11.848 ms.
Best time for 3840K FFT length: 11.720 ms., avg: 11.769 ms.
Best time for 4000K FFT length: 12.414 ms., avg: 13.174 ms.
Best time for 4096K FFT length: 12.522 ms., avg: 12.621 ms.
Best time for 4480K FFT length: 14.083 ms., avg: 14.232 ms.
Best time for 4608K FFT length: 14.476 ms., avg: 14.525 ms.
Best time for 4800K FFT length: 14.892 ms., avg: 14.959 ms.

Best time for 23040K FFT length: 89.475 ms., avg: 89.791 ms.
Best time for 25600K FFT length: 93.668 ms., avg: 94.010 ms.
Best time for 26880K FFT length: 104.536 ms., avg: 104.938 ms.
Best time for 28672K FFT length: 113.834 ms., avg: 114.190 ms.
Best time for 30720K FFT length: 119.908 ms., avg: 120.381 ms.
Best time for 32000K FFT length: 122.827 ms., avg: 123.289 ms.
Best time for 32768K FFT length: 130.638 ms., avg: 131.371 ms.

Best time for 4K all-complex FFT length: 0.014 ms., avg: 0.014 ms.
Best time for 5K all-complex FFT length: 0.019 ms., avg: 0.019 ms.
Best time for 6K all-complex FFT length: 0.024 ms., avg: 0.024 ms.
Best time for 8K all-complex FFT length: 0.033 ms., avg: 0.033 ms.

Best time for 3840K all-complex FFT length: 11.698 ms., avg: 11.760 ms.
Best time for 4000K all-complex FFT length: 12.451 ms., avg: 12.772 ms.
Best time for 4096K all-complex FFT length: 12.730 ms., avg: 12.790 ms.
Best time for 4608K all-complex FFT length: 14.561 ms., avg: 14.829 ms.
Best time for 4800K all-complex FFT length: 14.845 ms., avg: 14.939 ms.

Best time for 30720K all-complex FFT length: 121.944 ms., avg: 122.321 ms.
Best time for 32000K all-complex FFT length: 123.451 ms., avg: 123.814 ms.
Best time for 32768K all-complex FFT length: 133.054 ms., avg: 133.378 ms.

Timing FFTs using 3 threads.

Best time for 4K FFT length: 0.015 ms., avg: 0.015 ms.
Best time for 5K FFT length: 0.020 ms., avg: 0.020 ms.
Best time for 6K FFT length: 0.028 ms., avg: 0.033 ms.

Best time for 3456K FFT length: 7.527 ms., avg: 7.605 ms.
Best time for 3584K FFT length: 7.941 ms., avg: 8.213 ms.
Best time for 3840K FFT length: 8.219 ms., avg: 8.291 ms.
Best time for 4000K FFT length: 8.908 ms., avg: 9.793 ms.
Best time for 7680K FFT length: 17.306 ms., avg: 17.408 ms.
Best time for 8000K FFT length: 18.455 ms., avg: 18.523 ms.

Best time for 25600K FFT length: 65.096 ms., avg: 65.394 ms.
Best time for 26880K FFT length: 71.275 ms., avg: 71.601 ms.
Best time for 28672K FFT length: 78.050 ms., avg: 78.405 ms.
Best time for 30720K FFT length: 82.601 ms., avg: 82.891 ms.
Best time for 32000K FFT length: 84.828 ms., avg: 85.411 ms.
Best time for 32768K FFT length: 89.833 ms., avg: 90.240 ms.

Best time for 4K all-complex FFT length: 0.014 ms., avg: 0.014 ms.
Best time for 5K all-complex FFT length: 0.019 ms., avg: 0.019 ms.

Best time for 3200K all-complex FFT length: 6.769 ms., avg: 6.888 ms.
Best time for 3456K all-complex FFT length: 7.448 ms., avg: 7.518 ms.
Best time for 3840K all-complex FFT length: 8.304 ms., avg: 8.341 ms.
Best time for 4000K all-complex FFT length: 8.589 ms., avg: 8.844 ms.
Best time for 4096K all-complex FFT length: 8.912 ms., avg: 9.258 ms.
Best time for 4608K all-complex FFT length: 10.066 ms., avg: 10.145 ms.
Best time for 4800K all-complex FFT length: 10.541 ms., avg: 10.631 ms.
Best time for 5120K all-complex FFT length: 11.366 ms., avg: 11.519 ms.
Best time for 5760K all-complex FFT length: 12.719 ms., avg: 12.766 ms.
Best time for 6144K all-complex FFT length: 13.715 ms., avg: 13.749 ms.
Best time for 6400K all-complex FFT length: 14.198 ms., avg: 14.469 ms.
Best time for 6912K all-complex FFT length: 15.871 ms., avg: 15.998 ms.
Best time for 7680K all-complex FFT length: 17.278 ms., avg: 17.516 ms.
Best time for 8000K all-complex FFT length: 18.400 ms., avg: 18.467 ms.

Best time for 32000K all-complex FFT length: 85.093 ms., avg: 85.458 ms.
Best time for 32768K all-complex FFT length: 90.998 ms., avg: 91.616 ms.

Timing FFTs using 4 threads.
Best time for 4K FFT length: 0.015 ms., avg: 0.015 ms.
Best time for 5K FFT length: 0.020 ms., avg: 0.020 ms.

Best time for 3200K FFT length: 5.794 ms., avg: 5.905 ms.
Best time for 3360K FFT length: 6.081 ms., avg: 6.138 ms.
Best time for 3456K FFT length: 6.237 ms., avg: 6.308 ms.
Best time for 3584K FFT length: 6.744 ms., avg: 6.870 ms.
Best time for 3840K FFT length: 6.835 ms., avg: 6.882 ms.
Best time for 4000K FFT length: 7.631 ms., avg: 8.146 ms.
Best time for 4096K FFT length: 7.579 ms., avg: 7.643 ms.
Best time for 4480K FFT length: 8.274 ms., avg: 8.855 ms.
Best time for 4608K FFT length: 8.427 ms., avg: 8.471 ms.
Best time for 4800K FFT length: 9.101 ms., avg: 9.183 ms.

Best time for 8000K FFT length: 15.238 ms., avg: 15.285 ms.
Best time for 28672K FFT length: 62.049 ms., avg: 62.554 ms.
Best time for 30720K FFT length: 65.657 ms., avg: 66.615 ms.
Best time for 32000K FFT length: 69.140 ms., avg: 70.221 ms.
Best time for 32768K FFT length: 72.525 ms., avg: 73.595 ms.

Best time for 4K all-complex FFT length: 0.014 ms., avg: 0.014 ms.
Best time for 5K all-complex FFT length: 0.019 ms., avg: 0.019 ms.

Best time for 3200K all-complex FFT length: 5.636 ms., avg: 5.820 ms.
Best time for 3456K all-complex FFT length: 6.042 ms., avg: 6.124 ms.
Best time for 3840K all-complex FFT length: 6.848 ms., avg: 6.983 ms.
Best time for 4000K all-complex FFT length: 7.408 ms., avg: 7.929 ms.
Best time for 4096K all-complex FFT length: 7.490 ms., avg: 8.052 ms.
Best time for 4608K all-complex FFT length: 8.232 ms., avg: 8.701 ms.
Best time for 4800K all-complex FFT length: 8.891 ms., avg: 9.006 ms.
Best time for 5120K all-complex FFT length: 9.488 ms., avg: 9.615 ms.
Best time for 8000K all-complex FFT length: 14.928 ms., avg: 15.083 ms.

Best time for 30720K all-complex FFT length: 65.655 ms., avg: 66.737 ms.
Best time for 32000K all-complex FFT length: 68.448 ms., avg: 69.153 ms.
Best time for 32768K all-complex FFT length: 71.936 ms., avg: 73.224 ms.

Best time for 61 bit trial factors: 1.814 ms.
Best time for 62 bit trial factors: 1.852 ms.
Best time for 63 bit trial factors: 2.092 ms.
Best time for 64 bit trial factors: 2.112 ms.
Best time for 65 bit trial factors: 2.495 ms.
Best time for 66 bit trial factors: 2.911 ms.
Best time for 67 bit trial factors: 2.880 ms.
Best time for 75 bit trial factors: 2.802 ms.
Best time for 76 bit trial factors: 2.798 ms.
Best time for 77 bit trial factors: 2.799 ms.

Those are good times for a 3.88GHz! I got approximately the same for a sbe (i7-2600k) OC'ed around 4GHz too. For higher FFTs the memory (speed/channels) also come in play.

I'm looking at putting together a new Ivy Bridge system with mild overclock, nothing too crazy on air.

What are decent cases these days, power supplies, etc...

I'm interested in one of the new Asus Z77 motherboards since I heard that they did some interesting work with the memory subsystems to increase performance. What speed of DDR3 should I be looking at now, I think Z77 can support up to DDR3 1866? Is Corsair still a good RAM manufacturer?

I'm not trying to build the cheapest possible system, but want good quality parts.

[QUOTE=Jeff Gilchrist;298682]I'm looking at putting together a new Ivy Bridge system with mild overclock, nothing too crazy on air.

What are decent cases these days, power supplies, etc...

I'm interested in one of the new Asus Z77 motherboards since I heard that they did some interesting work with the memory subsystems to increase performance. What speed of DDR3 should I be looking at now, I think Z77 can support up to DDR3 1866? Is Corsair still a good RAM manufacturer?

I'm not trying to build the cheapest possible system, but want good quality parts.[/QUOTE]

The RAM goes passed 2133 if I recall correctly, though over course anything over 1600 is technically overclocking (still better than 1333 for SB). I remember hearing about the ASUS memory thing, but haven't heard anything since the initial review (haven't really looked either to be honest). As of last year Corsair and GSKILL were probably the biggest names, followed by Kingston. I'm sure the others are mostly good; have my sticks are Mushkin, but that's at 1333. For awesome/cheap air cooler, the Hyper 212+ is always a gem. It costs $25 and got 25C lower temps than stock/piece-of-junk heatsink. (Still reporting for SB of course.)

Some Z77 boards supports SSD-Caching, the only thing i know which could be important for some things

perhaps these RAM-Modules could be interesting for you

[URL]http://gskill.com/products.php?index=463[/URL]

high frequency + better timings + 1.5V support

[URL]http://www.bequiet.com/de/powersupply[/URL] (Made in Germany without "Bang" included [URL]http://www.pcgameshardware.de/aid,878970/China-Boeller-Hantol-Silent-700-Watt-mit-Fake-PFC-aus-Zement/Netzteil/News/[/URL] )

[URL]http://www.coolermaster-usa.com/category.php?category_by=0&category_id=19&category_name=Mid Tower[/URL]

enclosures without much "bling bling"

found something interesting about the memory on z77-boards of asus

[URL]http://www.youtube.com/watch?v=TOrUyVbNDO4&feature=youtu.be[/URL]

[QUOTE=Redarm;298699]found something interesting about the memory on z77-boards of asus

[URL]http://www.youtube.com/watch?v=TOrUyVbNDO4&feature=youtu.be[/URL][/QUOTE]
That's a darned nice video -- adjustable desktop TDP for the win (not-GIMPS-withstanding :razz:). I haven't even got to the memory part yet, but ASUS is looking pretty smart :smile:

Edit: Memory is at 28:45.

[QUOTE=Jeff Gilchrist;298682]What are decent cases these days, power supplies, etc...[/QUOTE]

Take a look at Antec Titan 650, it is marketed as a server case but it really is a workstation case. It lacks some server features (intrusion detection) and has audio outs, a workstation feature.

Anyway, this case has plenty of room, and it comes with a decent 650W Power Supply.

I have recently added a [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16811147107"]BlackHawk[/URL]; it is kind of a Newegg's own [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16811119160"]HAF 932[/URL] knockoff, but it works and vents well, can't complain. Depends on your demands.

[QUOTE=Dubslow;298701]That's a darned nice video -- adjustable desktop TDP for the win (not-GIMPS-withstanding :razz:). I haven't even got to the memory part yet, but ASUS is looking pretty smart :smile:

Edit: Memory is at 28:45.[/QUOTE]

Nice video. So 4 x 4GB DIMMs at 2133 MHz is speed spot apparently with the ASUS boards. Hmmm Sabertooth Z77 or Maximus V Gene...

Benchmark Data for overclocked Intel Core i5 3570K
 

[QUOTE=Brain;298609]I hope that tomorrow my i5 3570K will be delivered. I will use a Corsair H100 water cooling and will post my results here. Additionally, I bought new Corsair DDR3-2000 MHz CL9 RAM as Prime95 version 27 is memory bound.[/QUOTE]
[CODE]Intel(R) Core(TM) i5-3570K CPU @ 3.40GHz
CPU speed: 4296.40 MHz, 4 cores
CPU features: Prefetch, SSE, SSE2, SSE4, AVX
L1 cache size: 32 KB
L2 cache size: 256 KB, L3 cache size: 6 MB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
TLBS: 64
Prime95 64-bit version 27.7, RdtscTiming=1
Best time for 768K FFT length: 3.673 ms., avg: 3.728 ms.
Best time for 896K FFT length: 4.458 ms., avg: 6.621 ms.
Best time for 1024K FFT length: 4.963 ms., avg: 4.997 ms.
Best time for 1280K FFT length: 6.422 ms., avg: 6.503 ms.
Best time for 1536K FFT length: 7.755 ms., avg: 7.824 ms.
Best time for 1792K FFT length: 9.254 ms., avg: 9.275 ms.
Best time for 2048K FFT length: 10.360 ms., avg: 10.781 ms.
Best time for 2560K FFT length: 13.161 ms., avg: 13.201 ms.
Best time for 3072K FFT length: 16.373 ms., avg: 16.430 ms.
Best time for 3584K FFT length: 19.694 ms., avg: 26.449 ms.
Best time for 4096K FFT length: 22.105 ms., avg: 33.370 ms.
Best time for 5120K FFT length: 28.694 ms., avg: 30.367 ms.
Best time for 6144K FFT length: 34.559 ms., avg: 34.632 ms.
Best time for 7168K FFT length: 41.648 ms., avg: 72.306 ms.
Best time for 8192K FFT length: 48.394 ms., avg: 49.078 ms.
Timing FFTs using 2 threads.
Best time for 768K FFT length: 1.957 ms., avg: 2.198 ms.
Best time for 896K FFT length: 2.291 ms., avg: 2.330 ms.
Best time for 1024K FFT length: 2.578 ms., avg: 4.825 ms.
Best time for 1280K FFT length: 3.305 ms., avg: 5.679 ms.
Best time for 1536K FFT length: 4.023 ms., avg: 4.088 ms.
Best time for 1792K FFT length: 4.778 ms., avg: 4.890 ms.
Best time for 2048K FFT length: 5.353 ms., avg: 5.384 ms.
Best time for 2560K FFT length: 6.804 ms., avg: 7.721 ms.
Best time for 3072K FFT length: 8.424 ms., avg: 8.509 ms.
Best time for 3584K FFT length: 10.124 ms., avg: 10.195 ms.
Best time for 4096K FFT length: 11.378 ms., avg: 11.699 ms.
Best time for 5120K FFT length: 14.853 ms., avg: 15.435 ms.
Best time for 6144K FFT length: 17.758 ms., avg: 18.194 ms.
Best time for 7168K FFT length: 21.321 ms., avg: 21.479 ms.
Best time for 8192K FFT length: 24.884 ms., avg: 24.993 ms.
Timing FFTs using 3 threads.
Best time for 768K FFT length: 1.341 ms., avg: 2.218 ms.
Best time for 896K FFT length: 1.588 ms., avg: 1.692 ms.
Best time for 1024K FFT length: 1.798 ms., avg: 1.824 ms.
Best time for 1280K FFT length: 2.315 ms., avg: 2.359 ms.
Best time for 1536K FFT length: 2.825 ms., avg: 3.133 ms.
Best time for 1792K FFT length: 3.391 ms., avg: 5.513 ms.
Best time for 2048K FFT length: 3.773 ms., avg: 3.795 ms.
Best time for 2560K FFT length: 4.842 ms., avg: 4.936 ms.
Best time for 3072K FFT length: 5.928 ms., avg: 7.048 ms.
Best time for 3584K FFT length: 7.206 ms., avg: 10.469 ms.
Best time for 4096K FFT length: 8.075 ms., avg: 8.468 ms.
Best time for 5120K FFT length: 10.483 ms., avg: 14.638 ms.
Best time for 6144K FFT length: 12.602 ms., avg: 12.688 ms.
Best time for 7168K FFT length: 14.868 ms., avg: 15.045 ms.
Best time for 8192K FFT length: 17.343 ms., avg: 18.094 ms.
Timing FFTs using 4 threads.
Best time for 768K FFT length: 1.056 ms., avg: 1.462 ms.
Best time for 896K FFT length: 1.269 ms., avg: 1.444 ms.
Best time for 1024K FFT length: 1.459 ms., avg: 2.345 ms.
Best time for 1280K FFT length: 1.936 ms., avg: 2.019 ms.
Best time for 1536K FFT length: 2.371 ms., avg: 2.569 ms.
Best time for 1792K FFT length: 2.814 ms., avg: 2.934 ms.
Best time for 2048K FFT length: 3.219 ms., avg: 3.687 ms.
Best time for 2560K FFT length: 4.120 ms., avg: 4.891 ms.
Best time for 3072K FFT length: 5.065 ms., avg: 5.304 ms.
Best time for 3584K FFT length: 6.148 ms., avg: 7.262 ms.
Best time for 4096K FFT length: 6.971 ms., avg: 7.386 ms.
Best time for 5120K FFT length: 11.012 ms., avg: 13.474 ms.
Best time for 6144K FFT length: 13.106 ms., avg: 15.480 ms.
Best time for 7168K FFT length: 15.849 ms., avg: 17.387 ms.
Best time for 8192K FFT length: 18.081 ms., avg: 19.929 ms.
Best time for 61 bit trial factors: 1.643 ms.
Best time for 62 bit trial factors: 1.675 ms.
Best time for 63 bit trial factors: 1.894 ms.
Best time for 64 bit trial factors: 1.919 ms.
Best time for 65 bit trial factors: 2.264 ms.
Best time for 66 bit trial factors: 2.633 ms.
Best time for 67 bit trial factors: 2.617 ms.
Best time for 75 bit trial factors: 2.537 ms.
Best time for 76 bit trial factors: 2.535 ms.
Best time for 77 bit trial factors: 2.536 ms.
[/CODE][LIST][*]Stable for 24h now (fresh..., in-place large ffts)[*]+50mV core voltage[*]43x101 MHz[*]Dual channel 2000MHz-XMP-CL9-RAM @ 1818MHz[*]No IGFX used[*]Hottest core @82°C with Corsair H100 water cooling and ambient 20°C[*]Blue screens start at multi >= 44 with "In-place large" ffts, not with "Blend"[/LIST]Blue screens / errors / warning and temps > 85°C are "No Gos" for me. Will continue testing.

Reason for average value timing peaks is likely to be an IO process I ran.

Wow... that's a pretty low clock for such an excellent cooler. It's too bad IB is crap at over clocking, despite the lower TDP.

Those numbers do look pretty darn good though.

Anyone know how good westmere was at overclocking? Was it as good as nehalem?

I have a suspicion that we might be in for a run of:
- on tick, new architecture same process reasonable improvement. Overclocks well
- on tock, new process minor improvements. Lower power suited for laptops. Doesn't overclock well.

I wouldn't be surprised if haswell(the next tick) overclocks just as well as sandy bridge.

From the Ivy Bridge wiki page, the reason for the high OC temps:

[url]http://en.wikipedia.org/wiki/Ivy_Bridge_%28microarchitecture%29[/url]

[QUOTE]Ivy Bridge's temperatures are reportedly 20°C higher compared to Sandy Bridge when overclocked. Overclockers[who?] speculate that this occurs because instead of using fluxless solder to transfer heat from the die to the Integrated Heat Spreader (IHS), Intel used Thermal Interface Materials (TIM) to cut cost. The thermal paste has much lower thermal conductivity, causing heat to build up on the die.[20][21][22] Impress PC Watch has proven that this is in fact true.[23][/QUOTE]

Maybe the immersive cooling techs (liquid nitrogen) might reach higher speeds than sandy bridge OCs.

-- Craig

Lame. Cheapass Intel. I'd rather pay the same price and get the same OC performance rather than have them be $20 cheaper. (To be fair, on the whole, this is probably a win for Intel, as most people don't give a hoot about OC.)

Cool down
 

[QUOTE=Brain;299239]Will continue testing.[/QUOTE]
I regarded my high temps with suspicion. I didn't mention before that one of my four cores was >= 5°C hotter than the others. This made me wonder that the cooler wasn't fully functional.
I reinstalled the cooler with new heat paste. All temps went down about 10 to 15 degree. :blush: Now the hottest core shows 72°C instead of 86°C when both GPUs are heated up.
Never trust the pre-glued atomar-thin heat paste. I noticed that 15% of the core area hadn't had any direct contact before... arg.

Rewind to 2000, the Age of AMD ... I hope I can find my old lapping plate ...

[QUOTE=PageFault;299600]Rewind to 2000, the Age of AMD ... I hope I can find my old lapping plate ...[/QUOTE]

A piece of glass worked pretty well for me in those days. What were you using?

An industrial lapping plate, which is a surface ground plate with a grid of grooves to channel away excess lapping compound. I hope it turns up ... I'll be needing it soon - the benchmarks posted here indicate how hoplessly obsolete my old junk is.

For an Intel 3770K CPU system, which of these 3 RAM modules would be go for? This is what is available:

1) [URL="http://www.ncix.com/products/?sku=65761&vpn=993996&manufacture=Mushkin%20Enhanced"]Mushkin Enhanced Redline Frostbyte 16GB 4X4GB PC3-17000 DDR3-2133 9-11-10-28 Quad Channel Memory[/URL]

2) [URL="http://www.ncix.com/products/?sku=66024"]Kingston KHX2133C11D3K4/16GX 16GB 2133MHZ DDR3 NON-ECC CL11[/URL]

3) [URL="http://www.ncix.com/products/?sku=66138&vpn=F3-17000CL9Q-16GBZH&manufacture=G%2ESkill"]G.SKILL F3-17000CL9Q-16GBZH Ripjawz PC3-17000 16GB 4X4GB DDR3-2133 CL9-11-10-28 [/URL]

I realize the Kingston Hyper-X has looser timing but if they are a lot more stable/better build quality than I would consider them. I have never owned Mushkin or GSkill so not sure what kind of quality they are. The price for all three are roughly the same.

I read a few months back on one review on Newegg that for Kingston's higher-frequency memory sticks, they didn't actually come that way; you had to manually OC them in the motherboard up to their "rated" frequency, sometimes with an increased voltage, and when the costumer complained, Kingston replied back with "Sorry, but that's OC." Well why the hell was it rated like that then?

I'll try and find that review back, to make sure I'm not misremembering.

GSKILL and Corsair are the biggest names in memory; right now I have two Mushkin sticks and two GSKILL sticks, 1333 MHz, and they both work fine. If you want more nuanced opinions, try [url]http://www.hardocp.com/;[/url] they are to hardware what this forum is to number theory.

Edit: [url]http://www.newegg.com/Product/Product.aspx?Item=N82E16820104297&SortField=0&SummaryType=0&PageSize=10&SelectedRating=-1&VideoOnlyMark=False&IsFeedbackTab=true#scrollFullInfo[/url]

It's worth your time to read the bad review, but it's also probably worth your time to read responses to it. Certainly that Kingston makes a difference between "Plug n Play" and having to manually OC it to its rated frequency is a pretty big negative in my book. (In your place I'd go with GSKILL because it's what I have already and because it's a bigger name than Mushkin (or so I think).)

I have had a very good experience with GSKILL (Ripjaws series). Fast and reliable. Definitely better than Kingston in my opinion.

I am considering replacing my obsolete junk (pentium D 3.4). If I understand correctly, the best cpu is i5 3570k.

I don't know squat about memory these days. What I want is 2 x 8 GB sticks. Are the fancy types actually useful for LL testing performance? Perhaps this depends on the boart ...

Which will be a problem. I need a 64 bit pci slot for my SCSI card. Supermicro and Tyan offer these and I'm more inclined to choose a workstation / server board anyways but they don't offer much in the way of bios tweaks. Of course my statements here may be mutually exclusive.

I guess I can keep my geforce 8400GS. One thing I won't be doing is running a buch of high end gpu cards.

I don't have much budget due to an extremely unfortunate recent event but it does look possible for around a thousand dollars.

Anyone care to comment about the boart / memory?

From a value standpoint, yes that proc is best. About the memory, are you sure you want 2x8GB? That's very high density, and will be quite a bit more expensive than 4x4GB which is the same total memory. Faster memory will help LL, especially with the AVX on [SI]B processors. I don't know much about other peripherals, but PCI slots are still standard on many desktop motherboards. Where are you planning on ordering from?

I can recommend some memory if you give a price range; motherboards are harder to pick, but you should be able to find something decent for ~$150 or so. (Are you planning on overclocking?)

Jeff, are you planning on getting an ASUS mobo with their supposed memory speed advantage?

I'm not too keen on overclocking but may use a higher multiplier, leaving a generous safety margin. I have been looking for a boart and this 64 bit slot is going to be a problem. I'm hearing to try it in a regular pci slot, knowing that it will default to 32 bit speeds if successful.

I like 8 GB sticks since they leave room for a future upgrade. I could do with 2 x 4 GB if it was necessary for budget reasons. Last thing I want to have is piles of low capacity sticks - there will soon be a day when "64 GB is not enough" ...

AVX - just how useful is it? I prefer to remain on server 2003 64 bit version. The devil I know ... What newer OS supports AVX and won't plain gross me out?

Ok Dubslow, recommend some mem sticks for approx. 100$

[QUOTE=PageFault;299931]...the best cpu is i5 3570k....[/QUOTE]
Taken out of context, I (and anybody) would certainly argue to this... But you know better your context, price/performance, power/performance requirements, etc.

AVX is +25% speedup over the SSE FFTs, which were the default until v27. With faster memory, you might even squeeze a few more percent out of it.

I had no idea about 32-bit or 64-bit for PCI slots -- as far as I know that only referred to addressable memory space.

I'll give links for 2x8GB and 4x4GB. (They'll have to be edited into the post, so check back over the next hour.)

In general, I'm giving G.SKILL examples, but if it matters Corsair or Mushkin should be good as well.

[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16820231497"]4x4GB DDR3-1600 $90[/URL]

[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16820231429"]4x4GB DDR3-1600 $93 (more and better reviews)[/URL]

Food for thought: [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16820231455"]2x4GB DDR3-1866 $60[/URL] If all you're doing is LL tests, then this might be better with the extra memory speed, assuming you use AVX. With P-1 (or if some non-GIMPS workloads) you might actually need the memory.

The cheapest DDR3-1600 2x8GB that I would feel safe buying is [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16820145357"]$110[/URL]. Many of the reviews stated that they had problems getting two of those kits = 32GB up and running, and one of the manufacturer responses stated that a straight up 32GB kit would work better than two of these 16GB kits. I'm not sure if this applies to more expensive/other brand 2x8GB mem kits.
Addendum: The [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16820231489"]cheapest G.SKILL 2x8GB[/URL] set also had the same note from the manufacturer.
[quote=G.SKILL]We are sorry to hear you are having issues. This is a 16GB kit, so it is only guaranteed to operate as a 16GB kit by itself in a single computer. For 32GB, you should purchase the 32GB kit:

[URL]http://www.newegg.com/Product/Product.aspx?Item=N82E16820231490&Tpk=F3-12800CL10Q-32GBXL[/URL]

This kit is pre-matched, tested, and guaranteed to operate in a single computer. When purchasing performance DRAM memory, always purchase a single kit of the capacity you need. If mix matching kits, it can cause problems such as what you are experiencing. For any further questions or issues, please feel free to contact us directly. [/quote]
[quote=Corsair]We're sorry to hear that you're having issues with our two memory kits, but please understand that the 32GB memory kits are designed to be more stable than two separate kits. Please contact our Customer Service department at 888-222-4346 and we'll work with you to find a solution to this problem.[/quote]


If you truly need to expand to more than 16GB, then that might be okay, otherwise I'd still go with the 4x4GB.

@LaurV: Certainly, but within the $1000, and depending on what else I needed, I'd probably go with the same proc. The larger cache on the 3770K would be nice, but it's very arguably not worth the extra $100.

[QUOTE=Dubslow;299934]Jeff, are you planning on getting an ASUS mobo with their supposed memory speed advantage?[/QUOTE]

Yes, that is the plan. Considering the Sabertooth Z77 right now.

Jeff.

[QUOTE=PageFault;299937]
there will soon be a day when "64 GB is not enough" ...

[/QUOTE]

Mostly true, but application specific. For example, if we ever tackle
RSA-1024, it will require 32 to 64Gbytes of memory PER CORE to
do the sieving. i.e. if one has a multi-core processory, one will need
(say) 256Gbytes+ of memory to use multiple cores.......

[QUOTE=R.D. Silverman;299966]Mostly true, but application specific. For example, if we ever tackle RSA-1024, it will require 32 to 64Gbytes of memory PER CORE to do the sieving. i.e. if one has a multi-core processory, one will need (say) 256Gbytes+ of memory to use multiple cores.......[/QUOTE]As you say, application specific.

One specific change to the NFS application could well be multiple cores sharing a common copy of the bulk of the memory. It's not at all obvious to me that machine architectures won't migrate in that direction over the next few years, especially given the hints that GPUs have provided over the past few. We've known for decades how to implement multi-ported memory and, for almost as long, how to put more and more intelligence into the memory controllers to take load off CPUs. What has been lacking, IMAO, is an adequate commercial incentive for such architectures to become mainstream.

Looks like I'm out of luck with my SCSI card. There are sever boards with the PCI-X slot but they support Xeons and not the modern ones. Needless to say this is out of my budget and not interesting from a performance perspective. If I proceed, I'll have to buy a 19160. SAS cards are on PCI express but this does nothing for me since I have LVD disks.

I did a search and came up with riser cards but they typically host PCI express on PCI-X - the reverse of what I need. Anyone hear of a riser card in my direction?

I can get Asus workstation / server boards for about the same price as their "hot rod" overclocking oriented boards. I do notice the server boards have the resources put into PCI express, which I don't plan on using at this time. Are the OC boards reliable? My current board, Asus P5MT, is a server model with no bios options but has been rock solid - 7 years of service, prime95 always on 24/7. Wth windows server 2003, the machine has never crashed and on occasion went 2 years between a boot ...

How much do you have invested in the LVD disks? My strong temptation with such things would be to cast them into the sea and get modern ones - the servers that we bought at the office to replace the last generation with SCSI are getting to the point where the service contract can no longer be renewed, and a modern 2TB Seagate Barracuda significantly beats the sequential read rate of a Cheetah 15K.4 from 2004.

(yes, it seeks faster)

[QUOTE=PageFault;300012]Wth windows server 2003, the machine has never crashed and on occasion went 2 years between a boot ...[/QUOTE]Clearly, it wasn't on the net during those two years. Either that, or you didn't care much for applying security patches. :wink:

IME, a major cause of downtime is erratic power supply. At home I don't have UPS so the systems go down at least once or twice a year regardless of their connectivity and patch levels. In each of my two previous jobs, we lost far more time from government-mandated electrical safety checks than we ever did from crashes or scheduled software maintenance. Bloody Elfin Safety...

Paul

It was on the intarweb all that time, running behind a firewall and otherwise unattended. I was in Africa ...

I regularly checked our team stats to see if it was updating, in the event of an outage my brother had a copy of the key ...