Intel enters Round Two at 45 nanometres
First INQpressions One QX9770 vs Dual X5482 - The Duel at 3.2 GHz
Products: Intel QX9770 and X5482 processors
Website: www.intel.com
Find at: your local Intel distributor
Price: around US$ 1,400 upward per piece
2-socket board: Supermicro X7DWA
1-socket board: Asus Maximus Extreme
AS MENTIONED a few times here, Intel is continuing its high-end domination with the second round of 45 nm Penryn parts, both running at 3.2 GHz off fast FSB1600 - first on biggie workstations, then on the high end PCs. These fancy chips are worth more than their weight in gold right now - keep in mind they do depreciate quite a bit over time, contrary to the precious yellow metal.
And, as you saw earlier this week, the chips have quite a bit of headroom to play with - even as a daily-use 4.5 GHz system.
We ran both of the same-clocked offerings, the X5482 dual socket Xeon and QX9770 single socket Core 2 Extreme, on a set of 64-bit benchmarks comparing the per-thread performance of identical cores with varied memory and chipset surroundings as well as multi-thread scaling. No overclocking, no gimmicks - just pure default speed, of course with voltage and memory adjustments where appropriate.
So, both CPU and FSB clocks are same - an apples to apples comparison, as much as one can be these days.
The dual-socket workstation platform was the same old SuperMicro X7DWA used for the world's first 3 GHz X5472 Harpertown test we did just before the recent Frisco IDF, upgraded to X5482 in just over 10 minutes - half of it spent in fine applying of fresh thermal grease.
It's a full size Seaburg board with dual PCI-E x16 v2 slots and 16 GB FB-DIMM DDR2-800 CL5 Nanya RAM running at a cool 1.5 volts, plus an Adaptec SAS RAID 1 with twin mirrored Seagate 146 GB 15K rpm HDDs. The standard ATX size Asus Seaburg board, which I originally hoped to test the new chips with, is unfortunately delayed. Finally, the graphics was powered by Leadtek Nvidia Quadro FX5600 GPU surrounded by 1.5 GB memory.
The single-socket "enthusiast PC" config used, again, the same Asus Maximus Extreme X38-based mobo we took for the most recent QX9770 overclocking review. Both systems ran WinXP SP1 64-bit; yes, the 64-bit SP1 version seemed more responsive either way compared to the 32-bit XP SP2. On the 64-bit benchmark side, there was Sandra XII for synthetic stuff, PovRay 3.7 for some ray tracing, then CineBench 10 for more of the same.
Now, the maturity of Seaburg FSB1600 platform and its BIOS showed: the X5482 was auto-set to a low 1.21 volts Vcore, while, on the X38, somehow the default QX9770 setup was 1.33 volts for exactly the same clock and FSB - compare it to the 1.21 volts for the QX9650 on that same board, and something isn't right.
However, look at the CPU-Z 1.42, the brand new versions - seemingly, it " discovered" a voltage level that could burn through any modern CPU here! Oh boy, over 4 volts! Either this is a super-rugged CPU, or some voltage reading fixes may be necessary, I guess.
Interestingly, the SuperMicro BIOS, after changing the CPUs and applying Zaward's high conductivity HSC-W thermal compound, reported unbelievable temperature reading for - air cooled - twin X5482 Xeons. One was at 12 C, the other at 16 C! Supero Doctor, the vendor's Windows monitoring tool, reported the same. When running PovRay, the temperatures jumped to 23 and 26 C, still too low to be true. I ran Sandra XII temperature monitor instead, and, well, the readings were more real: 43 / 45 C in Windows idle, and up to 61 / 63 C under full load - still quite a feat for an aircooled setup.
Here are the benchmarks:
Bench X5482 QX9770
---------
Povray 1CPU 653.4 654.1
Povray allCPU 4882.8 2465.3
CineBench10 1Cp 3930 3954
CineBench10 AllCp 24413 15372
SandraXII CPU 106634 57362
SandraXII FPU 87144 44975
SandraXII MMint 695538 352103
SandraXII MMfp 542186 275311
SandraXII memInt 8930 9074
SandraXII memFP 8913 9087
Both the 8-core Harpertown Xeon and the 4-core Yorkfield provide unmatched performance across the (benchmark) board. And there is more to improve: put those upcoming Skulltrail CL3-3-3-5 FB-DIMMs to feed the Harpertown, or DDR3-1800 CL6 memories onto the QX9770...
In summary, together with the uni 4.5 GHz QX9770, the dual 3.2 GHz X5482 is the fastest machine ever reviewed by The Inq. The huge CPU numbers are matched by vastly improved memory scores for more system balance, a further warning to AMD that they may lose even the last performance hold out - the memory - even before Nehalems come in.
Even before that, in a month or two in fact, the QX9775 "unlocked" version of the X5482, will be making its rounds on the SkullTrail platform for enthusiasts and overclockers. Keep in mind there may be more than one 'suitable' board platform to run these at, say, dual 5 GHz supercooled clocks - Taiwanese are working on gaming dual-socket Xeon boards too.
If willing to spend the top dosh, you can have these CPUs real soon now - and if on budget, in a few months the Q9450 Yorkfield will give you a good feel of this performance in a mainstream PC anyway. µ
Good: Absolute top performance, at least till dual QX9775 comes on a Trail of Skulls
Bad: This is a very expensive toy, mind you
Ugly: I'd like it with silent liquid cooling, rather than listen to a jet turbine fan orchestra
Beers: 8

Comments
Temps
You reported 12C for air cooled processor.Where did you test it?
Standard room temperature is 20C, how air cooled processor can be lower than that?
Re: temps
Pretty sure they were just pointing out that the motherboard BIOS's temp monitoring was giving bogus figures. He said later on that in Windows he got temps of 41 or so.Wait..
8-core CPU?.. + caneland anyone?
32 cores in one box?
I'm wet?
Fool
Actually read the bloody article, Mr Smart Alec - sorry - Gerilart, and you'll see the real temps:"I ran Sandra XII temperature monitor instead, and, well, the readings were more real: 43 / 45 C in Windows idle, and up to 61 / 63 C under full load - still quite a feat for an aircooled setup."
Simple "experiment"
Have you ever tried pointing a floor fan at your face before, Gerilart?Umm...
Wait until you read the WHOLE article, before commenting...4 volts?
i think 4 volts would fry the processor as seen on cpu-z screenshot4v wouldn't just...
4v wouldn't just burn the CPU up. You would see a mini-nuclear explosion coming from your case, blinding you and then melting into the ground causing the China syndrom.Other than that this looks awesome. If I could afford this I would get them. Well the Skulltrail though. I just imagine the ownage of a dual socket setup running even at 4GHz each with 4GB ram each and a quad video card setup. So much ownage.
RE: Temps
Gerilart wrote:"You reported 12C for air cooled processor.
Where did you test it?
Standard room temperature is 20C, how air cooled processor can be lower than that?"
Maybe if you had read the whole article you would know how that was possible... dumbass!
oops..
I misread something, no 32 cores, yet.