I'm trying to figure which bencbmark best represents CPU performance. It seems like most are affected to various degrees by RAM, GPU, mobo, and HD. For instance, different sources gives a WinRar score for FX-62 of 621 and 1145 kb/sec. Obviously, differences in the system minimize the value of the benchmark. These effects should be eliminated as much as possible. On the other hand, overclockability and the multi-threading ability of dual core are integral to the CPU.
I understand that usage (ie games, floating point calculations, encoding, etc) is an important factor in which benchmark is best. I'd like discussion on this. Games are probably the most important use of CPU extreme speed. But game benchmarks apparently depend greatly on the GPU, and so have limited usefullness in comparing CPU quality.
Hopefully, if the best benchmark can be decided, it will be used by many sources creating a large database of the performance of many CPUs over a considerable period of time.
My own preliminary thoughts is that CPU performance depends on such factors as clock speed, special CPU instructions that reduce the number of assembly language instructions, and internal cache size and speed.
Which of the many benchmarks best represents CPU performance?
yep. There is no one tool to check everything. It would be nice as you indicate but that is not real world testing. We all use many different programs and utilities in our everyday computer usage so using many different testing tools best emulates real world environments.
There are three problems with using lots of benchmarks.
First, it complicates things greatly. THG is currently using 38 benchmarks. Huh? Which should I use? How do they all fit together? While the CPUs often maintain the same order, sometimes they don't. When and how is this important?
Second, it doesn't neutralize the variability the other components introduce, as mentioned above.
Third, this makes price/performance determination difficult. THG arbitrarily uses a 50% game / 50% application mix. But is that really a good choice? This thread can discuss this.
What is being suggested is that which benchmark you use depends on how you use your computer. What I'm suggesting is that the current system is so complicated that it is not useful. Maybe a good guide on when to use what would help (like use low-res BMs to minimize GPU effects). But when do you look at Quake, or Sam, or 3DMark06-Graphics, or look for Far Cry BMs. And how can you realistically combine them to get a single useful index?
Thanks for the referral to PC Wizard. A very comprehensive sys info program.
But if you throw at me one single bench that states it all --- well you are doing a very serious disservice to youself as well as others because certain architectures are better at certain things....
The premise of your argument is a onesize fits all, and this is sorta not possible. I do agree though that it is inappropriate to develop a price/performance curve from an arbitrary data set and draw conclusions in a singular fashion.... Tom's or Xbitlabs price/performance charts are fundamentally flawed. Jack
Good points. I guess I want a single number to fit my digital mind.
Actually, we do use one-size-fits-all to an extent. A size 8 shirt will imply a certain neck, chest, sleeve length, and sleeve diameter. Manufacturers found this simplification necessary and useful. I think some kind of similar simplification would be useful.
It seems to me that the only good way to choose a CPU is to generate some kind of price/performance curve. The question is how to make the performance index most realistic for me.
OK, I'll ask a specific question. I do a lot of Video processing. Should I use the THG benchmark: 3DMark06-Graphics, Divx, Lame, Mainconcept, Media Encoder Streaming, Photoshop CS2-Rendering, Photoshop CS2-Converting, Pinnacle, Premiere, WMA, Xvid, or somehow combine these to find the best value CPU. I have not seen any guidelines for this. How can I, and others, make all this information useful?
Yeaaaa for AMD; Boooo for Monopolie$
P4D's were also much cheaper than the X2's, did this make them a great value??
Actually, I'll admit that I recommended somebody to buy a Pentium D 820 system about a year and a half ago. The guy was a CAD user but didn't have all that much money to spend on a new machine- his old one was a Pentium II 350 for crying out loud. An X2 3800+ would have blown the D 820 out of the water, but the X2 3800+ wasn't introduced yet, and the X2 4200+ was about $500. The Pentium D was about half that and the same price as an A64 3500+. The D 820 would outperform a 3500+ handily in the multi-threaded CAD app the guy was using, so I knowingly suggested the D 820. He was and still is pretty happy with the machine and since it runs well enough, I think I suggested decently. The fan can get noisy, but it's not that bad.
I almost made a Pentium D machine for myself. I built my machine in early 2006 as my old laptop wasn't working well as my only machine for a good while and I was getting antsy. The Pentium D 900s had just came out and so had the very first of the Conroe benches. I had seriously debated buying a D 920 and a good 975X board and then putting a C2D in it later, but I decided upon a reasonably-prices 939 board an and X2 4200+. I'd mostly done so because I'd have hated to throw away a $250 D 920 that worked perfectly fine before it was so horribly outdated. (I'm still running a P4-M 2.2 GHz laptop. Why? It's only in the beginning stages of dying and not dead yet.) The X2 was a lot better chip than the Pentium Ds were and I simply got the best that was there at the time.
The Pentium Ds got seriously cheap before the X2s did, so in the latter part of 2006, they would not have been a bad buy. Sure, they were slow and hot compared to the Core 2 Duos and A64 X2s, but compared to the similarly-priced A64 single-cores and Pentium 4s, they were a much better value. But today, I'd not recommend one as a decent dual-core chip in the X2 3600+ is about 75 bucks, roughly as much as the cheapest PDs are.
I'm aware of SPEC and the difficulites of specifying any benchmark. What I'd like to see in this thread (or an article) is tips on when each of the 38 THG benchmarks should be used. I guess I'm asking for a 'benchmark to benchmarks'. If I play Far Cry, is Quake or Sam the better BM, and why. And how about the multiple video and audio choices I mention above. Straight calculations are shown with Super Pi but also Mersenne Prime and Passmark (10,000 systems).