Barcelona Vs Phenom

[Dudeson]

So we know how Barcelona performs...not as fantastic as speculated so i wanna know how Phenom/Agena?? will perform. With Barcelona being a server cpu they aint goina benchmark games or anything so how will it go against Core2duo? Will the new motherboards (AM3/AM2+) and higher frequency Ram (667>) give the desktop beauties a larger advantage??

 

[zenmaster]

It will be a tough call. We will need to wait and see.

AMD's processor will be able to provide more horsepower by far and be able to come much closer matching the C2D in performance. It may even beat it Clock for Clock, but they will need to work very hard at getting the clock speeds up since there are rumors that even 4.0ghz may be a shipping clock at some point for the up coming 45nm version.

While AMD will close the gap in regards to performance, they will have a tougher time on the price side. These chips will use the same 65nm process but will be a larger die and as a result be more expensive to manufacture. The results is that the bottom end chips may no longer be significantly less expensive than the C2Ds. However, they may still maintain a price advantage on Mobos.

If nothing else, I think these chips will be good enough to ensure Intel keeps prices down and releases chips with higher default clocks.

 

[elbert]

So we know how Barcelona performs...not as fantastic as speculated so i wanna know how Phenom/Agena?? will perform. With Barcelona being a server cpu they aint goina benchmark games or anything so how will it go against Core2duo? Will the new motherboards (AM3/AM2+) and higher frequency Ram (667>) give the desktop beauties a larger advantage??

From the benchmarks of barcelona Phenom should @2GHz compare to a 2.4GHz Kentsfield. This may be a problem for Intel as currently its intels number 2 CPU. No matter which wins as AMD is returning to be competitive as just 1 month ago the best they had was the 8th fastest CPU.

Going against a C2D will make the C2D look like a X2 because at that time crysis, multithreaded for quads, will be used in the benchmarks. If AMD comes out with a 2.7GHz Phenom before penry then AMD may have the top CPU as barcelona seems to have a 300MHz advantage. Intel will be forced to break 3GHz with Kentsfield and at that point who knows as Kentsfield does seem to have the highest clock of the 2. Intel could market a 3.5GHz Kentsfield which would require Phenoms top reported OC of 3.1~3.2GHz to match.

Looking at both I would say AMD will have the second fastest CPU and will push Kentsfield to it max to hold the top CPU. Penry will hold the top 2 CPU's and will be pushed near its max. The big question here is will either company screw up and not have an octal core CPU ready for 2h of next year.

 

[Slobogob]

I don't know why people expect the performance to improve by a large factor because of the new boards and different RAM. Claiming that a 2GHZ phenom will be on par with a 2.4GHZ Kentsfield is a bold claim. The processor was developed for the server segment and we have seen how it performed there. Power requirements will drop a little more, heat will drop a little more and, on the desktop, there will be a little performance increase thanks to faster memory and no ECC.
I would be surprised if faster memory (lower latency memory) would make a big difference as it was with the older am2 K8 processors. The new L3 cache has brought some nasty latencies and i somewhat doubt that low latency RAM will have a huge effect. Higher Memory bandwidth might improve performance a little though. But that's all speculation.
While most people seem to be disappointed by the new K10 i actually like what i see and i hope AMD can offer the Phenom spin off at an attractive price. Competing with the Q6600, as it is currently priced by intel, could be a really ugly affair for AMD.

 

[backy51]

I don't know why people expect the performance to improve by a large factor because of the new boards and different RAM. Claiming that a 2GHZ phenom will be on par with a 2.4GHZ Kentsfield is a bold claim. The processor was developed for the server segment and we have seen how it performed there. Power requirements will drop a little more, heat will drop a little more and, on the desktop, there will be a little performance increase thanks to faster memory and no ECC.
I would be surprised if faster memory (lower latency memory) would make a big difference as it was with the older am2 K8 processors. The new L3 cache has brought some nasty latencies and i somewhat doubt that low latency RAM will have a huge effect. Higher Memory bandwidth might improve performance a little though. But that's all speculation.
While most people seem to be disappointed by the new K10 i actually like what i see and i hope AMD can offer the Phenom spin off at an attractive price. Competing with the Q6600, as it is currently priced by intel, could be a really ugly affair for AMD.

From everything I've read, the K10 core is very data hungry. I suspect that it will benefit from 1066 Mhz DDR2, although I may be wrong. I'm very anxious to see some overclocking reviews. I doubt that this early stepping will OC very well, but we need to get an idea of how we can expect K10 clocks to scale. I don't think we'll be seeing 3.0 GHz quad cores this year. Also we haven't seen the new dual cores yet.

 

[Falken699]

Woah, hold a sec. Let this puppy ramp before you judge the performance, these are fabbing issues NOT architectural chip issues. I am sure if Barcelona was 3Ghz on 45nm in an AM3 board, we'd be seeing a repeat of Opteron in 2003, no doubt. Just on it's dynamic power saving features integrators would be buying this thing in droves.

 

[Reynod]

I think there might be some surprises but at stock speeds Phenom will prob just fall short of the new respin of Core2 ... not by much though.

I'm just thinking gaming tho.

I found a review last night that raised a few interesting points regarding L3 cache on the Barcy ... slightly higher latency that is tied to the memory controller.

I wonder with regards to overclocking the memory controller how much might be gained there ...

Plus we don't know how much headroom it will have - note the voltage for the 2.5Ghz chips suggests that with this revision there isn't likely to be the same stellar result many are getting here on Conroe ... and I imagine Penryn ... for overclocking.

It's all a bit up in the air.

I'll look for that link.

Iv'e got a few AM2's here ... keen to see how a "lazy" drop in upgrade will go ... hopefully the mobo won't have a prob either.

 

[Slobogob]

Woah, hold a sec. Let this puppy ramp before you judge the performance, these are fabbing issues NOT architectural chip issues. I am sure if Barcelona was 3Ghz on 45nm in an AM3 board, we'd be seeing a repeat of Opteron in 2003, no doubt. Just on it's dynamic power saving features integrators would be buying this thing in droves.

The Problem is, the competition is no longer sleeping as it was during the Netburst era. Customers don't care for ramping problems or driver issues. They want a working product and will buy what works. At the time AMD, with all it's troubles, delays and paper launches, reaches 3GHZ on 45nm the competition will be ahead by a long shot.
AMD always has brought something new to the table and most of their products are innovative. The idea to produce a monolithic Quad is good and i really like it, but if manufacturing troubles and delays are the result, then maybe another approach would've been better. I hope they can get the speed up fast, they sure need to.

 

[continuation]

Woah, hold a sec. Let this puppy ramp before you judge the performance, these are fabbing issues NOT architectural chip issues. I am sure if Barcelona was 3Ghz on 45nm in an AM3 board, we'd be seeing a repeat of Opteron in 2003, no doubt. Just on it's dynamic power saving features integrators would be buying this thing in droves.

You're not going to see a repeat of Opteron in 2003. You'll see 8-cores 16-threads Nehalem on 45nm long before you see 3GHz quad-core Barcelona on 45nm.

 

[elbert]

I don't see why people expect performance to not be where it benchmarks at as the 2GHz barcy was holding its own against the 2.33GHz xeon. That was with 667 memory which we know has little effect on the Xeon but does effect X2's and old optys. In reviews we have seen about a 300MHz advantage for barcy so the phenom should compare with a 2.4GHz Kentsfield. Not beat just compare nor be on par as on par would be a bold statement.

The latency is only seen in the low end barcys if you read the anantech review the latency scales with clock. In the comming months the higher end barcys will see lower latency. If you note the charts on this site the lantency goes way down between the 2350 @1.8GHz to 2360SE @2.0GHz. Going from 23ns at 1.8GHz down to 19ns latency at 2GHz leaves a good bit of improvement when it gets up to 2.2~2.4GHz. The 2.6~2.8GHz barcys being the 2.2 and the 3GHz barcys the 2.4GHz. A 2ns decrease per 100MHz and if that holds giving some for diminishing return the 3GHz barcy may only have L3 cache latency of 12~13ns. http://www.techreport.com/articles.x/13176/3

Adding the L3 cache in this way was undoubtedly a tradeoff for AMD, but it certainly carries a hefty latency penalty. This penalty may become less pronounced when Barcelona reaches higher clock speeds. AMD says the memory controller's speed can increase as clock frequencies do.

I cant wait to see if the reported 3.1GHz Phenom was true as thats pretty good for a 65nm quad. Don't worry as I took my spoon full of salt on that 1 as it came from one of the less reputable sites.

 

[accord99]

TechReport's preview with more desktop/workstation like applications show that Barcelona is clock-for-clock slower than Clovertown. I'd expect a 2.4GHz Kentsfield to be faster than 2.6GHz Phenom on the desktop.

 

[elbert]

TechReport's preview with more desktop/workstation like applications show that Barcelona is clock-for-clock slower than Clovertown. I'd expect a 2.4GHz Kentsfield to be faster than 2.6GHz Phenom on the desktop.

The review gives this information for the sake of completeness as he states.

This beta version of POV-Ray seems to have a problem with single-threaded tasks bouncing around from one CPU core to the next, and this causes especially acute problems on NUMA systems. Since the vast majority of the computation time for these scene involves such single-threaded work, things turn out badly for the Opteron 2300s.

From a flawed review the reviewer states they are a MHz for MHz equal.

As a result, Barcelona is sometimes faster, sometimes slower, and oftentimes the equal of Intel's Core microarchitecture, MHz for MHz.

The latency and power test are the only good test from this article.

Check the conclusion in this article where the 4 test estimates the Opteron 2350 (2 GHz) at 2.33~2.4GHz Xeon "Clovertown". A test of Fritz Chess engine was estimated 1.8GHz and 3DS Max 9 at 2GHz. Surprise an Intel optimized Linpack has the 2350 equal to a 1.9GHz Clovertown. Thats a 4 to 3 in favor of the 2350 having a better clock for clock performance. One test had the 2350 equal to a 2.7GHz Clovertown.
http://www.anandtech.com/printarticle.aspx?i=3091

Im not suggesting a 2GHz phenom can beat a 2.4GHz kentsfield but it should be the Q6600 a good run.

 

[accord99]

Even with the potentially faulty POV-ray benchmark, the 2GHz Opteron only beats the 2GHz Xeon in SPECjbb2005 (using a configuration more suitable to the Opteron platform) and Myri-Match while losing in Valve VRAD, Cinebench, Euler, Panorama Factory, picColor and Windows Media Encoder from the TechReport review.

 

[Reynod]

http://techreport.com/articles.x/13176/3

Interesting section on the L3 cache latency.

 

[vsary6968]

I understand how you all feel about the performance. Because AMD benchmarking the 2.0 Ghz Barcelona against the 2.33Ghz of Clovertown and that make alot of different in frequency and perform.Not only that and also it does not support DDR@-1066 memory. And the it does not release the right Stepping revision. When the correct revision was fix, and then you will see about 3% to 5% increase in performance in Floating point and integer. But this not bad at all. This Barcelona 2.0Ghz stepping revision is good for overclocking. But if you want a good performance you have to wait until they finishing fixing and I think it is B3 stepping what it should be. By that time I think they will fix it and support DDR2-1066 memory.

Agenda or Phenom will be a great performance for the Desktop. It will be increase in Gaming. Halo 2 will be about 21% increase in performance better than K8. I am only comparing clock for clock. But I know for sure is that Phenom X2, X4, or FX will be alot better than the C2D. Until the release of the Wolfdale and Yorkfield hit about 3.2Ghz and above. But they are not ready to hit anything over 3Ghz until 1H2008. Only 1 more month left for the Phenom family to be out. I also heard that Phenom FX will support DDR2-1066 but not sure. But I will looking into it.

 

[elbert]

Even with the potentially faulty POV-ray benchmark, the 2GHz Opteron only beats the 2GHz Xeon in SPECjbb2005 (using a configuration more suitable to the Opteron platform) and Myri-Match while losing in Valve VRAD, Cinebench, Euler, Panorama Factory, picColor and Windows Media Encoder from the TechReport review.

You forgot the folding @home where 2350 beats the 2GHz Xeon in 4 out of 6 benchmarks. This is only 1 review and on the anandtech reveiw the barcy is the clear winner. The conclusion of the techreport review was stated

equal of Intel's Core microarchitecture, MHz for MHz.

The anandtech review gives the barcy a MHz to MHz advantage very near the 2.33GHz Xeon. I may be a bit high on my 300MHz advantage but it is an advantage none the less. The Xeon's advantage is a higher clock where it could hit 3.4~3.6GHz. I dont think barcy at 65nm can break 3.0GHz. Phenom on the other hand could but its unlikely of being higher than 3.3GHz.

 

[d4NjvRzf]

Note the CPU-Z screenshot of this benchmark site http://www.techwarelabs.com/reviews/processors/barcelona/index_3.shtml

The benchmarks so far released are probably all based on the B1 stepping, not the B2 stepping that's supposedly capable of higher clocks. In particular, Dave_Graham here http://www.xtremesystems.org/forums/showpost.php?p=2421094&postcount=41 seems particularly optimistic about this new stepping.

 

[accord99]

You forgot the folding @home where 2350 beats the 2GHz Xeon in 4 out of 6 benchmarks.

There is only 4 tests on folding@home, there are two additional "averages" based on the first 4 tests. Plus Amber and Tinker are all but irrelevant as the gromacs core is the only core really being used today.

 

[accord99]

Note the CPU-Z screenshot of this benchmark site http://www.techwarelabs.com/reviews/processors/barcelona/index_3.shtml

The benchmarks so far released are probably all based on the B1 stepping, not the B2 stepping that's supposedly capable of higher clocks. In particular, Dave_Graham here http://www.xtremesystems.org/forums/showpost.php?p=2421094&postcount=41 seems particularly optimistic about this new stepping.

The tested 2.5 at least at Anandtech are stepping B2.
http://images.anandtech.com/reviews/cpu/amd/Barcelona/barc2.5.jpg

 

[Falken699]

Yes, I know it is a manufacturing / margin game ONLY at this point... I guess I am just a bit of an AMD fanboy

I wish they would get TSMC to run them a nice big batch of 45 nm 3Ghz chips and sell them at a huge markup until they can trickle some 2.4-2.6 Ghz chips into the channel to showcase their architecture a bit while it still has a chance to shine.

I heard Intels Fab 38 in the States is a few weeks away from crapping out TONS of Hi-K 45nm chips, so I think this will be interesting to see if AMD can survive the next year

 

[elbert]

There is only 4 tests on folding@home, there are two additional "averages" based on the first 4 tests. Plus Amber and Tinker are all but irrelevant as the gromacs core is the only core really being used today.

I love the word irrelevant. I guess most all the tests are irrelevant as most aren't real world but simulation tests. Some of the Anandtech benchmarks are of real world programs like WinRAR, MySQL, and 3DS Max 9. These 3 programs are relevant in both server and desktops. My son uses MYSQL for a game server so you cant get more relevant.

 

[wolverinero79]

Intel doesn't have a F38. AMD's F30 is being retooled to become F38. Intel's F32 is supposed to deliver 45nm parts. Maybe that's what you meant.

AMD's fabs relate to the number of years since the company was founded. I don't know what Intel's represent other than they're somewhat sequential (he he, if Intel did AMD's way of naming, they'd have to do like F38A, F38B, F38C, etc.).