i7 980x Now or Wait for Sandybridge?

[Steven Davis]

Welp, it's a running. An I7

EVGA 01G-P3-1373-AR GeForce GTX 460 (Fermi) Superclocked EE 1GB 256-bit GDDR5 PCI Express 2.0 x16 HDCP Ready SLI Support ...

EVGA X58 FTW3 132-GT-E768-KR LGA 1366 SATA 6Gb/s USB 3.0 ATX Intel Motherboard

COOLER MASTER Silent Pro RSA00-AMBAJ3-US 1000W ATX12V v2.3 / EPS12V v2.92 SLI Ready 80 PLUS BRONZE Certified Modular Active ...

Patriot Viper II Sector 7 Edition 12GB (3 x 4GB) 240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Desktop Memory Model PV7312G1600ELK

Intel Core i7-950 Bloomfield 3.06GHz LGA 1366 130W Quad-Core Processor BX80601950


Western Digital Caviar Black WD1002FAEX 1TB 7200 RPM SATA 6.0Gb/s 3.5" Internal Hard Drive -Bare Drive

Pioneer Black Blu-ray Burner SATA BDR-206BKS - OEM

COOLER MASTER Hyper N 520 RR-920-N520-GP 92mm Sleeve CPU Cooler Intel Core i7 compatible


Other than not having any power when I first hit the button, lol, it is running pretty good. Windows 7 is interesting, it's much better thatn VISTA.

[Randall Leong]

Originally Posted by Scott Chichelli

warning people do not be misslead by "VA" VA is not watts.

if you have a 1000W power supply you need a 1000 W UPS. 1KA (1000VA) is NOT 1000W

I've learned that by reading up on VA ratings for UPSes. 1000VA is actually only 576W (for 120V/60Hz markets). For a 1000W UPS the VA rating has to be at least 1750VA.

By the way, I have decided to give Sandy a spin now that a limited number of revised B3-stepping P67/H67 motherboards are in stock (with more coming in next week). The first part (an Asus P8P67 Pro motherboard) is now in my possession. I also ordered 16GB (4 x 4GB) of RAM, and will purchase a 2600K within the next few days. I will be holding on to my current i7-950/X58 setup a little while longer because I will be running tests on both systems with the new memory. (This idea came after I made a claim that the X58 platforms really need astronomically expensive disk setups just to perform as well as a Sandy Bridge system with a more modest but still multi-disk RAID setup.) The loser between the two (in my own environment and with equal KISS disk setups) gets returned or sold.

[Scott Chichelli]

looking forward to your report.

Scott

[Randall Leong]

Originally Posted by Scott Chichelli

looking forward to your report.

Got the CPU today. Still awaiting the arrival of the 16GB of RAM in the mail...

I'm glad I got the RAM at a good price. Otherwise, I would have had to limp along with only 8GB of RAM on that Sandy Bridge system (via four 2GB modules).

That said, I will be testing both my current setup and my new setup (both stock and overclocked) with 16GB, and also test my current system at stock with 12GB (in both its current memory configuration of six 2GB modules and with three of the four 4GB modules) and my new setup with 8GB (both with four 2GB modules and with two 4GB modules).

A report on all of the results will be posted both in a future post in this thread and on the PPBM5 site.

[Randall Leong]

I just received the RAM this morning.

I spent the better part of the evening putting the new system together. Will be testing it at both stock speed and some overclocked speeds. I decided to forego the 8GB tests on the new system, and stick with 16GB.

Here are the stock speed results of my new 2600K versus my old i7-950, both with 16GB of DDR3-1600 RAM (all other components except for the motherboard are the same for both systems):

• i7-2600K: 220 seconds overall (90 seconds AVI, 37 seconds MPEG-2, 82 seconds AVC, 11 seconds MPE); Performance Index: 184.9
• i7-950: 247 seconds overall (100 seconds AVI, 43 seconds MPEG-2, 93 seconds AVC, 11 seconds MPE); Performance Index: 202.8

And yes, the Gigabyte motherboard that I used for the i7-950 claims to support "full" triple-channel with four DIMMs of equal size.

For a fairer comparison, I re-ran the same test on the stock-speed i7-2600K system with only 8GB of RAM, and compared it with the results of an otherwise identically equipped stock-speed i7-950 system with 12GB of RAM:

• i7-2600K (8GB): 288 seconds overall (89 seconds AVI, 102 seconds MPEG-2, 87 seconds AVC, 10 seconds MPE); Performance Index: 263.5
• i7-950 (12GB): 297 seconds overall (99 seconds AVI, 93 seconds MPEG-2, 94 seconds AVC, 11 seconds MPE); Performance Index: 265.4

The stock i7-2600K system, in my testing, does outperform the stock i7-950 system due in large part to the higher stock clock speed of the 2600K. The i7-950 system with 16GB of RAM, although it was noticeably faster than that same system with 12GB of RAM, was hobbled slightly by its memory controller actually operating in the hybrid Flex mode in which the extra 4GB ran in single-channel mode rather than true triple-channel.

One more note: I discovered that the highest-performing stock i7-2600(K) system that's currently on the PPBM5 results list (with a total time of 264 seconds) did not have a RAID array at all. Instead, it relied on a single 2TB, 7200RPM hard drive for everything except the OS. As a result, that system was hobbled by a slow disk subsystem. For comparison, the stock i7-950 system in the PPBM5 list that Harm mentioned in the Adobe forums delivered a 243-second result mainly due to its 24GB of RAM running in true triple-channel mode despite its two 150GB 10,000RPM hard drives in a RAID 0 array underperforming several two-disk RAID 0 arrays with 7200RPM hard drives (although it was still slower than my stock-speed i7-2600K results with 16GB of RAM).

Results have been submitted for posting on the PPBM site.

[Scott Chichelli]

while PPBM is ok for an idea a better benchmark it to take the same footage on both systems and render it out EG AVCHD to H264.

Scott

[Randall Leong]

Originally Posted by Scott Chichelli

while PPBM is ok for an idea a better benchmark it to take the same footage on both systems and render it out EG AVCHD to H264.

To make that comparison meaningful, I have to use the exact same video and disk components on both systems. (This means that the only variables allowed are the CPU, motherboard and RAM amount.) If any of the disk or video components differ even slightly between the two systems, the results would be definitely skewed one way or the other.

The comparison that you gave between the i7-2600K and the i7-980X earlier is somewhat meaningless because there are too many variables between the two systems - disk subsystems, graphics card drivers and memory speeds, as well as the degree of overclocking. For all I know the memory in the 2600K system was running at its official DDR3-1333 speed while the memory in the i7-980X system was held back to DDR3-1066 speed. And in practice dual-channel 1333-speed memory actually delivers greater bandwidth than triple-channel 1066-speed memory.

[Scott Chichelli]

all ram in my systems always runs at 1600 nothing less.
for my benchmarks we always use the same disks etc we always use 2 sets raid 0 and standard Sata OS (unless doing a drive benchmark) then everything is the same but the Drives.
we have also tested ssds as media drive, temp file drive OS drive etc (pointless)

the only variables are the Mobo, CPU
we may have a different videocard in there but really for the test we are doing it matters not.
in fact here the SB has the lower video card and still beats the 980x

I7 2600K 3.4GHZ Turbo to 4.7GHz
16GB Blackline 1600 CL 9
470GTX
3 Layer - 31:35
4 Layer - 34:35

I7 980X 4GHZ
12GB Blackline 1600 CL 9
570GTX
3 Layer - 32:30
4 Layer - 35:25

as far as ram quantity..
note the 8 gig and slower video vs the 16gig and faster video.
with the same GHZ CPU.. not a huge performance difference at all.
break it down to seconds.
40:49 = 2449
40:05= 2405
about a 1.74% better performance with double the ram and a better video card.

I7 2600 3.4GHZ Turbo to 3.9GHz
8GB Blackline 1600 CL 9
460GTX
4 WD 1Tb Sata 64 Meg Cache 600 Drives in 2 Raid 0 arrays
3 Layer - 37:35
4 Layer - 40:49

16GB Blackline 1600 CL 9
570GTX
3 Layer - 36:17
4 Layer - 40:05

now 2 things i need to comment on the AVCHD to H264 test is too lite on a system to show any serious differences.
we picked it 2 yrs ago as it was (and still is) the most common workflow. (we also do a red 4k which does show better)

i also think the PPBM is too lite and too short as well. but for a downloadable test you have no choice..

so we started adding the lightning effect and have not finished collecting numbers for everything.. (too stinking busy :-) )

Dual Xeon X5680 CPU's at 4.0GHz
48 GB DDR3 1600 Blackline at 1600
4 WD RE4 2Tb Sata 64 Meg Cache Drives in Raid 5 array
580GTX
3 Layer - 31:00
4 Layer - 31:11

3 Layer w/Lightning - 1:11:52
4 Layer w/Lightning - 1:26:36

I7 2600K 3.4GHZ Turbo to 4.5GHz
16GB Blackline 1600 CL 9
470GTX
2x raid 0
3 Layer w/Lightning - 1:46:37
4 Layer w/Lightning - 2:05:44


so again as i said you need to test both systems with YOUR normal workflow and make them as identical as possible to see which better suites you.

Scott
ADK

[Randall Leong]

Originally Posted by Randall Leong

Thanks for the proof that the LGA 1155 platform is ill-suited for anything more than a KISS drive configuration. Although there are technically four PCIe lanes open for PCIe expansion cards on the LGA 1155 platform (after accounting for onboard USB3, PCIe-to-PCI bridge and additional SATA/IDE controllers), the few motherboards that have PCIe x4 slots actually share the bandwidth of that slot with the PCIe x1 slots on those same boards. This means that ANY card plugged into a PCIe x1 slot would have forced the x4 slot to run in x1 mode. That hurts the performance of most discrete hardware RAID controllers. The only other place to put a RAID card would have been the second PCIe "x16" graphics slot (which is bifurcated from the main PCIe x16 slot, and thus both slots would have been forced to run in x8 mode). This does hurt MPE GPU performance.

A few high-end LGA 1155 motherboards have onboard PCIe lane repeaters. Unfortunately, while they create additional PCIe lanes, they do not change the total bandwidth of the CPU's integrated PCIe controller. So instead of 16 PCIe lanes operating at full PCIe 2.0 bandwidth, you now have 32 PCIe lanes that are artificially restricted to PCIe 1.0 bandwidth -- 2.5 GT/s instead of 5.0 GT/s.

Originally Posted by Randall Leong

The condensed version of my above post is:

Unless someone has only an Nvidia graphics card and a discrete hardware RAID card (and no other expansion cards whatsoever), the current Sandy Bridge (LGA 1155) platforms do not allow both the graphics card and the RAID card to be run at their full speeds at the same time. Either the disk performance or the GPU performance would suffer, especially if that configuration also includes a sound card that's of better quality than the onboard one.

I was partially wrong about the above quotes. Looking at other posts in other forums, I discovered that even the PCI-e x16 slot dropping to x8 speed does not reduce the performance of current Nvidia GPUs because even the fastest of such GPUs (including the GTX 580) do not take full advantage of x8, let alone x16. This means that if one has a P67 motherboard with two x16 physical PCI-e slots that both run electrically at x8 when both slots are filled (these include the Asus P8P67 Pro or higher, the Gigabyte GA-P67A-UD4 or higher, the MSI P67A-GD## series and the Intel DP67BG), one can run a hardware RAID card from the likes of Areca with virtually no MPE performance penalty. Forget about using a hardware RAID card on a P67 or any other motherboard with only one PCI-e x16 slot; the two just won't work together. And if the secondary physical x16 slot runs in x4 or x1 mode (as is the case with the Asus P8P67 or P8P67-LE, or the Gigabyte GA-P67A-UD3 series), the disk performance (corresponding to the AVI portion of the PPBM5 benchmark) will be less than optimal - and might not be any faster than a simple two-disk RAID 0 array on the onboard Intel software RAID controller.

I've also discovered that low MPE performance with any given Nvidia GPU is due to an improperly tuned system and/or improper graphics driver settings and/or an excessive number of processes running in the background, not the limitations of the PCI-e bus in current mainstream Intel platforms.

As such, the PCI-e limitations of the LGA 1155 platform can potentially reduce performance with future components. In practice, however, the performance reduction with current components is virtually nil.

As for the replacement for the current LGA 1366, there will be none (technically). Based on ever-changing plans, there will be no desktop CPU from Intel that uses the LGA 1356 socket. All of the new Intel CPUs higher than the current 2600K will be LGA 2011 only. However, the desktop i7 Extreme LGA 2011 CPUs will be gimped to only 24 PCI-e lanes (instead of the full 40 PCI-e lanes in the server Xeon versions of the same CPU). That's only four PCI-e lanes more than the 20 PCI-e lanes in the current LGA 1155 CPUs (plus any additional PCI-e lanes available from those on the PCH). Because of this gimping, the forthcoming LGA 2011 platform replacement for the current i7-9xx series will be much less attractive than one would expect since the only CPUs that are equipped to support a hardware RAID controller would cost more than $2,000 for each CPU. The current LGA 1155/P67 platform theoretically has 28 total PCI-e lanes, of which four of those from the CPU are disabled during manufacturing (leaving only 16 PCI-e lanes available from the CPU) and anywhere from four to six from the P67/H67 PCH being eaten up by the motherboards' onboard devices (e.g. USB 3.0, an extra SATA 6.0 Gbps controller, etc.).

That said, there will be LGA 1356 CPUs on the market - but they (according to current plans) will be sold only as Xeon processors for single-CPU servers. And even if the higher-end i7s were to be available in LGA 1356, that socket would not be much if any better of a choice than LGA 1155: The LGA 1356 CPUs will be designed with only 24 PCI-e 3.0 lanes in the CPU - still not enough lanes unless the chipset can access more than 16 on-CPU lanes.