[Ross]

I was a bit bored last night and decided that since I hadn't fired up the 920 in while, I'd do a short water-cooled bench session with identical speeds between the P55-UD6 + Core i7 860 (1156) and the X58-UD4P + Core i7 920 (1366).

There's nothing very scientific here, just comparative test on a whim. What it's not, is a comparison of newly released h/w against some tests in the bank from many months ago when the X58s and 920s were new. These were all done back-to-back on the same cooling, HDD, OS, vid card, memory, etc. The only things that changed were the mobo and CPU.

1156 setup
Core i7 860 ES
Gigayte P55-UD6 (F3 BIOS)

1366 setup
Core i7 920 retail (D0)
Gigayte X58-UD4P (F7 BIOS)

Same hardware/software used in both setups:
2x2GB from a Kingston KHX16000D3ULT1K3/6GX kit (dual-channel)
Gigabyte 260 GTX SP 216 (stock cooling)
VelociRaptor HDD
Odin 1200W PSU

Win XP SP3
Forceware 190.62

NexXxos XP water block
2x BIPII rads in series with 1x Panaflo 120 H1A on each
MCP-355 pump

Both systems were setup as:
- 21x200 boot & bench (920 = 20x + Turbo, 860 = 21x w/o Turbo)**;
- All cores + HT were kept on for all benches;
- Uncore set 18x on both. This is the highest setting for P55/860 and lowest setting for X58/920;
- QPI set 18x on both. This is fixed on P55/860 and an underclock from the default 19x on X58/920 (and adjustable higher);
- Memory had manually set primaries/tRFC, everything else was left on AUTO.
- All EIST, speed step, C6/C7, etc. was disabled;
- ALL voltages (including Vcore, Vdimm, Vtt, IOH, PCH, CPU PLL, etc.) were left on AUTO;
- PCIe freqs were left on AUTO;
- Vid card clocks were the same saved profile and clocks kept purposely low (700/1509 (linked)/1200) to avoid heat causing inconsistencies;

** I am not aware of anything changing (ie. internal timings, etc.) with "Turbo" being enabled to get the 21x multi on the 920, however, there is always that possibility and I didn't think of it until today. At some point, I may do a 20x run with both setups just to check.

I just did a couple benches each since it was impromtu. I only used a single vid card so PCIe lane divisions didn't come into play...the card was getting the full 16x on both boards.

These were the settings used for the benches. All mem subtimings are AUTO settings.

P55-UD6 + Core i7 860



X58-UD4P + Core i7 920



3D06 - P55



3D06 - X58



SPi 32M - P55



SPi 32M - X58



wPrimes - P55



wPrimes - X58



Everest bandwidth - P55 (note: "not optimized" for Lynnfield yet)



Everest bandwidth - X58



There is definitely a hint in the Everest screens that there is more going on than meets the eye between the two (if the bench results didn't already say so). Even though the CPU, mem, QPI and Uncore are all at the same speed, the X58 still has a faster latency and slightly more bandwidth. I always take Everest results with a grain of salt because they can move around a bit, but SPi 32M is wholly dependant on CPU and mem. Given the CPU and mem are the same speeds (with slight differences in a few subtimings), the difference in 32M is very noticeable.

So even an X58/920 "crippled" to P55/860 specs and in dual-channel mode, it still out-performs 1156, if just by a bit. It would not take much for the P55/860 to match the X58/920 at P55 settings. With identical settings as tested above, but with 1000MHz 7-8-7-88 mem timings on P55 instead, wPrime is nearly identical and 32M is less than 1s difference (3Ds don't improve as much however).

On that note, between these two boards, it is much more difficult for me to run 1000MHz+ mem on the X58/920. On the other hand, the P55 board does just silly things with the same memory (>DDR3-2500 9-9-9-24). Suffice it to say, if it just takes a bit more memory clocks or a few timing tweaks to nearly equal X58/920 at the same setup, the P55 should be able to do it pretty easily if your mem is up to it.

Long story short, Core i7 1156 is honestly not that far off of Core i7 1366 performance when set up evenly (and I know some game tests have 1156 beating 1366). Either way, as far as "mainstream" goes, P55/1156 is pretty damn good!

[Ross]

P55/860. Same settings as above except 1000MHz 7-8-7-20 mem instead of 800 6-7-6-20. Much more comparable to the X58/920 32M run above (8m 43.218s). 3Ds do not gain nearly as much on P55 with increased mem. I'll do some 3D compares when I have time.

[Ross]

Reserved again in case I do more tests

[grishenko45]

nice post!!! its odd that intel say that i5 is going to be mainstream - but the "i7" 1156 is as fast as an i7 1366!!

[grishenko45]

nice test!! its odd isnt it - that intel say i5 is mainstream - but the "i7" 1156 is as fast as an i7 1366!

[Daddyjaxx]

Well...they are both I7's. The only difference between the two is triple or double channel memory and PCIE which you negated by using one card and running both in dual channel. The 860 already starts out with a multiplier head start over the 920. You have to overclock the 920 more to get the same speed hence the slight score increase. You could see the same results comparing a 920 to a 920 on two different boards or a 920 vs a 940 on the exact same board at the same speeds with the 920 winning there too. I would expect any I7 to benchmark practically identical at the same speeds.

[Ross]

The 920 isn't OC'd "more". Multi's only matter if you are running defaults or have fixed multi's. 940 is faster than a 920 at default only, but they are the same cores and a 940 set to 20x IS a 920. 920 vs. 940 in the same board at the same multi/BCLK will be identical (within a margin). The above benches are not using higher BCLK at a fixed multi to match CPU speeds, it's the same BCLK and multi, so they are clocked identically.

Both are called Core i7 yes, but they are different, so is the chipset. Core i7 1156 is high-end maintstream and the goal was just to see how close 1156/P55 performance would be to 1366/X58 if X58 were downgraded and running at P55 limitations. The performance options of X58 are obvious and uncorked (ie. default setup or OC'd), the X58/9xx obviously has better performance per clock, but I wanted to see them compared on a level field.

[nzaneb]

Nice work Ross! Are you going to run any other 3D benches?

[Gautam]

Excellent test Ross. Pretty surprising actually. It's enough to matter for benchmark nuts like ourselves. 7 secs in 32M, 600 pts in 06...and that's with the uncore held constant and dual channel for the X58 as well. Don't know why the Lynnfield is lagging so much, but it's clearly noticeably behind...that 06 CPU score hit...damn.

[Ross]

Originally Posted by nzaneb

Nice work Ross! Are you going to run any other 3D benches?

At some point, but not on water this weekend

Originally Posted by Gautam

Excellent test Ross. Pretty surprising actually. It's enough to matter for benchmark nuts like ourselves. 7 secs in 32M, 600 pts in 06...and that's with the uncore held constant and dual channel for the X58 as well. Don't know why the Lynnfield is lagging so much, but it's clearly noticeably behind...that 06 CPU score hit...damn.

Yeah, it's a hit. I'm surprised X58 is still that far ahead at same everything. 600 in 06 is pretty big. I did a quick 32M on P55 same settings everywhere except mem @ 1000MHz 7-8-7 and it's almost dead even in 32M (like .5s diff). I'll post the screen later when I get home. 3D06 with the raised mem is still like 450-500 points behind.

The 860 does OC a lot higher than the 920 on DI, so some 2D 860 LN2 runs might be in order this weekend. Come to think of it, I don't think I ever had the 920 on LN2. That might need a round for 3Ds too

[Brolloks]

Nice data Ross, can you put another GTX 260 in their to see how the SLI scaling is between the two boards?

[Mega_Death]

When the i7 800 series drops in price a little more it will be a great value. But as long as the prices are simialr the x58 platform is the better choice.

[Ross]

Originally Posted by Brolloks

Nice data Ross, can you put another GTX 260 in their to see how the SLI scaling is between the two boards?

I will do an SLI on both of these setups, but I'm pretty sure it will be more of the same...X58 out in front by a noticeable margin.

Originally Posted by Mega_Death

When the i7 800 series drops in price a little more it will be a great value. But as long as the prices are simialr the x58 platform is the better choice.

Point taken. I don't think it's going to take too long for the prices to come down though for either 1156 CPUs or mobos. I guess it depends how well they sell. Either way, it's just a matter of what you're willing to spend for what you want to get. Similar performance for noticeably cheaper always makes the decisions easier

[atm743]

the P55 i think was made just for main stream computers for manufactures like dell, HP, Gateway, acer .....

being that they integrated the north bridge in the CPU, only dual channel motherboards should be cheaper to manufacture and sell wholesale to these companies BUT like you said, the performance compared to the 1366 slot I7's are pretty close.

in the coming year, the market is going to be flooded with Core i5 and core I3 machines, as well as some 1156 slot i7's for the higher end entertainment computers.

being that gulftown (6 cores 12 threads) will be in Q1 as well as sandy bridge.

http://www.brigs-family.com/?p=159

after looking here, the 8 core 16 thread Nehalem-EX (Beckton)
# up to 8 cores, up to 16 threads
# 24MB shared L3
# 45nm
# 90/105/130W TDP
# 4 FBD2 interfaces
# 4 QPI links
# Socket-LS (LGA 1567)
# 44 bits physical address
# 48 bits virtual address

makes me

[Ross]

Core i9 has already been benched. It's sick. 6/12 >6GHz frozen. OMG. I'm saving already

[Boulard83]

nice reading !!!

[disk11]

I can't wait for the gaming tests.

I was curious about the costs of the varible components so I look 'em up:
860 setup: $539.98 (just newegg)
920 setup: $509.98 (newegg cpu, ewiz mobo)

Even using microcenter for processor prices, the 920 is still cheaper. Any word on when pricedrops will occur?

[Alien432]

Excellent work Ross !

P55/i7 860 quieter and cooler but price wise i7 920 is a better value to me.

[dudleycpa]

Thanks for the info. Ross - Nice work!

[4GHZ_or_bust]

socket 1156 is made for mid range mainly and has lower cost vs 1366. The biggest reason for cost difference has to do with the RAM. 1156 is dual channel and 1366 is tri channel. Designing tri channel takes more work, needs more traces, and costlier multilayer board than dual channel. OTOH tri channel has better RAM/CPU performance than dual channel.

Another consideration when chopping for cheaper 1156 vs pricier 1366 is the PCIe support. At this time, 1156 has PCIe support built into CPU (pretty much killed off Northbridge chip) but is limited to 16x so those using high performance dual or tri video card in SLi or Crossfire will take serious hit but using single video card only, performance between both should be very close.

Bottom line, if you plan to use mainly 1 video card (or 2 lower/mid range video cards that won't saturate 8x each), don't really need the faster CPU/RAM performance and don't plan to get the i9 at all, 1156 is much cheaper and both can overclock quite well.

[hokiealumnus]

Excellent comparison Ross, thanks!

[Ross]

I tossed up the 32M screen of the P55/860 with 1000MHz mem instead of 800 in post #2. I will try to get some more compares (3Ds) done maybe tomorrow. If not, during the week.

I spent all last night fighting it on LN2. It all handles cold just fine, but I think I am reaching the limits of the IMC/mobo (~240BCLK benchable). 245BCLK was about my max ever just for a screen, so 240 benchable isn't bad. There seems no way around the limit though unless I really start loosening things up (which defeats the purpose). More cold and more voltage isn't working, so I'll try some finesse today. If that doesn't work relatively quickly, I may give the 870 a give go on cold in the UD6 before I run out of LN2.

Originally Posted by disk11

I can't wait for the gaming tests.

I was curious about the costs of the varible components so I look 'em up:
860 setup: $539.98 (just newegg)
920 setup: $509.98 (newegg cpu, ewiz mobo)

Even using microcenter for processor prices, the 920 is still cheaper. Any word on when pricedrops will occur?

Keep in mind 1156 is brand new. 1366 has been around a while at this point. I remember it wasn't long after Q6600s came out that you could find them on ridiculous sales for $199 at Egg, Frys, Microcenter, etc. Come to think of it, I've seen $199 sales for the 920 at the same places. Either way, check again in a few weeks, and the holiday shopping season is right around the corner so there's bound to be some deals

[disk11]

I hope I have a job to take advantage of said deals. My current system is getting long in the tooth.

[Gautam]

massman suggests that you try a newer BIOS on the P55, and if possible, try a C0 in the X58 with an older bios.

And also, I ask that you make sure that the RTL's between the two were consistent in your testing if possible.

[Ross]

I don't have a C0 available. Have massman send me a PM so I can give him my shipping address I'm on the 870+UD4P now. BIOS F4m for that board, but I'll see if I can dig up anything newer. It's pretty much copy/paste scores of the 860+UD6 at the same settings.

I'll take screens with RTL showing all future benches just for you , but I don't recall seeing anything different than 54 on channel A (56 on B IIRC). As I am not making any changes to the BIOS, h/w or mem tweaks on the desktop during these benches, it stands to reason RTL isn't fluctuating

So far, the only thing I've really seen in my normal benching (ie. not for this thread) that makes them change is CAS (and/or primaries in general) and *possibly* the 2:12 ratio. I'll take a note anytime I see something change it from now on and will try the 2x1GB and 2x2GB to see what impact that has on them too...I don't *think* it makes a diff, but it's not like I haven't been wrong before

I did 3Ds (except 01 and AM3) on the 870+UD4P in single and SLI to see scaling. I'll post them in the next day or two, but it's about what you'd expect: decent(?) gains in 3D03 (50%+ IIRC), not so much in 3D05 or 3D06. If I get some time this week, I want to bench both setups at default system settings too. Same OCs as above, just run the X58 the way it supposed to be (tri-channel, 19x default QPI, etc.).

[Gautam]

I don't recall seeing anything different than 54 on channel A (56 on B IIRC)

Screenies not needed, that's good enough for me. Just wanted to make sure.

massman thinks maybe newer BIOSes might be optimized better (and likewise a C0 on an older BIOS would perform poorer)

I think what you've found is consistent with what everyone else is getting on P55...it's just not clear why.

[Porvalsh]

Great thread, I was just thinking about the intel line-up today and with the price drops, their performance to price ratio is much more inline with AMD. Its hard to pick AMD now, since the system components are much the same, but perform better.

[Ross]

Originally Posted by Gautam

Screenies not needed, that's good enough for me. Just wanted to make sure.

massman thinks maybe newer BIOSes might be optimized better (and likewise a C0 on an older BIOS would perform poorer)

I think what you've found is consistent with what everyone else is getting on P55...it's just not clear why.

OK, I will at least verify those are the RTLs being used and if anything changes, I'll make a note and try to find out why.

They probably still want to sell 1366?

LOL, unlike if you put an E8600 on an older BIOS

[Ross]

P55-UD4P, F4r beta is a bit better than the F4m I was using. ~1.5s faster in 32M and +100 on 3D06 (at least it cracked 22K). Mem AUTO secondaries are visibly different (tighter). I'm not sure of anything internal with F4r, but it's a small step in the right direction.

[Ross]

More info on the front page