2CPU.com » News » June 2001 » GamePC Multiplies 1.4 GHz x 2!

Terry was kind enough to point out that Chris over at GamePC has written up a piece on the Athlon MP and 760MP. He also ran his benchmarks with 2 1.4 GHz Athlons in that bad Tyan Thunder K7.

Amazingly, the Dual Thunderbird Athlon dropped dramatically in this test, and even while the test is multithreaded, came back with a score less than that of a single processor Thunderbird system. Truly, something isn't right here. Unsure of my own testing, I completely re-did our testing systems from scratch (freshly formatted and what not), and came back with the exact same numbers. Something tells me there's a serious multiprocessor issue with the Thunderbirds, which is why AMD pulled "official" support for them at the last second.

Interesting to say the least. This trend seems to continue throughout the review in certain benchmarks. You'll probably want to take a peek.

Jason_Dickey SMP Veteran Posts: 510 Joined: 2000-06-25

#9614 Posted on: 06/20/2001 03:24 AM Seems like the Thunderbird has a high penalty for a single-threaded task switching processors (as Win2K normally does dynamically swap the task between processors). Perhaps if a Processor Affinity was selected, the scores would improve? I would really like to see the benches run again with processor affinity selected, and perhaps a look at the task manager would be nice as well.

voltron Master Dualie Posts: 1254 Joined: 2000-12-22

#9615 Posted on: 06/20/2001 03:38 AM I thinks omething is afoot as I saw a benchmark months ago running off of regular tbirds in dual. AMD or Tyan crippled it on purpose. I am losing respect for AMD. Would soemone be kind enough to test older 200mhz fsb athlons to see if there is a difference between the 266fsb ones. So try 1ghz or more and if you have to overclock do so. MSI K8N Neo2-F, bios 1B | 3800 64 X2 | 2x512MB Kingston 2700 cas 2.5 unbuff | 3950U2D | 10K 70GB | ATI Radeon 800XL 256MB AGP | Viewsonic VX800 | Audigy2 | 3Com 10/100 | Mad Dog Surepower 550Watts | Windows XP Prof SP2.

grindles SMP Duck Posts: 370 Joined: 2000-06-03

#9616 Posted on: 06/20/2001 04:34 AM I expect it is just a cache issue, hence the improved cache protocol type thingy on the MPs. AMD still rock in my opinion.

ReMeDy SMP Duck Posts: 12521 Joined: 2001-02-10

#9617 Posted on: 06/20/2001 05:11 AM Yeah i wonder how it would react with 512k cache on it. So is it safe to say the bottleneck is coming from the cache itself? |[size=1][url=http://forums.2cpu.com/showthread.php?s=&postid=279831#post279831][color=yellow]FAQ's[/url][/color][/size]|Heatware-Remedy{WcS}|Dual760|Beerology|

grindles SMP Duck Posts: 370 Joined: 2000-06-03

#9618 Posted on: 06/20/2001 03:58 PM I dont think the cache is a bottleneck in itself, I think it is the cache coherency slowing it down, the original thunderbirds dont have the same thing to keep the two caches coherent (MOESI??), so they dont stay correct all the time, leading to big latencys as everything goes to the RAM. Maybe... I could be wrong, but it seems logical to me.

grindles SMP Duck Posts: 370 Joined: 2000-06-03

#9619 Posted on: 06/20/2001 03:59 PM Here is an interesting idea for Hooz, or whoever has the tyan, put some thunderbirds in, and do benchmarks with and without cache enabled, and with one and two CPUs. I bet you get a performance increase when you go to 2 CPUs with the cache disabled.

Jim SMP Newbie Posts: 0 Joined: 2003-12-16

#9620 Posted on: 06/20/2001 05:46 PM Originally posted by grindles I dont think the cache is a bottleneck in itself, I think it is the cache coherency slowing it down, the original thunderbirds dont have the same thing to keep the two caches coherent (MOESI??), so they dont stay correct all the time, leading to big latencys as everything goes to the RAM. Maybe... I could be wrong, but it seems logical to me. MOESI isn't something that's processor-specific. It's the coherency protocol used in AMD's (see DEC's) SMP implementation. I highly doubt that's the problem in this scenario. I could be wrong, though. I'll drop the guys at Aceshardware a line on this issue. Maybe they can shed some light.

mckennma Unregistered

#9621 Posted on: 06/20/2001 06:34 PM Can't someone come out with a Dual server board with no problems? It seems like every board since the i840s has had major performance or technology issues.

dual_celeron SMP Hobbyist Posts: 2559 Joined: 2001-04-19

#9622 Posted on: 06/20/2001 08:18 PM I don't think cache size has anything what-so-ever to do with the performance. Either AMD has crippled the regular Athlons or AMD has crippled the Tyan bios. My dual Celerons only have 128k cache each and they work just fine together. It's the implementation. Didn't one of the testers use a revised or customized bios on the Tyan board and run dual Durons with success? Everything from an old P90 to dual AMD XP2000 and Sempron 2400. Run an HTPC in the living room. Laptop, wireless networking all over the place at home & work.... blah blah blah

Attiq ASMP -Bring it on! Posts: 27 Joined: 2001-06-17

#9623 Posted on: 06/21/2001 02:10 AM There is saying when your aiming too high, you will always get it wrong at first, my advice is have faith and let AMD correct thier flows, after all we all need thier cheap alternative to Intels Megabucks solutions... Enterprise 512k is The machine to die for, got it off a bloke who almost died when he was given one.

mckennma Unregistered

#9624 Posted on: 06/21/2001 05:05 AM All manufacturers rush their product to market to be the first on the block. This usually causes problems. Testing is forsaken in return to being the first. Tyan has done this over the last 3 years and has always come up short. The Tyan Thunder i840 was a great example of Intel and Tyan both rushing. I have several servers and my workstation to upgrade. I have been waiting for over a year for an excellent server board. I have been reviewing motherboards over the last year. I do not care about the cost. I just want a board with few problems and maximum expandability.

2CPU.com » News » June 2001 » GamePC Multiplies 1.4 GHz x 2!