Dual AMD 1.2ghz Palomino's Benchmark :(

[NDC]

Take a look here:

http://www.2cpu.com/Hardware/thunder_k7/cpu.html

That's pretty weak if you ask me! Look at the ALU Could this be correct on the CPU benchmark?????? The Multimedia benchmark looks better though...


My Dual P3 @ 935Mhz get's higher than that on the CPU benchmarks...


<IMG SRC="http://ndcmj.hihome.com/Benchmark_VP6_CPU.gif" border=0>

<IMG SRC="http://ndcmj.hihome.com/Benchmark_VP6_MMedia.gif" border=0>

[jaida]

I think those benches are not too bad considering the 760 mp chipset is very new and still has some bugs to work out like any other new chipset. The multimedia benchmark is quite impressive but also the memory benches that ddr performance is excellent on that board.

[jad1097]

Speaking of WAV files, I just wanted to add that all of the problems I have been experiencing with my Live MP3+ 5.1 card on other motherboards has thankfully vanished. AMD's Viper southbridge is a leaps and bounds ahead of VIA's 686B and I never had a hiccup with the Live in this machine. You know it's stable if Liveware doesn't crash it ;-)

If you simply must have the latest and greatest that there is... This is it! This board is rock solid, loaded with features, and (according to AMD) has a bright, upgradable future. If you can afford it, this is definitely the cream of the crop. I never experienced any of the "normal" bugs and quirks common to a new platform introduction. AMD and Tyan played their cards right this time, and their first step into the workstation/server space is a big one without a doubt.

[NDC]

Actually, the benchmarks didn't impress me one bit considering the CPU's are at 1.2Ghz FSB 266. The ALU is 4208 on the dual and 3366 on the single. How much of an increase is that from a single? Less than 25%! And ALU is what's going to make the most difference on most applications, right? Not FPU which is used mostly for applications that require calculations. And yes, the memory benchmarks are quite impressive, but nothing you'd say "WOW! That's a BIG difference!" in a real computing environment, just benchmark bragging rights. As for the MP760 chipset, it's not something they made yesterday, it's been worked on for years as I know. I'm wondering if this is something AMD can straighten out in the coming months.


However the RC5 crack rate was just outrageous! Thing cracks at 8.44 Mkey/sec!


I was actually considering forking out some cash when the price dropped on this dual AMD setup. But after looking at this CPU benchmark, I have second thoughts about building myself a dual AMD, for the time being. I certainly hope they get this straightened out! I'm dying to build one of these systems as my primary workstation.

[This message has been edited by NDC (edited 06-10-2001).]

[Mntsnow]

I too was saddened to see the lackluster performace of the AMD dually in the CPU benchmarks. I hope that it will be figured out and corrected by the time the boards come down the the "middle of the road" consumer price area as it would be nice to have a couple of 8 to 10 Mkey cracking machines under my desk!

[Dputiger]

My hope is that the lackluster performance of the 760MP may well be corrected by driver/BIOS issues.

After all, we've seen Pentium4 performance jump DRAMATICALLY just with driver changes--why can't Athlon4 do the same in a multi-processor environment. I'm betting it can.

[OuTpaTienT]

Ok, I have serious question here. Maybe a little off the cuff, maybe I'm missing something, but I want to know how it's even possible for the Athlon MP to have such poor Sandra test results and at the same time have such great test results in the numberous tests run at Anandtech ( AthlonMP review (http://www.anandtech.com/showdoc.html?i=1483&p=11) ).

If maybe Anandtech just ran a few benchmarks then I'd be a little more hesitant, but hell, I see at least 19 different graphs of results and the Athlon is ruling most of them or at least in a respectable position.

So is something wrong with Sandra? Or is Anand lying to us? Almost seems like it's got to be one or the other.


[This message has been edited by OuTpaTienT (edited 06-10-2001).]

[Bob The Great]

How much memory, and how fast is it? Also were you running any programs in the background. Some times programs like registration reminders and such are built really badly and take up allot of your systems resources.

[eagle1]

The thing is that those benchmarks from anandtech are testing everyday network and those "kind" of apps performance. You don't buy a dual system for gaming or for Windows 98, you buy them for network servers to manage databases, etc. I trust more the anandtech "real performance" vbenchmarks than those of Sandra... (at least for dual systems).

[NDC]

Well, I wouldn't doubt 2CPU.COM benchmarks either though... In any case, I guess we shall all see soon enough when people actually start using these systems. But dang! $600 for the Tyan mainboard's pretty steep! http://www.sysopt.com/forum/frown.gif

[OuTpaTienT]

It's not that I don't trust 2CPU.com, all they did was run a Sandra Sisoft benchmark, heck I can do that. But I've ALWAYS been suspicious of Sandra scores. It's not really the best benchmark in the world.

I would definitely require more data than Sandra results before I go raking AMD over the coals.

[eagle1]

Yep.. that's what I meant in a way. I think AMD is doing a good job in the proccessor market and if anandtech benchmarks are any indication of real work performance, let me tell you Intel has to get a heck of a gameplan to stop AMD from taking over.!

[RobRich]

Sandra is perhaps the worst synthetic processor benchmark currently available for the Windows platform. SiSoft does stress L1 cache bandwidth and reigster move operations, but the system itself does not prove to be a factor for the scores. This is definitely not acceptable when trying to benchmark a processor, especially for a SMP system when cache coherency plays such an important factor into overall system performance.

I've used Sandra within some of my articles, but you will notice I've discontinued use of the processor benchmarks. To contrast, Sandra does offer a great memory benchmark due to its derivative of the STREAM codebase, but that is about the extent of Sandra's benchmarking capabilities.

To better analyze processor performance, I generally recommend taking a look at real time numbers, such as media encoding, software 3D rendering, or application compiling. If benchmarks are to be utilized, then I usually look towards professional solutions, such as specInt or specFP. Or better yet, try running SPECveiwperf through a software OpenGL rendering engine, preferably SGI-based.

Check out this review article at AcesHardware:

http://www.aceshardware.com/Spades/read.php?article_id=30000184

Notice how the Athlon MP performs in a real-world workstation testing enviroment. I also remember a nice article being done over at Tom's recently about how the Athlon MP and P4 Xeon compare, might be worth checking out for those interested.

Catch ya' later,
Robert Richmond

[NDC]

Well, I honestly do hope that's the case with the poor benchmark on the AMD duals at 2CPU.COM. I think the guys at 2CPU.COM were wondering the same thing and retested to make sure, don't you think?

Here are some benchmarks that were done on Sandra, and they do show consistency. It appears that roughly each 100Mhz = score of 300 on the ALU benchmark... And to see that the dual scored only 4208 using sandra2001 just blew my mind.

How would you explain this consistency from 4 different AMD processors using Sisoft Sandra 2001?

Barney - 1.2Ghz = ALU = 3314

AndreRIO - 1.3Ghz = ALU = 3642

Mntsnow - 1.34Ghz = ALU = 3702

NDC - 1.4Ghz = ALU = 3925

http://www.sysopt.com/forum/Forum20/HTML/000003.html


I don't want these numbers mean, but from seeing consistency in calculations, I think Sandra does give you an idea... What do you guys think?

[doctj]

2Cpu's review is garbage in garbage out. As a scientist myself I know that if you choose the wrong test for an experiment you'll get the wrong results no matter how many times you repeat it. Like Rob said, SiSoft Sandra is not sophisticated enough to benchmark an SMP system. I'd never rank those results above Ace hardwares excellent real world benchmarks.

[NDC]

That still doesn't anwer the question I above. "How do you explain the consistencies?" Perhaps pure coincidence?

2Cpu's review is garbage in garbage out.

Really? That's pretty much the best site I know for SMP resource... Do you know of any better SMP sites?

Looks like Tom @ Tom's hardware Review site didn't have such a great review on the 760MP either. At least it looks that way from what Dputiger says...

http://www.sysopt.com/forum/Forum3/HTML/011613.html





If Sisoft Sandra was such a crappy bencmark program, I just don't see how you could see this consistency... OK, Sisoft isn't a professional way to benchmark CPU's, and may have it's disadvantages... But like I said above, it give's you pretty good darn idea...

SiSoft Sandra is not sophisticated enough to benchmark an SMP system


Looking at the comparison between my Tyan dual @ 897Mhz and Abit Dual @ 935Mhz and the CPU benchmark score difference of 205 seems to make sense to me. If the benchmark on the 897Mhz was higher than the 935Mhz, I would definetly say that Sandra is flacky.

Take a look for yourelf:

http://www.sysopt.com/forum/Forum20/HTML/000002.html

The reason Mntsnow's VP6 @ 953Mhz shows lower scores is because services were running and was connected on the net as a web server. That would definetly give you incorrect results on CPU benchmark regardless of what util you use to benchmark the CPU. Even browsing the net while running a CPU benchmark program will give you lower scores despite the fact that very low CPU utilization takes place when browsing the Internet. If it was running clean, I'd definetly expect higher scores using Sandra2001.

Whatever the case, thank you all for posting your opinions on this thread. http://www.sysopt.com/forum/smile.gif




[This message has been edited by NDC (edited 06-11-2001).]

[Hellmund]

I think what they mean is that Sandra is consistent with single processor benchtesting but it inadequate to properly test SMP.