- Saturday, 21 November 2009
- INQ Mobile
- RSS
The noblest of all dogs is the hot dog. It feeds the hand that bites it - Hot Dog Restaurant
Xeon W5590 breaks records - Part 1
LAST WEEK Intel announced speed upgrades to its Nehalem-EP Xeon lineup, enabling the new CPUs to match their top end desktop counterparts' clock rates. So, the dual CPU Xeon W5590 runs at 3.33GHz plus whatever Turbo speedup you get (see the 3.6GHz at CPU-Z), and the single CPU Xeon W3580 runs at exactly the same speed.
The dual CPU Xeon W5590 part is reviewed here, in an interesting combined configuration: Tyan S7010 new mainboard, instead of the old Supermicro, and the brand new updated Asetek LCLC cooling kit with dual CPU support and twin radiator to cool all the CPU machinery, as well as the new Asus Matrix GTX285 OC card, with a Thermaltake 1500W PSU powering everything. Also, the six total channels of 48GB ECC server RAM now run at the full DDR3-1333 speed across all 12 DIMMs simultaneously, thanks to the Tyan board's manual speed override capability.
The Tyan mobo also has a buit-in SAS disk controller as well as integrated graphics if you want to use it as a server. The only complaint is the absence of the second PCIe X16 slot for dual GPU capability.
The BIOS screen looks like a nerd's wet dream:
So yes, it would make a dream desktop. Other vendors, like Asus and Supermicro, have mainboards that would even support Crossfire or Quadro SLI for dual GPUs, if you feel like caching Google Earth in memory and playing with the whole Earth in interactive 3D at deep zoom level.
But then, remember, this is a high-end workstation setup, meant for complex 3D CAD and engineering modeling, virtual reality design, IC simulation or computational science - almost a personal supercomputer with near 100 GFLOPs performance in Linpack benchmarks.
In this initial quick review, I ran a couple of Sandra benchmarks on CPU, multimedia and memory throughput and, guess what, these are all record-breaking results, whether in MIPS, MFLOPs or other metrics:
Note the memory bandwidth - for the first time, we have breached the 40 GB/s mark. In fact, if the BIOS allowed latency setting changes and we used desktop DDR3-1333 CL6-6-6-16 modules, we'd easily exceed 45 GB/s here! That's one point we can check off our wishlist.
Either way, this is a firm leader on the workstation front right now. AMD's Istanbul would need to have its cache and memory systems beefed up, in addition to its core clock speeds of course, in order to come close.
In the second part of this review on Monday, we'll examine the Asetek LCLC cooling performance when handling all these benchmarks at continued 3.47GHz speed across 8 cores and 16 threads, plus several more tests. For now, enjoy these initial record-breaking results! µ
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
How much?
[i]But then, remember, this is a high-end workstation setup, meant for complex 3D CAD and engineering modeling, virtual reality design, IC simulation or computational science - almost a personal supercomputer with near 100 GFLOPs performance in Linpack benchmarks[/i]
Cool but...i have about 2 TFlops on my home pc, not counting the CPU of course. When will developers turn to GPGPU ??
Proper cooling is not executed with fans on the memory. Simply place a 24" boxfan so it blows perpendicular to the motherboard! That's how I do it! :P Seriously you will get incredibly improved temps.
I can assure you that a lot of mechanical engineers are drooling over this machine. Modern 3D modeling tools don't use GPGPU's. Programs like Pro/Engineer, SolidWorks, SolidEdge or Inventor mostly need super fast graphics, heavy memory bandwidth and all the memory capacity you can throw at it. SolidWorks recommends "x64 processor with 6GB or more" for the larger 3D models. It's quite difficult to even use all the 8 cores for 3D CAD but most mechanical engineers also use finite element modeling for stress analysis. Until GPGPU's can be used for FEM this machine will be the tool of choice for at lot of professionals.
Best regards
Tomas
Nice stuff too bad about SLI MIA, not one but 3 extra PCIe slots are missing. Convergence of CPU&GPU cycles and all. But yeah, this box should be able beat a GF260c216 on BarsWF handily, yes yes.
How it coordinates two processors into one 16x slot, if it does, for mere 6 grand. Usually theres primary side for graphics with secondary side. This might be new. If it configures automatically, great.
NEXT: also old as yesterday is HTML 5 standard here:
HTML 5 in its current draft form includes a number of significant advancements, notably several that make the Web a better foundation for applications, not just static Web pages. Among the present HTML 5 features are built-in video and audio, the ability to store data on a local computer to enable use of Web applications even when offline, Web Workers that can perform computational chores in the background without bogging down Web application responsiveness, Canvas for creating sophisticated two-dimensional graphics, and drag-and-drop for better Web application user interfaces.
Both apple & google claim to have already established html 5, so Microsoft wants industry standard set, so now blogging to that end. Sooner than previously reported.
neither 6 grand workstation or new html standard are guarentee of stable intenet connection to browser. CABLE is better. Remember $600 Cobalt modem from 1998? People are desperate for reliable connection, yet even today, modem & proxy server setting are loss to browser, refusing to find webpages while still actually on internet..its' continuos mess, what good is such power refinement if its stuck in mud? Forget reliable speed with dsl, too. Seems Hardware & software have advanced, yet missed entire point.?Workstation?
vondrashek
How much? How many KW?
I have a lowly Q9650 with 8GB ddr2 and I am running 24/24 - 7/7 a turbulence simulation code since last February and probably finish by December next.
I could do with something better.
GPUs are not much help for code dominated by serial calcs.
after your recent SSD desaster, even some more record-breaking sandra results! really can't wait for the second part, that stuff gives me chuckles every time.
To the GPU lovers ...
For the moment, you was given example of GPGPU usage that are corner cases, as long as you don t see a Spec_fp_rate score, this mean that the GPGPU is UNCAPABLE of running it ... then, it means that the GPGPU can not run the workload chosen by the industy to calculate the best processing machine ...
so, stop "bothering" the medias with corner cases running CUDA, and start providing Spec_Fp_rate scores, without this, you are just promoting something that IS NOT a generic solution and will end up as the MPACT! card from Chromatics ...
PS: This is my very personal opinion, prove me wrong, show me the Specs_fp and spec_int score, then you can speak again.
Like the Internet in 1992, virtualization in 2000, the GPU computing power is there to take over computing industry. Those that adopt it early reap the early benefits of it.
As for Intel, go make a Larrabee with crappy drivers, produce it in numbers before DNF comes out. I remember i740 and the cripple ware that comes with a netbook.
The CPU as we know it, is dead, long live the merger. And lets not hope the CPU/GPU goes the way of the soundcard, 'cuz that would suck.
But that, is my personal experience and opinion.... Just keep on doing the great job Francios, no hard NV feelings here.
Intel sucks.
Tyan makes excellent mobos mostly for the server market. So most of their stuff does not do SLI or Crossfire. Instead you plug great big hardware RAID cards into those PCIe slots.
So it's Tuesday. Where the hell is part 2?
Get an education.
The comment about that Tyan board is incorrect. The Ram/QPI over ride on that board doesnt work. It will still defualt to the standard settings for the cpu's installed. Also that board has a bios issue with Nvidia 295 GTX cards. When you get to windows half of the system devices wont load because of a resource conflict likely due to the Dual GPU's. Tyan is aware of the issue but has not come up with a fix yet.
Intel's website says that the W5590 does not support turbo boost (and
doesn't mention anything about whether it supports hyperthreading). I
emailed Intel presales and called them and both times, they
confirmed that the W5590 does not support hyperthreading. Are you sure
it supports turbo boost, and do you know if it supports
hyperthreading?
See http://ark.intel.com/Compare.aspx?ids=37113,41643,