AVX will give performance gains approaching 100% in a CPU-limited parallelisable FP-heavy application such as media encoding/transcoding, or applying photoshop filters. For branchy code like raytracing, probably more like 30-50% improvement, and that's assuming you're using double-precision...
I'm certainly glad it has 6 slots. My current 8-core Xeon workstation has an Asus 6-slot board with 12 gigs of RAM, and I easily use them all on a daily basis (testing machine learning algorithms on large datasets). When Nehalem comes out hopefully someone will make a 12 slot DP workstation...
I gather you're doings lots of floating point calculations. It makes a critical difference whether those are classic x87 floating point instructions, or SSE instructions. Pentium 4s suck at x87, but tend to beat Athlon64s on well-tuned SSE code. If you're compiling with Visual C++ and not...
The P6 design has always run 16-bit x86 code natively. In fact all x86 processors have had hardware level backward compatability without any sort of translation required. AFAIK the three main reasons the original P6 didn't work well on 16-bit code were; (1) the address generation and load units...
If you really have the money to burn, try this: up to 8 dual-core Opterons, 128 Gigs of memory, twin PCI Express and 4 way SATA raid in a standard tower case!
If you want a dual-processor system with FX-57 single-threaded speed, the simplest solution is just to build one with two Opteron 254s, which are available now and run at 2.8 GHz out of the box. I agree that overclocking a 4800+ to 2.8 GHz will have similar performance at much lower cost, though...
The two kinds of processor should look more or less identical to the software at the application level, so I'd expect problems to be similar on both platforms. This is just lousy programming; devs should've tested the game on multi-cpu machines when it was first written. To be fair though, the...
Run two simultaneous instances of SuperPi (16 or 32M). Average the times. Actually it would also be interesting to see what happens when you run four instances, so that all the processor resources are contended between two threads.
Not if those 'real applications' are single threaded. The 'real applications' I mostly run are heavily multi-threaded and thus see the same sort of improvement that those benchmarks do (in throughput-limited usage scenarios). Furthermore NUMA does not affect the total memory bandwidth available...
GamingGeek, you are partially correct in that data is not interleaved across all four banks in the way that data is interleaved across two banks in a dual channel configuration. Thus for a single threaded application, no performance benefit will be realised unless special coding is done down at...
Gamingphreek, I write applications for Opterton servers for a living and I can assure you that they do have 2 DDR channels per processor, i.e. 4 in a 2 processor box and more in larger systems. For two-way systems the hypertransport bandwith easily exceeds the maximum dual-channel memory...
Personally I'm planning to build a dual-dual-core workstation around Christmas, with two dual core Opterons on socket 940. 4 memory channels negates the bandwidth advantage of DDR2, and it looks like S940 is going to stay current for a lot longer than S939. I do a lot of compiling and...
Scsi: whether the cores are synchronous or not makes absolutely no difference to their power dissipation. The pentium and athlon dual core processors are synchronous for this reason. Perhaps you have confused this with asynchronous logic, which is a technology that really does have the potential...
The Cell cores will be synchronous simply because they're all driven by a single external clock generator; it would cost more and perform worse if they were run asynchronously.
'SoothingRelease', your claims are not even remotely plausible. Here is a rough analysis of the theoretical capabilities of the two processors.
Both processors achieve the bulk of their floating point performance by vector operations on 128-bit registers. Normally this means processing 4...
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.