View Full Version : barcelona info from china....
http://img2.pcpop.com/ArticleImages/0x0/0/443/000443176.jpg
http://img2.pcpop.com/ArticleImages/0x0/0/443/000443215.jpg
http://img2.pcpop.com/ArticleImages/0x0/0/443/000443216.jpg
http://img2.pcpop.com/ArticleImages/0x0/0/443/000443027.jpg
http://img2.pcpop.com/ArticleImages/0x0/0/443/000443028.jpg
http://news.mydrivers.com/img/20070328/S07043590.jpg
http://news.mydrivers.com/img/20070328/S07052323.jpg
mini cell ???
XS Janus
03-28-2007, 08:21 AM
hmmm...
Gam3Ra
03-28-2007, 08:23 AM
nice
Bennccchhhhmmmaaarrrkkkkksssss!
Weeeee Wwwaannnnnntttt Bbbeeennnnccchhhmmmaaarrrkkkssss!
:)
madcho
03-28-2007, 08:28 AM
:love:
Fatal Error
03-28-2007, 08:35 AM
Infos and more infos and more infos and more slides and more slides and more slides....
nn_step
03-28-2007, 08:36 AM
kinda a repeat but ok
zerotol
03-28-2007, 08:36 AM
slides slides slides tech bs no real result , hope this doesnt turn out to be another r600 release:rolleyes:
Shintai
03-28-2007, 08:38 AM
Old info again :/
Btw, it looks alittle bad for R600 if its only directly 50% faster than R580...
estimated X4 perf.
http://img235.imageshack.us/img235/4229/1454454em9.png
Shintai
03-28-2007, 09:47 AM
estimated X4 perf.
Love the estimate...anyway..aint it time to stop all the estimate/hopes. Else we could all read theinq/fudzilla.
madcho
03-28-2007, 09:50 AM
i think AMD compare clovertown 5355 2.66ghz and opteron X4 2.5ghz. Agena Fx could be clocked at 2.7ghz. intel need 2P Kentfield at 3.2ghz even more to be competitive in very high end.
sacha35
03-28-2007, 09:53 AM
Wow I want one, I want one.
perkam
03-28-2007, 12:31 PM
Anyone notice that in the GFLOPs chart where they have the two R580s and two R600s...the Gflop performance of the A64 and Barcelona are different by only a few 10s of GFLOPs ...
Perkam
flopper
03-28-2007, 11:33 PM
http://img245.imageshack.us/img245/943/k10je7.jpg
20gflops more...
Thats approx 55% faster
but is this core for core, or they are comparing a quad to a dual core?
someone speaking Chinese?
Chinese people do.
:D
The question is, how well do they overklock?
And' how do the spanish babes scale?
:sonic:
Trouffman
03-29-2007, 01:12 AM
I've seen IRL a Barcelona @ Cebit in the hand a great and very comprehensive guy :D...
So waiting for benchmarks too :/
but is this core for core, or they are comparing a quad to a dual core?
someone speaking Chinese?
2xdualcore K8 vs. single quadcore K10
Shintai
03-29-2007, 03:28 AM
Anyone know what the 9% faster than X5355 benchmark is? The specFP_rate is a bloated memory bench. The other one is alot more interresting.
gOJDO
03-29-2007, 03:55 AM
the other is OLT-P, which is a memory bandwidth benchmark
kl0012
03-29-2007, 04:52 AM
Actualy Intel Caneland platform with quad-FSB, 20% higher FSB freq. and bigger snoop-filter will not leave many chanses to Barselona in the server area.
Shintai
03-29-2007, 06:21 AM
5000X is already quadchannel FB-DIMM.
oldblue
03-29-2007, 06:57 AM
the other is OLT-P, which is a memory bandwidth benchmark
Not exactly. STREAM is a memory bandwidth benchmark. OLTP stands for "Online transaction processing (http://en.wikipedia.org/wiki/OLTP)." It is not a benchmark, but a type of server application, often related to database use. Typical OLTP benchmarks are TPC-C or the newer TPC-E. AMD do not specify which benchmarks they use to measure "OLTP performance".
SPECfp_rate (http://www.spec.org/cpu2006/) is basically a floating point high-performance computing (HPC) benchmark. It consists of a suite of fp-intensive applications that includes ray tracing (povray), quantum chemistry, protein folding, speech recognition, weather forecasting, etc.
safan80
03-29-2007, 07:55 AM
With all the powerpoint slides I feel like falling a sleep. that projector will probably break before any real hardware gets released! I'm dying for a live demo of all this cool hardware (barcelona & R600)! I'm starting to wonder if DNF will come out first. Anybody want to start a pool?
Sad to confirm from the page linked to by Oldblue that the SpecFP_rate is indeed a memory benchmark, or "throughput" as they call it. Not much real life performance gain from that one for 1P systems. 9% gain is what i would expect from Barcelona then.
http://www.spec.org/cpu2006/results/
Shadowmage
03-29-2007, 04:38 PM
Sad to confirm from the page linked to by Oldblue that the SpecFP_rate is indeed a memory benchmark, or "throughput" as they call it. Not much real life performance gain from that one for 1P systems. 9% gain is what i would expect from Barcelona then.
http://www.spec.org/cpu2006/results/
I don't think you know the definition of throughput.... :rolleyes:
Here are the benchmarks from SpecFP_rate.
http://www.spec.org/cpu2006/CFP2006/
These are not "bloated memory benchmarks" :rolleyes:
serialk11r
03-29-2007, 04:44 PM
42% faster than Xeon 5355?
Interesting...
Its strange, in China I never saw a single AMD machine for sale in any store...
Metroid
03-29-2007, 04:56 PM
What AMD is doing is a nonsense practice, hold cubs from the lions will not help anyway. :(
I don't think you know the definition of throughput.... :rolleyes:
Here are the benchmarks from SpecFP_rate.
http://www.spec.org/cpu2006/CFP2006/
These are not "bloated memory benchmarks" :rolleyes:
Thats the complete FP benchmarks, both SPECfp and SPECfp_rate, i see no way to determine which is used for what there. If your eyes stopped rolling for a moment you would see that SPECfp is Speed. SPECfp_rate obviously isnt, it is whatever throughput is. If its "eye-rolling-simple" i think you could have included in your post what it is? Feel free to enlighten. No need to treat me as an Intel fanboy, Im just tired of waiting for Barcelona real world performance numbers and is getting doubts about it, thats all.
I agree with those who say "we want benchmarks, now!". Lets see what it can do in the non "_rate" SPECfp and other FP benchmarks. My guess is still 9% for now =)
Anemone
03-29-2007, 07:21 PM
If AMD had something to show, they'd be showing it.
kl0012
03-29-2007, 08:19 PM
Here are the benchmarks from SpecFP_rate.
http://www.spec.org/cpu2006/CFP2006/
These are not "bloated memory benchmarks" :rolleyes:
These are "bloated memory benchmarks"
http://www.anandtech.com/tradeshows/showdoc.aspx?i=2956&p=5
SpecFP rate is nothing more than several SpecFP benchmarks running completely separate from each other, and it is well known that SpecFP is a bandwidth intensive benchmark. So running several of those benchmarks will only increase the bandwidth needed.
Shintai
03-29-2007, 10:59 PM
K8 already beats Intel in specFP_rate...but that dont change the balance on the benchmarks we run.
These are "bloated memory benchmarks"
http://www.anandtech.com/tradeshows/showdoc.aspx?i=2956&p=5
SpecFP rate is nothing more than several SpecFP benchmarks running completely separate from each other, and it is well known that SpecFP is a bandwidth intensive benchmark. So running several of those benchmarks will only increase the bandwidth needed.
uhhh... I knew it had to be bad but not that bad. Assumed it was a benchie where IMC would benefit but choosing a bench where K8 already beats Core 2 is too biased. Hmm, its actually good news also, this can mean affordable prices and no repeatition of X2's inflated prices.
Shadowmage
03-30-2007, 10:51 AM
These are "bloated memory benchmarks"
http://www.anandtech.com/tradeshows/showdoc.aspx?i=2956&p=5
SpecFP rate is nothing more than several SpecFP benchmarks running completely separate from each other, and it is well known that SpecFP is a bandwidth intensive benchmark. So running several of those benchmarks will only increase the bandwidth needed.
It's really not as simple as that.... you think that an industry-standard benchmark suite is nothing but a bandwidth benchmark?
http://www.spec.org/cpu2006/press/release.html
The throughput metrics, SPECint_rate2006 and SPECfp_rate2006, measure the rate at which a computer can complete a number of tasks in a given amount of time.
These are benchmarks running real-world computational-intensive algorithms used in scientific fields, usually.
I tried looking for evidence of the bottlenecks of the SPECfp benchmarks, and there (obviously) isn't a simple answer. Obviously though, the SPECfp benchmarks are floating point benchmarks. There tends to be a lot of ILP for these sorts of benchmarks, so things that would help include:
1. Being able to extract ILP (instruction-level parallelism)
2. Having many or having powerful FP units
3. Having large data caches
4. Large amounts of bandwidth
Here's a paper that analyzes the SPEC benchmarks:
http://escher.elis.ugent.be/publ/Edocs/DOC/P104_006.pdf
They identify a 4 bottlenecks and analyze pretty much every single processor. Core Duo/Core 2 Duo isn't included since they weren't released then.
My point being, it's not all about bandwidth. Take another example: Itanium. It absolutely destroys everything in SPECfp. Does it have crazy memory bandwidth? Not really. It does have #1, #2, and #3 though.
Now, does SPECfp work well as a measurement of so-called "real-world apps"? Well, not really. SPEC benchmarks are targeted towards the industry, not gamers or your grandma. I do think that there should be a strong correlation between SPECfp and multimedia benchmarks (video encoding/decoding, etc), though.
EDIT:
Oh, one more thing: the normal SPECfp is basically single-threaded. That's why they're testing the "rate" version, which exercises all 4 of the cores. This is fair to compare against other multi-core systems.
accord99
03-30-2007, 12:04 PM
EDIT:
Oh, one more thing: the normal SPECfp is basically single-threaded. That's why they're testing the "rate" version, which exercises all 4 of the cores. This is fair to compare against other multi-core systems.
And it is precisely the SPECfp_rate tests that are overly memory bandwidth dependent, so much so that it has little correlation to most enterprise and desktop applications.
Shadowmage
03-30-2007, 04:33 PM
And it is precisely the SPECfp_rate tests that are overly memory bandwidth dependent, so much so that it has little correlation to most enterprise and desktop applications.
I wouldn't say that they're overly memory bandwidth dependent. I would just say that Intel's current architecture has a poor and outdated bus architecture. Even Intel architects will tell you this.
accord99
03-30-2007, 05:40 PM
I wouldn't say that they're overly memory bandwidth dependent. I would just say that Intel's current architecture has a poor and outdated bus architecture. Even Intel architects will tell you this.
Only for applications that are similar to SPECfp_rate, which are rare. A QFX FX-74 should outscore a QX6700 in SPECfp_rate by more than 40% yet this advantage is never seen in the vast majority of applications.
gOJDO
03-30-2007, 08:37 PM
I wouldn't say that they're overly memory bandwidth dependent. I would just say that Intel's current architecture has a poor and outdated bus architecture. Even Intel architects will tell you this.
the fsb is just FINE, for desktop system. Because of the lack of ODMC, Core2 have less RAM bandwidth, thus they have lower SPECfp_rate score.
VulgarHandle
03-30-2007, 08:53 PM
and maybe AMD will make better use of that bandwidth, thus giving such a large lead on a "bloated memory benchmark"
funny, i see alot of people who are skeptical on these benchmarks that also use SuperPi 1M scores as a benchmark to show domination of A64, as if that's a real-world benchmark....
truth is, we won't know jack until NDA's are lifted
be skeptical, but be open-minded as well..
If AMD had something to show, they'd be showing it.
umm, how many true numbers from A64 did we have when it was first released?
just because Intel released stuff early, or, stuff was "leaked", dosn't mean AMD should follow suit
gOJDO
03-31-2007, 03:59 AM
@LOE
The primary idea is irrelevant, because we are talking about a side effect - bandwidth. The efficiency of the K8's ODMC lays in the direct connection to the CPU via a 128bit independent(not affected by other transfers) bus, with a 12.8GB/s throughput. The ODMC on Core2 platform is on the norhtbridge, and is connected to the CPU via a shared(CPU to RAM, CPU to GPU(s), CPU to HDD(s), CPU to other devices) 64bit parallel bus, with a 8.5GB/s of total bandwidth. The bandwidth depends of the latency, also. So, that is another advantage of K8's bandwidth.
Shintai
03-31-2007, 04:40 AM
gOJDO - the primary IMC idea is to reduce latency
the fact amd has more bandwidth is cause their MC is more efficient, not cause it is intregrated
Both yes and no. AMD made the IMC for 2 reasons.
1. Primary reason, servermarket with more sockets.
2. Remove all the problems there was with 3rd party MCHs.
Anything else is a sideeffect they didnt care about.
gdogg
03-31-2007, 05:00 AM
2. Remove all the problems there was with 3rd party MCHs.
Wow thats a great reason. Why not add south bridge to processor too then, then we would be set.
Especially an intel sb, heh.
I can't wait for intel to use an imc too then, cause we do have certain probs with motherboards due to the mch.
Both yes and no. AMD made the IMC for 2 reasons.
1. Primary reason, servermarket with more sockets.
2. Remove all the problems there was with 3rd party MCHs.
Anything else is a sideeffect they didnt care about.
I hope you consider multi-core along with multi-socket. AMD knew there would be massive benefits as CPUs went multi-core.
Whoever said the original X2 CPUs were "price bloated" is wrong. They were the best dual-core CPUs at the time. Intel cut prices when they realised they were losing overall..like AMD are doing now! Are C2D prices therefore bloated? errr.... not really.
At the time, X2 CPUs were priced totally acceptably.
But the past doesnt matter :)
gOJDO
03-31-2007, 05:08 AM
There were not problems with the MCH, but different MCH's were performing different and there was a performance delta of 10% from one MCH, compared to another. For example a K7 on a VIA KT266 and on a nForce2.
"They were the best dual-core CPUs at the time."
By that logic, Intel should be able to charge $10K for its quads.
http://www.techreport.com/reviews/2005q2/pentiumd-820/index.x?pg=1
Pentium D 820 $241
Pentium D 830 $316
Pentium D 840 $530
Athlon 64 X2 4200+ $537
Athlon 64 X2 4400+ $581
Athlon 64 X2 4600+ $803
Athlon 64 X2 4800+ $1001
Thank you price wars.
Whoever said the original X2 CPUs were "price bloated" is wrong. They were the best dual-core CPUs at the time. Intel cut prices when they realised they were losing overall..like AMD are doing now! Are C2D prices therefore bloated? errr.... not really.
At the time, X2 CPUs were priced totally acceptably.
How can you compare X2 prices with Core2Duo? Here Core2 launched at kr1500, X2 launched at kr4200-4300 (thats close to the E6700 pricetag). How F-ing long did we have to wait for X2 3800+? If X2 3800+ had been available earlier, or from day 1, the prices would have been "totally acceptable". But no, AMD were milking money, only high end first, squeeze every cent out of poor PC-addicted cows. Anandtech said like "for the first time in a very long time we recommend an Intel"(PD820). Would anyone have recommended that Smithfield bastard if X2 had reasonable prices?
No-one forced you to pay top-dollar the second they came out, dont forget that! AMD are in this to make money.
Were things different in the UK??? Seriously, IIRC the Intel dual-core was first out in shops, but once the AMD parts were available, the price war started. IIRC...3800 X2 started at <£300, and was still £200 when AMD had dominance.
I`m not comparing architecture, just pricing based on competition. The AM2 X2 series were much more expensive before C2D came out!
largon
03-31-2007, 07:26 AM
Quit whining, X2 was a huge success to AMD, there was no competetive chips out there.
AMD was struggling to sate the demand for DC chip and they could've charged more than they did but X2s would have been flying off the shelves regardless.
nn_step
03-31-2007, 11:50 AM
How can you compare X2 prices with Core2Duo? Here Core2 launched at kr1500, X2 launched at kr4200-4300 (thats close to the E6700 pricetag). How F-ing long did we have to wait for X2 3800+? If X2 3800+ had been available earlier, or from day 1, the prices would have been "totally acceptable". But no, AMD were milking money, only high end first, squeeze every cent out of poor PC-addicted cows. Anandtech said like "for the first time in a very long time we recommend an Intel"(PD820). Would anyone have recommended that Smithfield bastard if X2 had reasonable prices?
If you want the best you have to pay top dollar.
but if you think current X2 prices are bad, I might suggest that you check them. http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=50001028%2040000343%201302820275&bop=And&Order=PRICE
DeathReborn
03-31-2007, 12:09 PM
The prices we paid for X2's early on is what has enabled AMD to expand production & purchase ATI.
Your money went into the R&D for K10 & it's successors so just think about what your money lets AMD do for you.
If you want the best you have to pay top dollar.
but if you think current X2 prices are bad, I might suggest that you check them. http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=50001028%2040000343%201302820275&bop=And&Order=PRICE
Current prices are good. Im still talking about launch, K404 seemed to think they launched top to bottom, even comparing it with Core2 which DID launch top to bottom. Like Intel launched only E6700 and X6800 + 2MB versions of these, and then E6600 some months later?
Now its probably valid to consider AMDs limited production capacity into this but for marketshare (and thats what AMD want) it wasnt a good idea to delay X2-3800 (i think)
nn_step
04-01-2007, 10:33 AM
Current prices are good. Im still talking about launch, K404 seemed to think they launched top to bottom, even comparing it with Core2 which DID launch top to bottom. Like Intel launched only E6700 and X6800 + 2MB versions of these, and then E6600 some months later?
Now its probably valid to consider AMDs limited production capacity into this but for marketshare (and thats what AMD want) it wasnt a good idea to delay X2-3800 (i think)
why waste wafer space producing low profit dies, when you can make a hell of alot more money by making high demand, high profit dies.
AMD is playing the game as intelligently as possible. They know they will not always be on top, so they are doing what ever it takes to gain market share when they are down and then use that market share to expand their profit margin large enough to pay off any debts they accrued during their down time.
What I have said, IIRC held true for the UK availability. I cant even remember what the arguement is about with reading the previous posts, and it doesnt matter anyway.
My opinion is not worth more than anyone elses here, and very few ppl are adding anything of use to this thread, including me.
IF Barca is "all that," i`d expect AMD to release the top chip first because binning will probably be too important for them. Once they have enough discarded chips, they're release the lower chips to distributers. Maximum profit, minimum waste. Yup, different to Intel- partly due to the production capacity and partly due to the speed bumps of C2D.