Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Sandy Bridge E processors out today...

I spent an evening pondering on pretty much the same question as Philipe. I then discovered that, going by Passmarks' CPU Mark results, a singular Core i7-3930K is faster than a dual Xeon E5645 -- and quite a bit cheaper, too.

For the motherboard I picked the Asus Rampage IV Extreme because it is currently the only LGA 2011 mobo that has enough PCI-E lanes for my needs. Barely; you can have one x16 slot and three x8 ones. That should be good enough for a Quadro 4000 in the x16 slot, followed by a RED Rocket, a GeForce GTX460 and a SCSI adapter in the x8 slots. I bought 32GB of Corsair Low Profile Vengeances because I had reason to believe DIMMs with tall heat spreaders would not fit under the Noctua cooler.
 
Can anyone who has built a Core i7-3930K system comment on how the system performs?
 
Brandon, I ordered the components only yesterday... but I will try to remember to comment on it once all the parts are here and the system is up & running.
 
Mike P. said:
With those kinds of savings you could essentially buy another updated system later this year guilt-free (that said, IvyBridge is supposed to be pretty much the exact same as SandyBridge-E, except have ondie graphics and 20% heat/power reduction.)
Click to expand...

Ivy Bridge will certainly be slower than Sandy Bridge-E for multi-threaded apps. It is still quad core, and Intel is focusing on power efficiency and graphics - CPU performance will be a minor boost over Sandy Bridge. By that time, Sandy Bridge-E will get a minor refresh too, with i7 3940K and 3970X. Besides, Intel has to protect its high-end - Sandy Bridge-E will be the fastest desktop CPUs till Ivy Bridge-E hits late 2012. After all, the top Ivy Bridge is Core i7 3700, Sandy Bridge-E is Core i7 3800 and 3900. The only wild card here is overclocking, but it won't be able to make up for the lack of 2 cores and only a minor IPC boost. Plus, trigate fabs are aimed at high power efficiency in the 2 GHz - 4 GHz range.

Also, we must remember that the Sandy Bridge-E die is identical to Sandy Bridge-EP. It contains 8 cores and 20MB L3 cache. The only reason why only 6 cores are enabled is to fit in under 130W. In the future, with process improvements and a new stepping, 8 cores at 130W is certainly no stretch of the imagination. We can see a Core i7 3980X or something that is 8 core, sometime in 2012. Sandy Bridge-EP Xeon E5s bring 8 cores at 150W in January.
 
Philipe Ratton said:
Very new to the PC world, tired of waiting and a bit scared of apple.

What dou you guys think for a CS5/Resolve system:

Intel Core i7-3960X Extreme Edition 3.3GHz (3.9GHz Turbo) LGA 2011 Six-Core Desktop Processor
ASUS Rampage IV Extreme/BF3 Extended ATX Intel Motherboard
G .SKILL Ripjaws X Series 32GB (4 x 8GB) 240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Desktop Memory
EVGA GeForce GTX 580 (Fermi) 03G-P3-1584-AR Video Card
EVGA GeForce GTX 560 Ti (Fermi) 02G-P3-1568-KR Video Card (GUI)
Click to expand...

I just built nearly that exact same system. I suggest you go with 4GB RAM modules, as in 8x4GB. You can go with the 2133MHz G.Skill Ripjaws and do 32GB for less money that way. That motherboard has 8 DIMM sockets and SB-E supports quad-channel memory. So you want to install either 4 or 8 memory modules and make sure they're all identical. What do you plan to use for SSD/HDD's? I kinda went all out (or at least spent a bit) and have a 480GB Mushkin Chronos 6Gbps SSD for my system/apps drive and then 4 x 2TB Seagate Barracuda XT's (attached to an Areca 1882LP RAID card) in RAID-0.

The i7-3960X is probably overkill for most people unless they're doing a lot of rendering or intense computational stuff. You can save yourself a bit of money by dropping down one model of CPU. You still get 6-core. The difference is smaller cache (which is what makes the difference in the computationally intense stuff) and it is clocked at 3.2GHz instead of 3.3GHz.

(2X) Intel Xeon E5645 2.4GHz LGA 1366 80W Six-Core Server Processor
EVGA Classified SR-2 270-WS-W555-A2 HPTX Intel Motherboard
(4X) G.SKILL Ripjaws Series 12GB (3 x 4GB) 240-Pin DDR3 SDRAM DDR3 1333 (PC3 10666) Desktop Memory Model F3-10666CL7T-12GBRH
EVGA GeForce GTX 580 (Fermi) 03G-P3-1584-AR Video Card
EVGA GeForce GTX 560 Ti (Fermi) 02G-P3-1568-KR Video Card (GUI)

any advice is great appreciated !
Click to expand...

The Xeons are due for an update soon. And the i5520 chipset especially is feeling very dated (it's 3 years old, after all). But that's a nice system and while it's clocked slower, it has a heap of memory bandwidth and quite a bit of muscle if you run a lot of multithreaded applications. I'm confused on your RAM selection you have (4X) in front of it which implies you're buying more than the 3x4GB (12GB) you list later on the same line. With this configuration you can take advantage of a triple-channel memory interface for each CPU. So RAM works best installed in matching groups of 3 for each CPU RAM bank.

Personally, I would go with the Sandy Bridge E system, probably with the 3930 CPU. ...Unless you really feel you need the 3960X. Which I would recommend if you're doing a lot of rending or other computationally intense multithreaded stuff. If you're playing games, editing in Premiere and playing in Resolve, the 3960X is probably overkill.
 
Intel Core Extreme Edition i7-3960X Six-Core Socket LGA2011, 3.30Ghz, 15MB L3 Cache, 32nm (Retail Boxed) (BX80619i73960X) $1,199.99

Asus P9X79 PRO Socket 2011 Intel x79 Chipset Quad-channel DDR3 2400(O.C.)/2133(O.C.)/1866/1600/1333/1066 MHz 4x PCI-Express
3.0 x16 Dual GLAN 8-CH High Definition Audio 4x SATA 3.0Gb/s 4x SATA 6.0Gb/s 2x Power eSATA 6Gb/s 6x USB 3.0 12x USB 2.0
Bluetooth V2.1+EDR ATX $319.99

G.SKILL Ripjaws Z Series DDR3 1600MHz (PC3-12800) 64GB (8x8GB) Quad Channel Kit (F3-12800CL10Q2-64GBZL) $1,099.99

OCZ (AGT3-25SAT3-480G) Agility 3 2.5" 480GB SATA3 6GB/s Solid State Drive, Read: 525MB/s, Write: 410MB/s $699.99 - main
OCZ (AGT3-25SAT3-480G) Agility 3 2.5" 480GB SATA3 6GB/s Solid State Drive, Read: 525MB/s, Write: 410MB/s $699.99 - swap/scratch
OCZ (AGT3-25SAT3-480G) Agility 3 2.5" 480GB SATA3 6GB/s Solid State Drive, Read: 525MB/s, Write: 410MB/s $699.99 - working drive

Maximus Config
PNY NVIDIA Quadro 6000 (VCQ6000-PB) 6GB GDDR5 Dual-Link DVI/Dual Display Ports/Mini Stereo PCI Express 2.0 x16 Graphics Card $4,099.99
Nvidia Tesla C2075 GPU Computing 448 CUDA Cores, 3 GB DDR5 Memory PCI-E 2.0 $2,549.99


DroboPro Dual Disk Redundancy 1,499.00 <- i know some of you hate these. I've had no prob so far.
Hitachi Deskstar 3TB 3.5" SATA3 6Gb/s 7200rpm 64MB Cache OEM (0S03208) $299.99 X 8


drool
 
Patrick OMara;898671 Nvidia Tesla C2075 GPU Computing 448 CUDA Cores said:
Unless you're running one of the 2 or 3 apps that actually support this, why bother? And with one of them, it's kinda like "meh..."

DroboPro Dual Disk Redundancy 1,499.00 <- i know some of you hate these. I've had no prob so far.
Click to expand...

I know a guy who sells mopeds that cost almost as much as a new Harley...

Really though, if you're building that, looks like fun. :)
Click to expand...
 
Patrick, check the Vertex 3 SSDs specs in comparison to the Agility. The sustained writes on the Vertex are favorable.

Are you going to OC this? If so, I would recommend getting some G.Skill mem that has tested to clock higher at like 1866 or 2000. For my build I did Corsair Vengence 1866 CL10 (4x8) kit paired with the P9X79 Deluxe and the 3930k and it works very well. Stable OC to 4.9 GHz, on H100 closed-loop CPU cooler.
 
@Jeff K,

I'm really interested to see how your build benches in comparison to my 3930k system. Your data disk i/o will clearly be better with the hybrid drives in RAID 0 and the Areca controller. But I'm interested. If you get time, please share.
 
jon, ya sustains are important but this ssd should be able to handle about 7 streams of scarlet footage at a time, that's max of what I'd need for sure. Especially if the ssd is not being hit by anything else.
 
I've done a little benchmarking, but I'm going to save most of that for when I get the thing properly water-cooled and overclocked. After I get through all this holiday madness... So far, even at stock speeds, it's very fast. I had it OC'd to 3.9/4.6turbo (I think that's what it was), but temps would climb on me so I throttled it back. I believe I was at +1.4V and it could've handled more speed just fine.

Drives I'm using are not the hybrid ones. I would actually prefer the WD 2TB RE4 drives, but with current pricing I went with the Seagates. I paid $244 each, the RE4's would've been $350 each at the time I ordered. Drive prices are creeping back down finally though. Can buy 1TB 7200rpm drives for under $100 again now if you look around. NewEgg has the Samsung F3 for $99 today after instant coupon code.

The new system is sitting next to my '09 Mac Pro in my home office and it's much snappier overall when running the same things -- Premiere, RC-X Pro 8, Photoshop, Lightwave, Modo, Maya... The last 3 just scream on this new system and a lot of that has to do with the GTX580s. I have Quadro 4000's in the Mac Pro... This Mac Pro is a 2.93GHz Nehalem 8-core (dual quad-core Xeon), 24GB, 1TB WD Black-X system / apps drive, 4 x 2TB RE4 in software RAID-0. Will be interesting to compare. All I know is I have way too much money in this Mac Pro system and I'm planning to move my Resolve operations to the new PC and ditch the Cubix box and second quadro card. It's been a good solution, but not cost effective now that Resolve for Windows is in public beta form.

App load times are obviously faster on the PC where they load from SSD. But still way snappier once loaded and running.

Patrick OMara said:
jon, ya sustains are important but this ssd should be able to handle about 7 streams of scarlet footage at a time, that's max of what I'd need for sure. Especially if the ssd is not being hit by anything else.
Click to expand...

Just a warning on SSD transfer rates. These Sandforce based SSDs use on-the-fly compression to pass more data across their interconnects. When moving large files that are already highly compressed (R3D, H264, etc..) you won't get the advertised speeds. Don't get me wrong, they're still blazing fast, but the compression makes an impact -- typically the biggest impact is on their marketing numbers. [/COLOR]
 
Jeff, thanks a lot for the info.
The system is cs5/resolve only I don't want to deal with any instability at all.
Maybe my past experience with PCs is just bad.
The memory modules would be 4x triple channel 3x4Gb, 48GB total.
But after reading your posts and some others, I decided for the 3930k.
 
Jeff Kilgroe said:
Just a warning on SSD transfer rates. These Sandforce based SSDs use on-the-fly compression to pass more data across their interconnects. When moving large files that are already highly compressed (R3D, H264, etc..) you won't get the advertised speeds. Don't get me wrong, they're still blazing fast, but the compression makes an impact -- typically the biggest impact is on their marketing numbers.
Click to expand...

Yeah, I currently have 3 SSDs rated at 275MB/s but I get a solid 200MB/s from them. I'd expect on 525MB/s that you'd get maybe a solid 375MB/s. 7 x 55MB/s . Just a rough estimate. Should be good. :)


+1 never believe the marketing numbers.
 
Patrick OMara said:


Yeah, I currently have 3 SSDs rated at 275MB/s but I get a solid 200MB/s from them. I'd expect on 525MB/s that you'd get maybe a solid 375MB/s. 7 x 55MB/s . Just a rough estimate. Should be good. :)


+1 never believe the marketing numbers.
Click to expand...


NEVER. But, the Vertex 3 has been independently benchmarked at around 500mb/s from multiple sources. I was looking into SSDs for work this week for one of our graphic artists.

Granted, I doubt, as Kilgroe points out, that it was benchmarked using compressed footage files. But the fact remains: the independent benchmarks don't place it too far below the advertised 525MB/s speeds.
 
SDK limiting

SDK limiting

Good to see more SB-e computers being set up for r3d processing.
So once you start looking at actually using some r3d's you will no doubt see that your v8 is firing on 3 cylinders. While the computer will indeed be faster than the old rig, it is still a disappointment in what it could be doing.
We need to ask Red to please multithread the SDK better!

I can light up all the cores on my system to do RCX derived r3d -> other codecs, but it's not easy to do and a clumbersome non GUI workflow.
To put it another way, my old i7 system would easily outperform my new SB-e system, if the SDK was optimized!
 
The incompressible data issue that Jeff mentioned has been fixed in the latest Sandforce chipsets (Adata and OCZ are the two major manufacturers that use Sandforce chipsets), but you really should look into a specific drives performance in this area before you buy. It is definitely worth getting a drive that doesn't suffer from the problem; when it's large amounts of incompressible data the speeds can clamp down quite a bit (half). Video production definitely falls victim to large amount of compressed data, so it's probably worth the extra $$$ for a drive that doesn't have the issue.

@Les: Are you just running three or four instances of RCX to transcode on all cores? Also, if you use Adobe Media Encoder to transcode instead of RCX, is it any faster (since Media Encoder should be better at using more CPU cores) or is that also limited by the SDK? Theoretically (I know, I know), if it takes 4xRT to transcode out of RCX, but it's only using 1 CPU/Core, then it should be possible to run/transcode in realtime if you could get it to use 4+ cores. Similarly, if there was a way to put a transcode going on each core, you could essentially render 4x/6x/8x faster (depending on how many cores) and it'd probably be way easier to do with a simple GUI or option in RCX (the 'amount of simultaneous exports' option is useless if all exports are only running off one core).

I also read somewhere that CS5 doesn't take advantage of more than 4cores (not even the additional threads of HT), which would make even hexacore a waste, but Jeff said it's actually efficient up to like 12-16 threads now. With that in mind, Hexacore (+HT) seems to be the route to take for price/performance, and that's exactly what SandyBridge-E is.

I was kind of hoping that some whiz's would come up with a way for r3d acceleration on GPUs as opposed to having dedicated hardware like an RR... I know it would shoot RR sales in the foot, but it'd be so much cheaper.
 
You must log in or register to reply here.
Back
Top