backtop


Print 93 comment(s) - last by jazkat.. on Jun 28 at 10:35 PM


Cinebench ran AMD "Barcelona" 1.6 GHz in 27 seconds. (Source: DailyTech, Anh Huynh)

Cinebench ran Xeon X3220 2.4 GHz in 17 seconds. (Source: DailyTech, Anh Huynh)
An early AMD "Barcelona" revision gets its first non-simulated benchmark

Earlier today, AMD announced that it successfully demonstrated Barcelona across the server market.  The company did not publically state how fast the processor was running, the stepping of the processor, the processor thermal envelope or the eventual ship date.

We had the opportunity to benchmark the AMD Barcelona, native quad-core on an early stepping. We only had a few minutes to test the chip, but we were able to run a quick Cinebench before we were instructed to leave.

The AMD benchmark ran on a single-socket, K10 CPU running at 1.6 GHz on NVIDIA's nForce Professional 3400 chipset.  According to the system properties, the AMD system used 4GB of DDR2-667.

The most similar Intel system we could muster up on such short notice was an Intel Xeon 3220.  The Xeon X3220 is clocked at 2.4 GHz, and ran on Intel's Garlow platform (Intel X38).  This system property profile stated the system utilized 4GB of DDR2-800.

Cinebench completed the default benchmark in 27 seconds for the 1.6 GHz K10; 17 seconds for the Intel Xeon X3220.  The Kentsfield Xeon was 58% faster with a 50% higher clock frequency for Cinebench.

Both systems ran Windows 2003 R2, 64-bit.

AMD partner engineers tell DailyTech the chip we tested was the latest revision silicon.  The same engineers claim 2.0 GHz Barcelona chips are making the rounds, with 2.3 GHz already on the desktop and server roadmaps. 

AMD's current guidance suggests a late July announcement for Barcelona.  However, when DailyTech tracked down the individual partners named in AMD's press release, all cited "optimistic September" ship dates for motherboards. 


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Not so good for AMD
By Proteusza on 6/6/2007 6:11:22 AM , Rating: 2
Assuming the K10 was clocked at 2.4 GHZ, it should have done the test in 18 seconds (assuming increasing clockspeed linearly decreased the time taken to perform the test).

How much of a difference will the slower memory bandwidth offered by the K10's DD2 667 make?




RE: Not so good for AMD
By defter on 6/6/2007 6:25:11 AM , Rating: 1
Memory bandwidth shouldn't make much difference in Cinebench.

These results are very poor, they show very small per clock per core performance increase compared to K8: http://www.hwupgrade.com/articles/cpu/27/the-last-...

dual core 3GHz K8 = 829
quad core 1.6GHz K10 = 814


RE: Not so good for AMD
By JackPack on 6/6/2007 6:40:32 AM , Rating: 2
Indeed, renders like Cinebench don't care about memory bandwidth.

Besides, Barcelona only supports registered DIMMs up to 667 MHz.


RE: Not so good for AMD
By spartan014 on 6/6/2007 7:46:07 AM , Rating: 2
Aren't these meant for use in servers.. Doesn't HT perform better than FSB architecture when all 4 cores are fully loaded? Sure, there won't be a 30% performance difference. But only by this benchmark, can we conclude it fully?


RE: Not so good for AMD
By Reynod on 6/6/07, Rating: -1
RE: Not so good for AMD
By defter on 6/6/2007 8:34:30 AM , Rating: 2
quote:
The single socket comparison is useless anyway condsidering servers


Single socket comparison is not useless, since Barcelona based CPUs will be used also in the desktop and single socket servers.

I'm sure that Dailytech would be happy to benchmark dual or quad socket Barcelona systems, but it seems that they got access only to a single socket system at Computex.


RE: Not so good for AMD
By dess on 6/6/2007 12:32:17 PM , Rating: 3
quote:
Single socket comparison is not useless, since Barcelona based CPUs will be used also in the desktop and single socket servers.

It is actually not true in this form. Barcelona is a HT2.0, non-split power plane part, intended to be used in place of the older Opterons. Desktop (and some of single socket server) parts will be based on Budapest, a HT3.0, split power planes part, what - according to some reports - enables it to be clocked (much?) higher than Barcelona, as well.

Also, Cinebench is scaling rather along the clock, not along true performance. See:
http://www.hwupgrade.com/articles/cpu/27/the-last-...
So it was just the worst selection to make a test in this case...


RE: Not so good for AMD
By Topweasel on 6/7/2007 11:07:12 AM , Rating: 2
Yeah looking at that only the Core 2 duo comes close to jump out of its spot MHz wise vs. the Performance and even then they seem to leap from maybe one position.


RE: Not so good for AMD
By AntDX316 on 6/10/07, Rating: 0
RE: Not so good for AMD
By vignyan on 6/7/2007 10:13:34 AM , Rating: 2
quote:
A truly free market requires many competitors


Agreed. So AMD must be coming up WRT performance not the otherways! :)

Also this is an early silicon. So cross your fingers for the actual stepping.

I am just thankful that Barcelona is a "real" project and not just a paper project! :)


RE: Not so good for AMD
By Calin on 6/8/2007 8:06:23 AM , Rating: 2
The performance of the early silicon for the AM2 processors didn't increase much in the last 3 months before launch. But I really hope this time it will


RE: Not so good for AMD
By Gibby82 on 6/10/2007 12:46:29 AM , Rating: 2
Early silicon for AM2? All AM2 was is a package change and the move to DDR2 support. Other than DDR2 it's K8/939 on an different package. Of course you won't see any performance increase.


RE: Not so good for AMD
By Gibby82 on 6/10/2007 12:47:36 AM , Rating: 2
*AM2 is just* a package change...


RE: Not so good for AMD
By knowyourenemy on 6/12/2007 1:44:17 AM , Rating: 2
The personal attack was uncalled for, but I do see your point. A little more tact, maybe. :/


RE: Not so good for AMD
By bob661 on 6/6/07, Rating: -1
RE: Not so good for AMD
By MDme on 6/6/2007 10:06:11 AM , Rating: 1
he did read it. The article mentions the K10 running at 1.6Ghz. He was ASSUMING that IF it did run at 2.4Ghz it would complete the task in 18sec.....which is 1/3 less time.

I would like to comment though that AMD systems would likely see more of a boost with a higher clock since the IMC would clock higher as well.

Anyway still based on this isolated specific incident, the K10 does not appear to be the C2D killer AMD purports it to be, not unless this is some ridiculously low power part where AMD can argue perf-per-watt metrics to server companies. (or it is sandbagged).


RE: Not so good for AMD
By PlasmaBomb on 6/7/2007 5:06:12 AM , Rating: 2
In AMD's forthcoming architecture the IMC will be able to be clocked separately from the cores, so who knows what the performance will be like.


RE: Not so good for AMD
By Master Kenobi (blog) on 6/7/2007 9:14:06 AM , Rating: 2
Intel has already proven IMC isn't the end all be all. Heck the C2D does just fine and the Memory Controller is on the north bridge.... IMC is good, but it isnt magical.


RE: Not so good for AMD
By KristopherKubicki (blog) on 6/7/2007 9:39:14 AM , Rating: 2
IMC sure has some distinct advantages -- but it also has some distinct disadvantages too. I really wouldn't call one approach superior to the other, especially when we're talking general computation.

Eventually the whole system is going to sit on the chip anyway :-P


RE: Not so good for AMD
By Master Kenobi (blog) on 6/7/2007 11:35:35 AM , Rating: 2
Chip? Heck by that point we will probably be growing them like crystals LOL.


RE: Not so good for AMD
By knowyourenemy on 6/12/2007 1:45:45 AM , Rating: 2
Do I detect Superman 4?


RE: Not so good for AMD
By dess on 6/7/2007 12:51:32 PM , Rating: 2
quote:
Intel has already proven IMC isn't the end all be all.

Really!? Then they should gone mad since then, because they use just that with Nehalem...
http://www.pcper.com/article.php?aid=382&type=expe...


RE: Not so good for AMD
By Master Kenobi (blog) on 6/7/2007 1:14:06 PM , Rating: 2
Again. IMC is not the end all be all. The fact that Intel still holds the crown, and is still not using an IMC, after AMD has been using one for around 6 or 7 years now demonstrates this fact. Intel moving to an IMC doesn't prove anything. It just means Intel has decided an IMC would be good for the CSI board style, and having it on the north bridge would be bad on the CSI system. Having the MC on the Northbridge seems to work just fine on BUS!


RE: Not so good for AMD
By Spoelie on 6/8/2007 11:11:33 AM , Rating: 2
opteron was released april 2003, athlon64 october 2003
that's 4 years, not 6 or 7. I doubt they already had prototypes 2-3 years ahead of launch ;)


RE: Not so good for AMD
By dess on 6/8/2007 1:32:17 PM , Rating: 2
Note that Intel needs those monumental caches just because of the abesent IMC... Partly due to the IMC, K8s could be faster than P4s, despite the much lesser L2 cache.


RE: Not so good for AMD
By Accord99 on 6/9/2007 4:09:07 AM , Rating: 2
Yet even with 1MB of shared L2 cache, the Pentium 2100 series remains a powerful processor and more than a match for the K8.


RE: Not so good for AMD
By dess on 6/10/2007 10:59:15 AM , Rating: 2
The 2100 series is quite a bit slower than an E6300 (2MB L2), let alone E6400 and up, at similar clocks, while the E6300 equals to a mid-range X2.


Am I understanding the math correctly?
By RussianSensation on 6/6/07, Rating: 0
By RussianSensation on 6/6/2007 6:28:19 AM , Rating: 4
Nevermind, I don't know what I was thinking. Twice as fast would position a Xeon at 13.5 seconds (or 100% faster). The 58% makes sense actually. My bad...Time to re-learn grade 3 math :)


RE: Am I understanding the math correctly?
By jpeyton on 6/6/2007 4:00:53 PM , Rating: 5
Just for comparisons sake (I know some of you were looking for this number), I ran CineBench 9.5 on a dual-Opteron workstation (4 cores total) with the processors running at 1.6GHz (Vista Ultimate 32-bit, 4GB DDR2 667, nForce Pro 3600 Chipset).

[b]29 seconds[/b] to render the scene in the multiprocessor benchmark.

DailyTech didn't mention what OS they used; if they do, I'll rerun the benchmark with it.


By KristopherKubicki (blog) on 6/6/2007 6:24:09 PM , Rating: 2
Sorry, it was Windows 2003 Server SP2, 64-bit. I updated that into the article.


RE: Am I understanding the math correctly?
By jpeyton on 6/6/2007 4:01:47 PM , Rating: 2
Just for comparisons sake (I know some of you were looking for this number), I ran CineBench 9.5 on a dual-Opteron workstation (4 cores total) with the processors running at 1.6GHz (Vista Ultimate 32-bit, 4GB DDR2 667, nForce Pro 3600 Chipset).

29 seconds to render the scene in the multiprocessor benchmark.

DailyTech didn't mention what OS they used; if they do, I'll rerun the benchmark with it.


By Proteusza on 6/6/2007 4:11:05 PM , Rating: 2
Good god. may god have mercy on AMD if that is true, because Intel wont.


By Furen on 6/6/2007 4:21:10 PM , Rating: 2
I think the test system was running windows server 2003 x64. Notice that Cinebench was the 64bit version, which does make a big difference.


By w0mbat on 6/6/2007 6:28:51 AM , Rating: 2
27 : 0,17 = 158,82 -> The K10 needed 58% more time to finish the task.


By defter on 6/6/2007 6:30:10 AM , Rating: 2
quote:
Or reduced 10/27 from K10's time which is 37% not 58%. Therefore, the performance increase is 37% given a 50% clock difference.


1/(1-0.37) = 1.58 => Xeon is 58% faster given 50% higher clock.

Or alternatively: K10 is 37% slower given 33% lower clock.

quote:
Had Xeon ran 58% faster than K10, its time to complete the task would have been more than half as fast (i.e. 11.34 seconds).


No, you forgot that if A is 100% faster than B then B is 50% slower than A.

If the Xeon would compete the task in less than half time, then it would be over 100% faster.

Maybe it's easier to just look at the Cinebench scores:
K10 = 814
Xeon = 1274
=> Xeon about 58% faster.


By leidegre on 6/6/2007 6:37:01 AM , Rating: 4
Well, what you express with the ratio 27/17 is basically how much more work can I do given that I'm doing this task in 17 seconds and not 27.

27/17 ~= 1,5882

Which then actually would have been a 59% performance difference.

You are describing given the time difference in performance measured for this task how many of those can I perform given I have 27 seconds, which actually is 2,7 as you managed to invert the result.

However, this type of silly guessing doesn't give you anything. I'm satisfied with the early benchmarks as they are, and for now we should not try and read too much into it. Let's hold off the flame wars until we have a little more facts ey?


RE: Am I understanding the math correctly?
By Awax on 6/6/2007 7:45:39 AM , Rating: 4
You have the choice :
* the K10 is clocked 33% slower than the Xeon and performs 37% slower (here, 100% is the Xeon)
* the Xeon is clocked 50% faster than the K10 and performs 58% faster (here, 100% is the K10)

So, on a clock/performance ratio point of view, the K10 isn't that great : a 100MHz boost on the Xeon is matched by a 117 Mhz boost on the K10.


By Spoelie on 6/8/2007 11:00:54 AM , Rating: 2
You are not taking into account the memory speed differences. The kentsfield was running in its most optimal memory configuration (unreleased x38 enthousiast chipset) with 20% faster memory, while the barcelona adds support for 1066mhz mem but is running only 667mhz. We do not know if the board it was running on was one that provides split power planes (am2/1207 or am2+/1207+?), the latter which should provide the processor a 5-10% boost at the cost of a few watts according to last heard information.

Not the best reference to make any judgments on performance per clock.


RE: Am I understanding the math correctly?
By RW on 6/6/07, Rating: -1
RE: Am I understanding the math correctly?
By penter on 6/6/07, Rating: 0
RE: Am I understanding the math correctly?
By penter on 6/6/07, Rating: -1
By casket on 6/6/2007 10:29:20 PM , Rating: 2
Distance= Rate * Time
Distance =30,000 (renders?)
Rate = Cinebench score (Renders per second?)
Time = seconds (17.25 vs 27).
**********
"A 2.4 Ghz Quad-core K-10 goes the equivalent speed of a 2.3 Ghz Xeon. It gets a 2300+ rating.

AMD's chip is 95.84% the speed of Intel, clock-for-clock."


RE: Am I understanding the math correctly?
By jhtrico1850 on 6/6/2007 7:51:36 PM , Rating: 2
Let's just choose an arbitrary number of 5000 pixels for a scene. If Barcy at 1.6GHz finishes in 27 seconds, that is 185 pixels/s. If Kentsfield at 2.4Ghz finishes in 17s, that is 294 pixels/s. Adjust the Barcy score of 185*(2.4/1.6) for 277.5 pixels/s. As you can see, 294 is 58% faster than 185, and 2.4 is 50% more than 1.6, and in this case, the Kentsfield is faster than the Barcy clock for clock with 294 vs 277.5.


RE: Am I understanding the math correctly?
By hoyanf on 6/7/07, Rating: 0
By GlassHouse69 on 6/7/2007 12:34:53 PM , Rating: 2
My only hope is that Kuma dual core has like a 25% speed increase or something like that over am2's of same ghz rating, AND they are like 120 dollars at most to buy. this would make intel's systems to cost 300-500 dollars more to purchase, considering the AMD boards normally run 65-75 dollars for full features and full speed. DDR2 is luckily at a super cheap price at the moment. combine all of that with a more than adequate chip, this will change the market.

gaming is all gfx card so long as you use 1920x1200 displays with "image quality" settings cranked.


nvidia nforce 3400
By duploxxx on 6/6/2007 8:30:30 AM , Rating: 3
ehh to my opinion nvidia nforce 3400 is a socket am2 board...
so we are seeing budapest here, not barcelona.

and if am2 you also have less performance due to the board, not to mention the early bios phase.

anyone care to explain?




RE: nvidia nforce 3400
By defter on 6/6/2007 8:38:43 AM , Rating: 2
The main difference between Barcelona and Budapest is the socket. Thus, people use term "Barcelona" to describe whole range of K10 based products, just like people are using "Penryn" to describe 45nm desktop and server products (even though Penryn is a codename for a mobile chip).

quote:
if am2 you also have less performance due to the board


Why would AM2 boards would have less performance? Actually, all Opteron 1000 series CPUs use socket AM2: http://www.amd.com/us-en/Processors/TechnicalResou...


RE: nvidia nforce 3400
By raven3x7 on 6/6/2007 3:10:49 PM , Rating: 1
No Budapest also has HT3 while Barcelona is still HT1. Also what makes you think that its impossible to make a Socket F nf3400 board. There is no technical limitation preventing it.


RE: nvidia nforce 3400
By Spoelie on 6/8/2007 11:17:20 AM , Rating: 2
RE: nvidia nforce 3400
By Proteusza on 6/6/2007 8:48:16 AM , Rating: 2
Yes Nforce 3400 is an AM2 board.

And this is budapest, because its a 1P processor.

There probably would be a performance loss associated with using an AM2 board, not sure how much though. In this case, probably a fair bit as it would have been using all 4 cores.

No way to tell until we get more detailed benchmarks with AM2 and AM2+ boards in a variety of different situations. And preferably closer to final silicon.


RE: nvidia nforce 3400
By coldpower27 on 6/6/2007 9:08:47 AM , Rating: 3
Actually, these are effectively what your going to get on Bareclona, albeit this is only 1P, as they are sticking to HT1.x and Socket F Registered DDR2 for the first interation of the server SKU.

The improvements to I/O with HT3.0 won't really affect Single Socket systems, and are more geared toward the multi Socket platforms. So your maybe looking at 2-5% more with DDR2-800 on the desktop.

It was always going to be a close race in the Uni-processor arena, it's only in Multi Socket systems AMD's server based architecture gets to shine.

Though it is a little worrying if they are suppose to be launching in July and they are only demonstrating silicon at 1.6GHZ vs the rumored 2.3GHZ launch speed.


RE: nvidia nforce 3400
By defter on 6/6/2007 9:26:01 AM , Rating: 2
quote:
In this case, probably a fair bit as it would have been using all 4 cores.


Some people are still clueless about AMD's platforms.

Please explain to me WHY Socket F board with one populated socket would be faster than Socket AM2 board (as long as CPU and memory run at the same speed)?


RE: nvidia nforce 3400
By raven3x7 on 6/6/2007 3:26:34 PM , Rating: 2
I think he meant AM2 compared to AM2+. Anyway if he meant socket F the opposite would be true as socket F uses registered ram which has a slight performance penalty and therefore an AM2 system would probably use slightly faster memory in most cases. Otherwise both Socket F and AM2 1P systems should perform identical


Quick, DT!
By TejTrescent on 6/6/2007 6:04:26 AM , Rating: 4
One down on the list of benchmark-less AMD products, now you just need to somehow get a benchmark of the HD2600 up and I think you'll put a lot of questions to rest.

... or, knowing the internet, it'll be the great flame war of our time.

Either way, thanks for the benchmark! Good to finally see one of any sort.




RE: Quick, DT!
By davegraham on 6/6/07, Rating: 0
RE: Quick, DT!
By jhtrico1850 on 6/6/2007 9:57:24 AM , Rating: 2
a) We can use simple math to extrapolate scores at certain MHz.
b) AMD has been touting "drop in" upgrade for Barcelona. Boo hoo
c) DDR 800 is not going to change the fact that these scores are similar to K8
d) Would you rather have benches now or later?


RE: Quick, DT!
By TejTrescent on 6/6/2007 12:26:45 PM , Rating: 2
Yeah, I'm not so much caring that AMD's a bit behind when you extrapolate the scores.

I'm just glad to finally see proof that K10's not gone the way of the Phantom!


RE: Quick, DT!
By mhahnheuser on 6/7/2007 2:36:48 AM , Rating: 2
Quite right. Enthusiasts like us want the muscle, but truth is the rest of the world plus dog wants, smaller, quieter, less power hungry cpu's. AMD is fighting on 2 fronts now, CPU market and GPU market. It's stetched itself pretty thin, but in the long term it means getting a greater variety of products in every market segment on the table. X2 is plenty fast enough for "Windoze any flavor." Hence seeing lower and lower powered Athlons appearing. Also the more I use the Cool and Quiet feature on my desktop the more I like it and makes me feel better about leaving it idling. I think this feature is highly underated. Also I managed to break a P4EE nForce4 mobo the other day. Culprit was heat, the offending part was the Northbridge, 1066MHz FSB and beyond has challenges. Look at the cooling solutions on the 680i mobos? Faster is not always better, and AMD needs to reposition itself in the short term to be a stronger competitor in the long term. Short sightedness is termed "myopia." (and before i'm plastered a fanboi, my 2nd system is based on C2D.)


Doesn't phase me too much
By Zurtex on 6/6/2007 8:51:11 AM , Rating: 2
Well this kind of benchmark doesn't phase me too much. Early silicon, much lower clockspeed, far from ideal benchmark conditions.

I'm waiting to put off making a new computer till the end of the year, even if AMD doesn't deliver equal or better chips than Intel, I can't imagine they'll be taking a step back, so Intel will undoubtedly cut their prices yet further.




RE: Doesn't phase me too much
By defter on 6/6/07, Rating: 0
RE: Doesn't phase me too much
By Zurtex on 6/6/2007 9:32:00 AM , Rating: 3
Sorry, by early silicon, I mean not final silicon. I remember the AM2+ having early benchmarks (1 or 2 months before launch) and it improving 3-6% in some sectors still before final launch.

Ideal benchmark conditions would not be described as "quick and dirty". They would also be CPUs given by AMD and their specs would be ones similar to AMDs roadmap.


RE: Doesn't phase me too much
By darkpaw on 6/6/07, Rating: 0
RE: Doesn't phase me too much
By paydirt on 6/6/2007 9:12:46 AM , Rating: 2
nVidia did not cut their prices when it was discovered that the 2600 did not outperform the 8800GTX. So don't hold your breath that Intel will cut their prices beyond the somewhat aggressive July 22nd price cut.


RE: Doesn't phase me too much
By Zurtex on 6/6/2007 9:33:11 AM , Rating: 2
Errr, yeah. But the K8 CPU is hot on the heels in the budget world of Intels CPUs. There's not a HUGE difference between getting a budget AMD platform and a budget Intel platform at the moment, in some cases I'd even advise AMD as the better platform. So if AMD can push the price performance with this new architecture and it can push out of the budget sector, Intel will either have to yield the crown or cut prices even more aggressively, given their current attitude I don't imagine their choice will be to yield.

I realise that that relies on a couple of ifs, but AMD have had plenty of time to do R&D, I don't seem them falling back with a new architecture. But only time will tell.


How is it compared to dual x2s?
By GoatMonkey on 6/6/2007 8:30:48 AM , Rating: 2
What do a pair of dual core X2s (around the same clock speed) run in this benchmark?




RE: How is it compared to dual x2s?
By defter on 6/6/2007 8:40:11 AM , Rating: 2
There aren't any 1.6GHz dual core X2s, but here are some K8/Core2 scores: http://www.hwupgrade.com/articles/cpu/27/cinebench...

3800+ (2GHz) Athlon64 X2 scores 558.


RE: How is it compared to dual x2s?
By PrinceGaz on 6/6/2007 9:31:25 AM , Rating: 2
I think he meant a 2P (dual-socket) system, as all X2's are dual-core. Of course he should have said dual Opteron 2xxx rather than dual X2. That way you would effectively have a quad-core K8 system to compare with the quad-core K10.


By GoatMonkey on 6/6/2007 10:54:24 AM , Rating: 2
Yes, I meant a pair of dual core Opterons.

Here's one that I found with a pair of 275s. It does 21.1 second at 2.2Ghz vs. 27 seconds for the new 1.6Ghz chip.

http://videoediting.digitalmedianet.com/articles/v...

There must be some closer (clockspeed) benchmarks out there.


Barcelona is slower than Athlon in this test...
By fikimiki on 6/6/2007 11:40:37 AM , Rating: 5
If we take Athlon 3800+ 2.0GHz which has 25% faster clock than this chip we get 279 points per core. Barcelona gets 203, so adding 25% more clock (2.0) should get 254. This is SLOWER than Athlon so this kind of test cannot be considered as a true performance result. Or K10 is an downgrade. ;)




By raven3x7 on 6/6/2007 3:31:20 PM , Rating: 2
I just have this feeling that AMD is actually sandbagging.


Did anyone look at the systems?
By Treckin on 6/8/2007 4:12:06 AM , Rating: 1
I just wanted to point out that the xeon was run on the x38 chipset... Supposedly the highest end enthusiast chipset that money cant even buy yet...
Honestly, this was a horrible benchmark IMO. It is worthless to compare a beta 1.5 server silicon chip to a well developed, second or third iteration production model clocked at nearly 175% of the speed. Also, the article didnt say whether we were looking at phenom or one of the lower derivatives... or if it was socket f or am+, or if it were the ominous + series... This was a poorly reported test, and should have been not reported on for reason of quality (or lack there of). I honestly expected more from DT on this one. I can only think that they were all giddy over barcelona + benchmark that they soiled themselves and impulsively posted faulty/unverifiable/unreliable/biased data.
This could be like comparing a super low end c2d to ... um .... a brand new quadcore 1 bizillian mb cache beast on a Brand




By Gravemind123 on 6/8/2007 1:02:15 PM , Rating: 2
It also shows that the Phenom was using DDR2-667 instead of DDR2-800, whether or not this benchmark relies on memory speeds as a factor could also be an issue. This is also only one benchmark, and we have to wait and see if this trend continues for other applications.


By murphyslabrat on 6/8/2007 1:37:00 PM , Rating: 2
I hate to be another naysayer, but as has already been pointed out, Conroe was kicking but and taking names well before the 1-month-till-launch point.

I will be getting an AMD CPU regardless of absolute performance. AMD has been very competitive with price/performance in the low- to mid-range segment, and I don't have $700 to spend on a CPU, let alone the system.

The bottom line is that Barcelona(at least at this point) is not the C2D killer that we-AMD fans-have been looking for. Although it will mean quad-core and improved IPC's for AMD, it isn't that magical Athlon64 II.


RE: Did anyone look at the systems?
By dess on 6/8/2007 1:41:06 PM , Rating: 2
Phenom is a brand name/moniker of the desktop series. This one in the test was an Opteron brand chip with a Barcelona core, so the server variant.


These chips could be in debug mode...
By extechman on 6/6/2007 5:27:09 PM , Rating: 2
I used to write system software for SGI, and when the new MIPS processors were coming out, we often had to run them in weird modes to work around a bug here or there. Some of them had some pretty substantial performance penalties, but the benefit of being able to find the next problem without respinning the chip made it worth it. At times, we had to turn off write buffers, shrink caches, put the chip in "in order" more, etc. AMD may be doing workarounds like those in this rev of silicon.

What I'm saying is that we still don't know anything about the performance of real Barcelona chips. Extrapolating the GHz won't give you the whole picture.




RE: These chips could be in debug mode...
By hstewarth on 6/7/2007 10:33:40 PM , Rating: 2
I think you might be confusing software with hardware. I don't believe that its same as software running in debug more than hardware. Most debugging on CPU's are done with ICE's which are done done with processor in realtime. I use such devices years ago debugging OS software.

Yes GHZ doesn't mean much.. it doesn't mean that Intel chip is 50% faster cpu than this chip. There is other factors like memory bus, cache and core instructionset.

Actuall with above in account and AMD Onboard memory controller and other things, I expect that this processor would do a lot better than it did.


By extechman on 6/14/2007 1:05:16 AM , Rating: 2
Hi, hstewarth

I was trying to be careful because I don't know exactly how this works on AMD CPUs, but what I meant was that in the CPUS I worked with (back in the day) there are a bunch of configuration bits that got clocked into the chip at startup to control its function. I believe that the same thing happens with modern x86 chips except that these bits come from the BIOS.

When we were debugging MIPS chips and new multiprocessor system architectures, we sometimes needed to change those configuration bits to tell the chip not to use certain performance features because they would cause the chip to produce incorrect results.

For example, if the chip was having trouble with cache coherent loads using an update protocol, we would switch the chip into invalidate mode - slower in some situations, but it worked. Or when the chips were new, we would have to make sure that the sysad bus (MIPS equivalent to FSB) ran at a particular fraction of the internal clock. Or we would have to disable out-of-order execution...

It's actually a lot like software debugging. It's all about testing as much logic as you can. You have to find as many bugs as you can before you spin the chip so sometimes you have to have to run at significantly lower performance just to get the OS to boot and run benchmarks properly. Then you fix a whole bunch of bugs and critical paths at the same time and respin the chip.

We don't know where AMD is in that process, but you'd have to hope that these chips are early silicon that was able to run correctly but had performance features turned off or not well tuned (what I soemwhat simplisticly referred to as "debug mode"). It's possible that the chips are being respun as we type.

If anyone would like to enlighten me on how AMD chips get configuration bits from the BIOS and what they control (do they still load microcode?), let me know!


Come again
By SavagePotato on 6/6/2007 3:18:26 PM , Rating: 4
Has dailytech not learned a lesson about posting oddball benchmarks of unreleased products in poor comparisons?




RE: Come again
By Lugaidster on 6/7/2007 10:50:44 PM , Rating: 1
I myself am not a Intel fanboy but for everyone that is alleging that this bench isn't valid, well. It probably isn't fair but let's remember that Conroe kicked K8's butt well before launch and without final silicon.

But in any case this isn't all that bad, if you can plug in this babies in current servers and double the amount of cores without incrasing power consumption (And as far as I know, this chips are consuming little), it's a win for AMD, because today still, opterons are better suited than xeons for multi-processor environments. Let's just hope frequency rises a bit.


Barcelona should be faster
By fikimiki on 6/7/2007 1:51:00 AM , Rating: 1
I made a calculation scaling down quad-core Intel CPUs in Cinebench.
It looks like Intel should get no more but 670 points at 1600MHz. Based on this Barcelona should be faster 21% and 2.3 GHz should beat 2.66 blue quad.
For Penryn they should have at least 2.7GHz beast.




RE: Barcelona should be faster
By hstewarth on 6/7/2007 10:44:12 PM , Rating: 2
Actually the figure for Intel should be 849 points if scale it.

1274
---- * 1.6 = 849
2.4

My estimated for the stated 2.7Ghz Penryn would be about 1792 using 25% increase on chip to chip speed up.

1274
---- * 2.7 * 1.25 = 1792
2.4


RE: Barcelona should be faster
By hstewarth on 6/7/2007 10:53:35 PM , Rating: 2
One more thing time in seconds are are to estimated but my guess around 10 seconds.

One more thing its likely to be less than 25% chip to chip ( same Ghz speed ) performance increase. But I assume this because larger cache of Core 2's have made a big difference - and the Penryns have more cache than core 2 - plus cpu improvements.


Simplified
By casket on 6/6/2007 10:23:59 AM , Rating: 2
Projection:
A 2.4 Ghz Quad-core K-10 goes the equivalent speed of a 2.3 Ghz Xeon. It gets a 2300+ rating.

AMD's chip is 95.84% the speed of Intel, clock-for-clock. (Assuming the cinebench scores are corect).
***********
So AMD has both an inferior chip and an inferior manufacturing process. This is bad news.




RE: Simplified
By GoatMonkey on 6/6/2007 10:59:11 AM , Rating: 2
They would never use a lower speed rating number than what they have already. They could do 2300x4 for a rating of 9200 if they wanted to.


As Yoda would have put it...
By DeepThought86 on 6/6/2007 6:15:49 AM , Rating: 1
Begin, the flame wars have




By spartan014 on 6/6/2007 7:07:07 AM , Rating: 2
May geForce be with you....!!!


Apple anyone?
By JayDeeJohn on 6/6/2007 12:02:17 PM , Rating: 2
OK, reading this Ive learned something Ive already known. You cant compare it apple to apple. Not with MB issues, socket issues, revision issues and is this going to be a released speed? Id hate to depend on this for anything other than its SOMETHING for info. OK, there is a 1.6 k10, anything else?




By ccbr01 on 6/6/2007 1:55:45 PM , Rating: 2
With the die shrink, Intel is going to be on top again because of the 3.something ghz Core2Quad. I feel sorry for AMD... :(




By jazkat on 6/28/2007 10:35:00 PM , Rating: 2
dont panic you will be gettin an amd cpu,
it was only running at 1.6, also
from amd:

Motherboard Problems

The motherboard we tested on had minimal HT functionality and wouldn't run at memory speeds faster than DDR2-667, most 3D video cards wouldn't even work in the motherboard. Memory performance was just atrocious on the system, but the motherboard manufacturers we worked with attributed this to BIOS tuning issues that should be fixed in the very near future.
the current boards required some heavy BIOS work before the new chips would even work, much less perform up to the expectations, In the end performance was absolutely terrible,

thats why the benches were crap and why amd would'nt let anyone test barcelona, the demos that AMD ran were on its own motherboards and they werent functioning correctly.


By crystal clear on 6/9/2007 12:22:16 AM , Rating: 2
quote:
Intel discusses quad-core processor server market: Q&A with Boyd Davis, general manager of server platform group marketing of Intel


Q: What was the main factor that caused Intel to bring its 45nm Harpertown server processor forward to this year?

A: Intel's product launch is directed by the demand of clients, which has nothing to do with the competition in the market. Currently, product development is almost complete therefore it is natural for the company to launch the Harpertown processor ahead of schedule. The company expects to mass produce the processor in the fourth quarter this year.

Q: In the current server market, competition among CPU makers over semiconductor technology has boosted the migration rate of new platforms. Can the enterprise market catch up in technology?

A: In the past, the server product lineup had a slow migration rate due to the long research time in developing the processor. However, with the assistance of nanometer processing, CPU technology now can advance faster, which shortens the product manufacturing time, sequentially. Intel used to have CPU as the unit for server migration in the past, but is now using platforms. The current platform is expected to last until Penryn processor launches in the fourth quarter.

Q: What are Intel's thoughts about the delay of AMD's native quad-core processor Barcelona?

A: AMD's so-called native quad-core processor has a difficult challenge in technical and manufacturing terms and even Intel would have difficulty facing such challenges. Intel currently still adopts two dies on one chip for its Harpertown processor. The technology is much easier, the product has higher yields and performance is almost the same as the native quad-core processor. Therefore, Intel will not launch native quad-core processors in 2007.

http://www.digitimes.com/mobos/a20070608PD206.html




Quad 2900 XTs Crossfire
By bfellow on 6/6/2007 8:15:45 AM , Rating: 1
Someone said to me for my future file server needs quad crossfire 2900XTs and dual-quad cores to store files. I was like the caveman making a rebuttal in the Geico commercial: Yeah I have one...What?




"If you can find a PS3 anywhere in North America that's been on shelves for more than five minutes, I'll give you 1,200 bucks for it." -- SCEA President Jack Tretton

Related Articles













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki