Sign in with
Sign up | Sign in
Your question

My own idea base on acehardware

Last response: in CPUs
Share
April 23, 2003 1:46:38 AM

http://www.aceshardware.com/read.jsp?id=55000251
http://www.tomshardware.com/cpu/20030422/index.html
http://www.ebns.com/showArticle.jhtml;jsessionid=P0JL0HYVTGEGWQSNDBCCKH0CJUMEKJVN?articleID=8800528

Unlike to Toms they have use 2 differente kernel.1 for intel that make use of HT 1 make use of AMD K7 and have X86-64 extention.Windows 2003 would have been faster for intel but many enterprise like linux due to low cost and much cheaper upgrade.

They have use almost the same config exception there were faster Xeon avaible strangely Aces have deceide to take 2.8GHZ and Placer.Placer vs Grand champion can be discuce but i dont think there a big difference.Xeon MP are faster on webserver app but slower in workstation.Windows 2003 should have been use for workstatation for bolt platform as it will show the differente between driver and others.

No PCI test as been made as i beat my right hand that Placer from intel is much faster on PCI and AGP (no AGP found in Placer)

No HD test is also made wich can show a 4 15K raid 5 can push on bolt platform.

Like toms time was running for bolt.A full test was not possible.Also a 4 way lantency for ht and memory can be interesting.As like i say before more CPU more the lantency will grow this is dont happen on Xeon but badwith scale on Opteron unlike Xeon.So xeon on a 4 way will have less overall lantency but much less bandwith.Also consider the need in I/O of a 4 way systmes extra bandwith is useful.Grand champion is a quad channel memory controleur the CPU can use only 1/2 but the gigabit and HDD can use the rest leaving all the bandwith possible (not the case on placer do only dual channel) on some case we see from toms RDRAM become faster even with much older platform it a shame that intel dont offer a 4I chipset for RDRAM wich is much faster under HPC (HPC= webserver sientific workload in others word traditional RISC market).

Some may get impress by canterwood with P4 3.0 GHZ unlike most say it the old willi spirit that live in it only software have change around it.

Future performance scaling

Opteron scaling

For sure we will 2.2 ghz higher may come only when intel offer nocona.

Driver will improve a good performace increase can be see in future version.Reason number 1 is AMD driver may be stable up to now but optimization to a extreme level as we see in intel software is not yet possible but time will do it trick.In others moving a car to 100 HP to 200 HP will increase the acceleration much more if we compare that a car that move from 200HP to 300HP.

Compiler wise i dont think linux can pull it out but MS may come to change all of this.
Tool for programing is also need Zero is avaible right now others that dose made for IA-32 so none are made for opteron.Ms can also change this

Chipset wise AMD offer all fearure need but how fast, good question no benchmark are been made to isolate the chipset yet.My own guess that it badly lag behind intel but Opteron is the long shoot like itanium it take more that 2 year to intel finish there own E8875 for itanium (a part of this delay was because original made for RDRAM PC800 that would have been faster).

The package is once again a copy of intel P4 exception ceramic is use vs nickel for intel.Benefit or draw back are out of my knowlege.

Server look well thank to ex IBM engineer.

Support wise

Sun annouce that solaris will be made for opteron but only in 32 bit.I short word this run also on VIA to transmeta CPU so there is still no official support from SUN

IBM annouce they will in the future built server around Opteron.In the same time they have built there own chipset for XEON and say in a very future have Ia-64 capacity, i can guess they are waiting for Madison.

Dell

Dell dont care that itanium or opteron as long as they sold a systemes but any change in there Pure intel strategy may make lose there discount price on P4, leaving Dell to use also AXP in there desktop ligne.They will just lose on this gaining market share in the low-end server but losing massive share in desktop where HP will be very happy but 1% of server market is much pofitable that 1% of desktop market.

Futjisu-Siemes

They have annouce Ia-64 for 2005 (i guess we can speak about that in 2 year)
Also Xeon/Opteron for this year soon.Right now they use SUN CPU.

IBM futjisu point of view is let them fight together. It not a bad choice for the last one but IBM must be sure that this is not touching Power market that seem to start to be affect by IA-64 with the incoming race in server also they get obsolet faster that they think it maybe not a good move as buyer of IBM dont care about the CPU brand they just want IBM reputation.

HP will be not to bad for them to use Opteron too but only to a low-end level they just want the high-end to fall under IA-64.

Software corporation Opteron or any others as long as they sold.For MS intel long reputation was there key for succes as for intel to stick to MS.At the best a Windows server 2003 for X86-64 may come in Q4 if not later.Also what is there plan offer it at all level or only with data base center that cost few 1K $ US.Data base wise all version is offer for all posible version.Also do they will ask a premiun price like they do with Ia-64 version.Does they got a little one they better like.

Position/price of the Opteron

Most of the marketing will go again Xeon and sometime tell to everyone they can beat Itanium.

Intel on others hand will try is best to kill it all is weight wich is pretty heavy.Cheap move may come from intel on on server side.

The big key of opteron is not that 64 is faster or not it just bring some market share so they got a good range software and large base of customer from the A64.AMD will try to bring X86-64 from the low-end and intel bring there Itanium from the high-end realm.

Result speaking

1 test a message board test.

Opteron score

A good start for Opteron much faster but it ligne of total rquest repond move a lot at 140 request the advantage over Xeon getting pretty small around 15% or less when at 110 request it at is peak leaving Xeon in the dust by far.i don see any explication except maybe software or else.in any case that a good win for Opteron.

The same benchmark but with Compress file before sendind

The 1 thing we see is Ht make a decrease in perf i not really buying the L2 story maybe just be the OS or so.
Now the Xeon and opteron switch place often as thing change but at peak request Xeon win.final small win for Xeon

Java Gzip

Once gain compression transaction show Xeon at is best (strong point of P4 architecture stream data inside the Cpu)

Opteron fall short really fast when request increase.This all thank to HT.

Spec JBB 2000 result i strongly suggest that AMD put a TPC submission to compare again others server.

Opteron is much faster that Xeon.As a thing we saw is Opteron perf decrease as request increase like Xeon but the decrease is much more important.

Overall of Opteron show better result that Xeon.Also if i remember well also at anandtech database benchmark AMD was getting good score at Message board so Xeon may have show better result in others test but does can Xeon can win the lose ground to Opteron.I dont think so unless software change.

1 opteron 0 Xeon

Database

Xeon is much faster that opteron.I dont take in account datamining it clear that is not make for P4 at all even a mere Athlon Mp that never win a singlew benchmark is faster i will like to see what happen in here.

i give this win to Xeon

1/1

Sientif workload Sienmark is getting very old but can give some insight how opteronis faster that a Athlon Mp.The plasma benchmark is much more useful it been formely by P4 and I850E that is offer the 4 best time.

It show how Opteron may show a great potential in sientific load it offer a good vector base and a good legacy FPu and very fast memory access.In short word Xeon got no chance i dont think that even 166 fsb Xeon can change this not even nocona but this also one of the best field of IA-64 at is turn even a 4 GHZ may not be faster that a Itanium2 considering that the 3 coming some will just crush i even more but is not the case of all enterprise have this need of power.

On 3D rendering the reverse happen we all se what happen on P4 after the activation of Ht and SSE2 opteron will never be able to mach here but the need of more that 4GB of ram may come a bigger need in 3 year that more power and itanium is not faster that P4 2.0A in 3D rendering.This may change as Intel will start to work on this.Toms seen to have the same result Xeon and P4 clear winner with a good margin



Webserver Opteron
3D rendering Xeon
Data base Xeon (may change)
Sientifc workload Opteron

FPU benchmark is really bad at Aces sorry.Life benchmark is close to what happen but still out of the map i be waiting to a X86-64 spec submission.(official not peak)


1 thing we see it opteron normaly got a good margin of winning when it win unlike the Xeon that barely win may just be a question of time before it 3 to 1.3D rendering gap seem to big for now maybe a 64 bit driver may help.On the intergrate memory controleur on steam benchmark from toms show that it not faster that a Xeon but the plasma benchmark dont say the same thing some we got mix result on this.

I wait to see more benchmark with others distro of linux others kernel and windows and others database.

[-peep-] french<P ID="edit"><FONT SIZE=-1><EM>Edited by juin on 04/22/03 09:51 PM.</EM></FONT></P>

More about : idea base acehardware

April 23, 2003 2:16:29 AM

TPC result from Rack server using a 4 way systemes

http://www.tpc.org/results/individual_results/RackSaver...

That around mid range low range Xeon MP so yes in every web server Opteron is faster that Xeon Dp but not yet to the level of Xeon MP.Maybe a benchmark with oracle or DB2 may show better result

[-peep-] french
April 23, 2003 4:37:43 PM

http://www.amd.com/us-en/assets/content_type/Downloadab...

FPU spec score with 1.8 GHZ is faster that the older speculation that was give long ago.

FP base 1122 with intel compiler 7.0 C++ and fortran.Once again they have use intel compiler it should prove that with actual compiler for X86-64 will end to be slower that intel compiler.

I guess i was right that 7.0 would offe more power on the Opteron.In regard to some persons.

[-peep-] french
!