Bulldozers first screens

[STEvil]

well, not many people with sr-2 rigs on here.. i'm one, but its not me...

[slaveondope]

Who would that be?

http://www.xtremesystems.org/forums/...&postcount=318

[FlanK3r]

Thanks for the tests Flanker! How about setting other cores to 800Mhz and let windows affinity at default(let OS bounce the threads of all 6 cores randomly)? Cinebench 11.5 score mimics the BD ES behavior,somewhat.

It was with 6 cores randomly...Only Superpi I tested with 3 diferent setings (random, affinity to core 0 and afinity to core at 1600 MHz)

Hm, 800 MHz core will be hard, I would have NB down under 2000 MHz. But think, with 800 MHz will be to much bad score.

For those interested,the chiphell poster had to delete/edit his original post.No more crippled Zambezi ES results .

Hehe, maybe it was DrWho at Chiphell forum ?

[informal]

Theoretically you are right, one core (Single thread) can issue, execute, result 4 micro-ops meaning it could use all the FP pipes (2x FMACs + 2x MMX). But i find it very hard if it could sustain it in real world usage.

The point is that pure single thread fp performance is going to be higher than MT fp performance(divided by number of cores). On top of that comes turbo mode.

[Aten-Ra]

The point is that pure single thread fp performance is going to be higher than MT fp performance(divided by number of cores). On top of that comes turbo mode.

Well again only for theoretical maximum, it would be very difficult to have 4x micro-ops from a single thread in order to feed all 4 FPU pipes.

Remember the design is been optimized for multithread/multitask and not single thread like Intel's Core architecture (per core).

[Mumak]

You seem to have a pretty old version of HWiNFO32.
Few hours ago I released a new public Beta (1230) with more improvements for Bulldozer support..

Some my benchmarks with x6 1090T@ 2800 MHz, turbo off, core 0 set 2800 MHz, others at 1600 MHz. OS is 32-bit win XP. So, look at this...

[informal]

Well again only for theoretical maximum, it would be very difficult to have 4x micro-ops from a single thread in order to feed all 4 FPU pipes.

Remember the design is been optimized for multithread/multitask and not single thread like Intel's Core architecture (per core).

How do you know for what workloads Orochi is optimized/designed? How about we wait and see how it does in real life workloads.Then we can say what AMD had in mind when they designed the thing.

[geo]

not saying the benchmarks are true, but why would AMD give out an ES to a guy who cant seem to figure out that the ES he's got is crippled ?

[keithlm]

but why would AMD give out an ES to a guy who cant seem to figure out that the ES he's got is crippled ?

Crippled ES chips could be used to test early motherboard and/or bios compatibility. During early motherboard development phases the only question is "will it work" and not "will it work well."

Later they would want to issue non-crippled chips for stress and performance testing; but they can keep much tighter control of the non-crippled chips. (If this truly is an early crippled ES sample it is probably 6 months old if not older... because at this time I would expect non-crippled chips to be issued by now.)

[Dimitriman]

Yes agree with above post. The ealier the ES revision the less likelly to be anything useful as an indication of final performance. But he could test it on some asus am3 boards and see if that will in fact work.

[LightSpeed]

IIRC, the first ES K8 tests were quite bad. Didnt really turn out that way did it ^^

However, I dont expect BD to have the same impact, certainly not. Ill be happy if it is faster by 10 - 15% compared to 2600k in MT workloads. Lets see what happens

[BeepBeep2]

...sooo what does this all mean that anyone who wants a high end cpu has no choice but should go with the 2600k...if not done already.
its even possible that once Bulldozer gets released Intel might increase the prices for the 2600 since currently theyre only 260euro which is pretty cheap for such a high end cpu...same as AMD did past year with the 5850/70 gpus

Cmon, were all smarter than this...
Four K8 90nm cores @ 3.2 Ghz could beat these scores...

[Chumbucket843]

Well again only for theoretical maximum, it would be very difficult to have 4x micro-ops from a single thread in order to feed all 4 FPU pipes.

Remember the design is been optimized for multithread/multitask and not single thread like Intel's Core architecture (per core).

they are not meant to all be used at once. heavily mixed ints and floats in sse is not only rare, but also very slow. at least that is what i have noticed.

How do you know for what workloads Orochi is optimized/designed? How about we wait and see how it does in real life workloads.Then we can say what AMD had in mind when they designed the thing.

http://www.anandtech.com/Gallery/Album/754#1
i did not see any direct statement that it is designed for thread level parallelism but words like "sharing", "throughput", "scalability", etc. all suggest that multicore perf is a major focus.

[XRL8]

yes, like they hid k8. not showing anything b4 launch is usually a good sign @amd.

about bd performance: whaka whaka!

They did show K8 before launch, there were also some previews if i remember correctly.
Theres one:
http://www.anandtech.com/show/883/4

[w0mbat]

with real benches?

[XRL8]

with real benches?

You said that AMD didnt show hammer, so thats a good sign that theyre not showing bulldozer.
They did show various hammers on few occasions ,and showed whole platforms to the reviewers (the solo platform).Thats more than we can say about BD.I didnt say they showed numbers.However they were more open about things back then.

[accord99]

First Hammer numbers were obtained June 2002 by Tecchannel who got 1 hour with a 800 Mhz Hammer system, roughly a year before the Opteron release and 15 months before the A64 release.

http://www.tecchannel.de/ueberblick/...mer_prototype/

[mars]

First Hammer numbers were obtained June 2002 by Tecchannel who got 1 hour with a 800 Mhz Hammer system, roughly a year before the Opteron release and 15 months before the A64 release.

http://www.tecchannel.de/ueberblick/...mer_prototype/

yes this is what i remember as well k8 wasnt a hush hush release but yea i remember the above quite well.... which is odd because i am perpetually stoned

i hate to say this but this is alot more akin to the k10 release than the clawhammer

[SEA]

OK, now we know what everybody thinks of upcoming release. Now let's wait and see unless you have some real news...

[Postmodum]

Maybe... the majority of the users are "gamers" and not overclockers anymore ... ehehhehe ( gamers usually stand for a brand and they defend it to the death! eheheh )

It's so sad to close your eyes to all the options you have ... it's like having your vision tuned to only see green or red or blue .... but not the 3 at the same time ...


Oh... did i say Vision? Sorry ... i wanted to say Vision FX tuned to only see green or red or blue ... ahahahhaha

[generics_user]

yes this is what i remember as well k8 wasnt a hush hush release but yea i remember the above quite well.... which is odd because i am perpetually stoned

i hate to say this but this is alot more akin to the k10 release than the clawhammer

Hammer suffered severe delays due to issues with SOI and the memory controller VERY shortly before launch; that's why benchmarkes were available half a year bfore launch and not 3 months

that said amd got much quieter after the barcelona fiasco, remember that they showed off 3ghz chips a long time before launch; kept on talking how it's going to destroy C2Q???

i prefer them to be quiet until they release BD instead of talking about huge performance and clocks and disappointing at launch...

[-Boris-]

Hammer suffered severe delays due to issues with SOI and the memory controller VERY shortly before launch; that's why benchmarkes were available half a year bfore launch and not 3 months

that said amd got much quieter after the barcelona fiasco, remember that they showed off 3ghz chips a long time before launch; kept on talking how it's going to destroy C2Q???

i prefer them to be quiet until they release BD instead of talking about huge performance and clocks and disappointing at launch...

Did AMD ever say that Phenom would be able to beat C2Q? We have people today who thinks AMD said BD would beat i7 just because fake slides stated that AMD said so. I'm sure these people will call AMD liars if BD don't beat i7.

[Particle]

Even if it managed large wins, if there were a few fringe cases where it didn't beat an i7 those same people would scream about how it's not any faster. When it comes to challenging Intel, a competitor's CPU would have to be undeniably devastating to Intel's current flagship in order for it to be viewed as even a legitimate equivalent.

This isn't specific to Intel or CPUs of course. It's just the psychological curiosity of brand leadership, perceived or real, and is rather intriguing.

[gallag]

There are a lot of hypocrites in here, fanboys complaining about fanboys lol.

[AbortRetryFail?]

Did AMD ever say that Phenom would be able to beat C2Q? We have people today who thinks AMD said BD would beat i7 just because fake slides stated that AMD said so. I'm sure these people will call AMD liars if BD don't beat i7.

IIRC, there was a gentleman from AMD who cherry-picked a single test of Barcelona and noted a 40% gain ...

The internets morphed this into BARCELONA IS 40 PERCENT MO' BETTAH.

Then came the TLB, and then the nOverclocking that wasn't semi-corrected until sb 710-50, ACC and the 'magic pins'.

Shortly thereafter the AM2+ 940/920 PhIIs came out.