AT Barcelona previews are up

[Viditor]

Originally posted by: Accord99

Originally posted by: Viditor
And you know this how?

By the virtually no effect on performance when changing the clock speed of the HT bus on current desktop AMDCPUs.

I think you are falling into a trap by equating the two for a number of reasons, but the largest is that we don't know what else the new revision will have (remember that changing to HT3 necessarily requires a new core revision).
For example, remember the substantial improvement the K8 had when they went to the E3 revision (Venice)? Improved performance, clockspeed overhead, memory, etc...
Also, since Barcy has significantly better throughput compared to K8, HT3 should have a comparatively better response.

But Barcelona is not...yet.

All Barcelonas should be enabled.
http://www.anandtech.com/IT/showdoc.aspx?i=3091&p=3

Actually, that doesn't say that Barcelonas are enabled at all...

 

[Brunnis]

Originally posted by: classy
On the desktop I think its pretty impressive to see 15% increase over K8, especially on a server board.

Both the K10 and K8 CPUs were tested on the same board. Testing both CPUs in a performance oriented desktop board would naturally lead to about the same 15% difference, since both would benefit from the faster board.

I don't know. It seems a lot of people are trying desperately to twist the performance results into something good. There's actually little to celebrate, other than the fact that AMD finally have something new out. Sure, performance, while hardly spectacular, is decent in multi CPU scenarios. Things look bleak on the desktop side, which is what I guess most of us here are primarily interested in.

I was personally hoping to be able to go for an AMD quadcore folding and HTPC later this year, but everything so far points to Penryn being a much better choice. I'd rather buy an AMD system, but it seems that I won't be able to justify it.

 

[Phynaz]

Nothing surprising here.

Just as AMD stated in their 2nd quarter conerence call, they will not have the fastest cpu, but are trying to have the most power efficient cpu.

Interesting how they acheived their goal by changing how they measure power consumption though.

 

[Viditor]

Originally posted by: Phynaz
Nothing surprising here.

Just as AMD stated in their 2nd quarter conerence call, they will not have the fastest cpu, but are trying to have the most power efficient cpu.

Interesting how they acheived their goal by changing how they measure power consumption though.

Ummm...NOBODY measures power consumption using TDP numbers.
They certainly have absolutely demolished the power/performance category though, even before the split planes and HT3!

BTW, I have read and listened to the CC many times, and I never heard or saw AMD say they weren't going to have the fastest CPU...could you point that out please?

 

[Regs]

In Viditor's defense AMD does have a few excuses.

Anand used a hack'ed motherboard to use PCI-Express graphics.

Phenom will be a cut-down version of the Barc to focus performance on mostly consumer applications.

AMD is working with DDR2 manufactures for low-latencys which the Barc needs.

Chipset drivers could push performance up from anywhere (realisticly) 1-5%.


It's still really too early to tell and I doubt AMD will be sending Anand a Phenom motherboard anytime before late november.

 

[Phynaz]

Originally posted by: Viditor

Originally posted by: Phynaz
Nothing surprising here.

Just as AMD stated in their 2nd quarter conerence call, they will not have the fastest cpu, but are trying to have the most power efficient cpu.

Interesting how they acheived their goal by changing how they measure power consumption though.

Ummm...NOBODY measures power consumption using TDP numbers.
They certainly have absolutely demolished the power/performance category though, even before the split planes and HT3!

BTW, I have read and listened to the CC many times, and I never heard or saw AMD say they weren't going to have the fastest CPU...could you point that out please?

Sigh...

Henri P. Richard

"As you know, and Hector was pointing this out as well, the focus in the data center switched quite a bit from absolute performance to performance per watt"

About a third of the way into the call, in response to a question from Uche Orji of UBS Warburg regarding the speed of Barcelona.

 

[Sunrise089]

Originally posted by: SexyK
you assume AMD's yields will improve to the point that 3.0 will be commonplace, but intel's yields will not.

Not at all - if that's the impression I gave then it was unintended. I specifically intended to suggest that Intel may be able to offer 4.0ghz chips all day, but NOT do so within a thermal envelope that the retail market will accept. I think Intel will offer the better chip on the high-end enthusiast crowd who provides aftermarket cooling and lots of case airflow when they overclock, but they may not be able to offer a faster chip than a 3.0ghz Phenom in the retail space.

 

[Arkaign]

Originally posted by: Sunrise089

Originally posted by: SexyK
you assume AMD's yields will improve to the point that 3.0 will be commonplace, but intel's yields will not.

Not at all - if that's the impression I gave then it was unintended. I specifically intended to suggest that Intel may be able to offer 4.0ghz chips all day, but NOT do so within a thermal envelope that the retail market will accept. I think Intel will offer the better chip on the high-end enthusiast crowd who provides aftermarket cooling and lots of case airflow when they overclock, but they may not be able to offer a faster chip than a 3.0ghz Phenom in the retail space.

That's grossly ignorant of existing performance of newer stepping Q6600s on even moderate air cooling (nothing exotic). Given that K10 probably won't exceed K8 by more than 20% (best case scenario, probably) clock for clock, AMD would need a 3.4Ghz Phenom to match the EXISTING Q6850.

AMD is down for the count on the desktop. But Barcelona is a nice way of keeping their well-earned lead in the Server arena. If AMD can combine this with a decent offering in their GPU division (read : 2950Pro/XT/XTX), then they can claw their way back to profitability.

Odds? 40/60?

 

[Stoneburner]

Looks like hector ruiz is focused too heavily on the high margin server markets. Is he transforming AMD into another SUN?

 

[Arkaign]

Originally posted by: Stoneburner
Looks like hector ruiz is focused too heavily on the high margin server markets. Is he transforming AMD into another SUN?

Don't knock it if it keeps them alive. I hate to see them descend into utter irrelevance/self-destruction on the desktop though.

 

[Nemesis 1]

Odds 10/90

 

[Regs]

Originally posted by: Stoneburner
Looks like hector ruiz is focused too heavily on the high margin server markets. Is he transforming AMD into another SUN?

If he was doing that he would of spared the company a whole lot of money by not investing billions (yes, billions) of dollars on fabs.

Though I do expect a lot of reconstruction and re organizing next year.

 

[Nemesis 1]

As long as Hector remains @ the Helm of AMD the odds of this ship sinking go up.

Hector has never done 1 thing good for AMD. It was Sanders who started the K8 project not Hector. Hector came into power just befor the relese of K8. But it was sanders vision that brought K8 to market.

Its really not looking good at all when you look @ complete picture. Penryn will bring lower cost parts better efficiency the lack of SSE4 support until sometime after they release SSE5 is going to really hurt amd in the Server as well as the DT .

Penryn is so much more future proof on the DT. One cann't even compare the 2 cores its that bad.

 

[Stoneburner]

Originally posted by: Nemesis 1
As long as Hector remains @ the Helm of AMD the odds of this ship sinking go up.

Hector has never done 1 thing good for AMD. It was Sanders who started the K8 project not Hector. Hector came into power just befor the relese of K8. But it was sanders vision that brought K8 to market.

Wall Street people tend to think backwards though. I bet they absolutely love him. I think it's all the cocaine...

 

[MadBoris]

What gets me is how muddy the waters are.

AMD only allowing people to review over a weekend, really speaks volumes.

If it was an Intel killer reviewers would have had the CPU's for weeks, instead all you can get out of a weekend is muddy reviews, with no clear cut conclusions but alot of speculating is all that we can do. Meh

Although the enthusiast market is a very small percentage of sales...
We'll see what a Phenom can overclock to compared to an OC q6600@3.2. I doubt Phenom will grab my $250, by beating a q6600 with that type of OC, when the time comes.

 

[Roy2001]

Guys, yes, Barcelona is faster clock to clock than C2, Phenom could be at least as good as C2. But they all do NOT matter.

What matter is yield/cost. Look at the K10 die. It is single 280+mm^2, Conroe based quadcore is 2X143mm^2, Penryn would be 2*100mm^2.

If they priced same as Intel CPU's based on performance, AMD would make much less money. That is based on assumption that AMD could release 2.5+Ghz CPU soon with good yield. If not, then AMD would still be in RED. We would still see 500M lost from AMD in next quarter.

 

[Arkaign]

Originally posted by: Roy2001
Guys, yes, Barcelona is faster clock to clock than C2, Phenom could be at least as good as C2. But they all do NOT matter.

What matter is yield/cost. Look at the K10 die. It is single 280+mm^2, Conroe based quadcore is 2X143mm^2, Penryn would be 2*100mm^2.

If they priced same as Intel CPU's based on performance, AMD would make much less money. That is based on assumption that AMD could release 2.5+Ghz CPU soon with good yield. If not, then AMD would still be in RED. We would still see 500M lost from AMD in next quarter.

Not true, look again. Barcelona is ~15% faster than K8, which in turn is ~20-30% SLOWER than C2. So, clock to clock, Barcelona/Phenom will probably be 0-~10% SLOWER than C2 clock for clock.

 

[sdsdv10]

Originally posted by: Regs
In Viditor's defense AMD does have a few excuses.

Anand used a hack'ed motherboard to use PCI-Express graphics.

In Anand's defense, he had no choice as it was AMD itself that sent server MBs to preview/test a desktop CPU.

AMD was kind enough to send us two servers, identically configured, from Colfax.

It should also be noted that there wasn't any software hacks. The simply removed the PCIe riser (which converted the x16 connection into two x8 connections) and removed the GPU back plate to allow it to fit. No beta drivers, no hardware mods to the MB or modifications of the GPU board itself. What you wrote implied they altered the software or hardware to allow it to work. All they did was remove some pieces so they two items could be fitted together. I'm sure Anand and company would have much rather used a desktop MB to preview the Phenom chip, but they didn't have access to one [at least not one they could talk about yet ]

 

[Roy2001]

Originally posted by: Arkaign

Originally posted by: Roy2001
Guys, yes, Barcelona is faster clock to clock than C2, Phenom could be at least as good as C2. But they all do NOT matter........

Not true, look again. Barcelona is ~15% faster than K8, which in turn is ~20-30% SLOWER than C2. So, clock to clock, Barcelona/Phenom will probably be 0-~10% SLOWER than C2 clock for clock.

That is not true, or at least we are ALL not sure. AT is using a server chip to simulate desktop CPU in non-optimized situation to give us some general idea. So it could compete with C2D, or close to. So the actual Phenom could be faster with desktop applications.

My point is, if AMD could crank up the frequency to 2.5+Ghz with reasonable power consumption and yield, it would be in better situation. Otherwise, it would not change the picture fundamentally. AMD would still be in huge red.

Yes AMD has a few cherry picked dies run @3Ghz, but how about the yield? AMD is already lag with the 280mm^2 die. If yield is low, then it will be a disaster for them as Intel would continue the price war. It could end up that AMD loses more money with K10 than K8 if Intel aggressively ramp up Penryn with higher frequency.

 

[Stoneburner]

I can see why everybody is focused on the phenom "preview," but isn't the chip clearly faster clock per clock than intel solutions in the server space? In fact, it looked like it handily won the comparison with the xeons. So these aren't dissapointing processors per se, but they may not be great at desktop needs. However, we don't know the second part for sure. So as it stands.

1. These are great chips for the server market.
2. These may/may not be great chips for the desktop market.

I'd urge everybody to avoid conflation.

PS: Gary key has already hinted a phenom preview on a rd790 board is coming soon. We may have more concrete answers before December.

 

[SerpentRoyal]

With more data to digest, I still stand behind my initial prediction that an overclocked Intel quad is 15% faster than an overclocked AMD quad. Overclocking a chip yields a more accurate true potential of these CPUs.

Of course AMD now has a quad that can compete with Intel based on price/performance. For the average non-overclocking PC user, this is a good thing.

 

[Nemesis 1]

Originally posted by: Stoneburner
I can see why everybody is focused on the phenom "preview," but isn't the chip clearly faster clock per clock than intel solutions in the server space? In fact, it looked like it handily won the comparison with the xeons. So these aren't dissapointing processors per se, but they may not be great at desktop needs. However, we don't know the second part for sure. So as it stands.

1. These are great chips for the server market.
2. These may/may not be great chips for the desktop market.

I'd urge everybody to avoid conflation.

PS: Gary key has already hinted a phenom preview on a rd790 board is coming soon. We may have more concrete answers before December.

Could you link or copy paste were K10 is faster than C2D. Because I really don't see what your referring to . I may be wearing blinders but that would go against my nature.

 

[nyker96]

I think Barcelona isn't strong enough to beat C2D with these results but it is probably good enough to just stay in the game with Intel which for AMD is important at this point.

 

[kknd1967]

not really faster than intel clock per clock if you read this
http://techreport.com/articles.x/13176/1

Originally posted by: Stoneburner
I can see why everybody is focused on the phenom "preview," but isn't the chip clearly faster clock per clock than intel solutions in the server space? In fact, it looked like it handily won the comparison with the xeons. So these aren't dissapointing processors per se, but they may not be great at desktop needs. However, we don't know the second part for sure. So as it stands.

1. These are great chips for the server market.
2. These may/may not be great chips for the desktop market.

I'd urge everybody to avoid conflation.

PS: Gary key has already hinted a phenom preview on a rd790 board is coming soon. We may have more concrete answers before December.

 

[Stoneburner]

Originally posted by: kknd1967
not really faster than intel clock per clock if you read this
http://techreport.com/articles.x/13176/1

This looks more and more like the k8 launch. From what I remember, at launch the k8 was trading blows with the p4c. The k8 architecture steadily improved, however, while intel dropped the ball with prescott. I doubt intel is gonna drop the ball again this time.