AMD's Barcelona to "stink" in ORACLE - Page 2

Last message on previous page:

ahHAAaaa...there is the point. We dont actually know what its going to do. Certainly it should perform better, but to what level? We can guess, but no one knows for sure until its out.

Think about this. Intels head shed are no bunch of angels. They hyped crapburst so much and for so long that a lot of people had become indifferent (to some level) to their BS. So had the hyping of C2D last year turned out to be nothing more than BS, it probably wouldnt have suprised or disapointed too many people.

During the same timeframe Inel was trying to convince everyone Crapburst was the "it" CPU, AMD built a reputation as a customer oriented company who delieverd what they said they would. Look at whats happend this year. People are dissapointed with AMD over 65nm and DDR2. Why? For the most part, they achieved what AMD had planned for them. Why did people expect more? One possibility, which I believe does have some bearing is hype. Only AMD didnt hype the stuff, the fanboys did. Somehow, somewhere, for whatever reasons, people came to expect more than AMD was going to deliver, and those people were disappointed when they didnt get what they were expecting.

If people spend the next 2~4 months running around hyping Barcelona, above what may achieve, it has the potential to cause the same if not worse disapointment. If everyone gets a 40% improvement set in their heads, and AMD only delivers 20%, people are likely to blame AMD and say they failed. And the ridiculous thing is, theres absolutely nothing wrong with a 20% improvment. I have seen no offical staements from AMD yet. Ive only seen what others have seen, Some unverified specs, some that Uarch analysis you talked about in another thread, "...some guy who works for AMD said some guy at Dell told him....bla bla bla" and a whole lot of fan boys making a whole lot of claims. Did you know the lost soul Sharikou is claiming a 400% increase?Too many people making to many guesses with far to few facts to back them up.....for now.

AMD doesnt need 'help' like that right now. It looks like they are going to continue lowering their prices. Yes, that will cut into thier 'wallet', but they can deal with that. The benefit of regaining the value crown and continuing to expand market share or retaining it or, worst case slowing its errosion, is much better for them long term then a few more bucks right now. Disapointing people could threaten any progress they make doing this.

I didnt say you did. I said you called BS on the unsubstansiated, overestimated claims the fanboys were making before Intel had said anything other than "C2D is comming". And rightly so. Even if C2D had come out with a 40% improvement, at that point in time there was nothing to indiacte they would. It wasnt for several months after the initial announcements that actual specs and benchmarks started showing up. Prior to that time, everything was just BS and guesses. But again, it was Intel, so who cared.


WTF are you talking about? I have never said anyone was lying about or hyping their architecture. Give it up.

How can you compare NetBurst to Barcelona? Barcelona is a redone K8. Unless they screw up everything it can't help but be as fast as the improvements allow, which they are quoting as 1.8x per dual core (fp) and I would say at least 1.4x (int).

Stop confusing me with other people. Oh I know what it is. You just can't stop. Let these companies do what they do and leave your personal opinions out of it.

It's from the medieval period apparently, around 13th century, and at that time it meant 'priest', as well as 'testacles' and 'nonsense' as it does today.... most famously used on the Sex Pistols' album, 'Never Mind The Bollocks', of course.

He is a very inflammatory poster, Baron, isn't he? I don't like to see people bashed, but when you start your post calling people idiots for having certain opinions, what can you expect? Anyway, we've discussed this before so I'll leave it......

Baron, Turpit's post was right on the money with this one.

Second your 1.8 and 1.4 is WRT K8, AMD seems to like that 40% and 80% number, high and almost but not quite unreasonable, and just enough to get your hopes up.... remember 40% improvement in 65 nm that got all excited.... show me where they got 40%???? Show me the data


First, it's AMD's thing not mine. Even Xbit believes the "hype." Second, I'm not sure which 40% you mean though. If you mean switching speed from DSL then I would say we have to wait for Kuma for a dual core to dual core comparison with Brisbane.

I didn't agree with his analysis. You can't compare AMD to Intel or Netburst to Barcelona. Just like you said, it is important to execute well and 65nm has been in retail for a while now. That automatically means ~3.4x (2x for 300mm, 1.4x for 65nm) the amount of chips from the same wafer starts. And in this case this is added inventory. This also means that all of the pics floating around combined with the infamous 100% Task Manager demo, means they are ready for volume.

I expect SPEC benches by the middle to end of the month. Let's see if it happens.

I am comparing netburst to K8...didnt you read? Netburst was crap. K8 wasnt. Comparison complete. K8L is a redone K8. How many people are going to expect more than k*l is going to deliver, because they were bombarded with so many opinions thats when the facts do start showing up they wont see them past the mound of fanboy generated BS?

It is mind boggling that you say this:

to me after I said

I guess you just didnt bother to read. Thats your perogative.

I guess the diference between you and I is that I think everyone should keep their opinions to themselves, while you think everyone should keep there opinions to themselves but you. Thats all you have right now, opinions, because there are no facts yet. Regardless, this is going the same way the C2D pre release went last year, and I guess theres just no stoping it. Que sera sera.

I think everyone should give their opinion, but you have yet to mention the several analyses of Barcelona architecturally and that's what I'm basing my statements on.

I actually wish XBit would update theirs with all of the newest info from AMD. I don't think they mention the sideband optimizations and a few other things that came out later.

But anyway as to the topic, there were a lot of inconsistencies in the blogger's understanding of Barcelona. That was the reason for my first post. I am sure benches will be released soon so we'll see if they are just "hyping" the product.

Though some people wouldnt agree, IMO I would like to see AMD and Intel each with 50% total markets share. Each company would be secure, and it would benifit us, the consumers. Unless the 2 companies started coluding.

Baron,

I think you may have misunderstood what turpit posted.

Turpit wasn't saying that K8L wasn't going to perform like we think it will(what AMD has said it would do, not fanboy claims). Turpit didn't say that 65nm didn't perform(even though we saw minor increases with performance with expensive memory). And he didn't say anything about K8 not performing.

The whole point Turpit was trying to make was that it's not AMD who is over hyping the products, it's us. I don't remember AMD saying we would see any major improvments in the die shrink(I may be wrong, just don't have time to double check). But, if you were to take a look in the forums you would have people making claims of a major increase in performance due to ZRAM or any other of POSSIBLE added goodies. But, AMD never claimed any of these things, all they told us is that they were shrinking the die. And short of frequency increases, why would we have seen performance increases? We wouldn't.... but, many enthusiasts continued to over hype the release. It was our fault.... not AMD's. Also can't remember if they made false claims on 4x4, I seem to remember they did, but don't have time to check.

Now, K8L, if people sit here and say how K8L is going to have a 100% performance increase over Conroe and other crap, well, it will be the same thing. We will be making claims that AMD, the profoessionals, never even made themselves. Hell, they are the ones making the cpu, why don't we leave the pre launch performance numbers to them since they know best.

Turpit wasn't saying anything about you, and not really saying anything about things you said(for the most part as far as I could tell). He was just saying that AMD has been doing what they said they would, with the few minor misteps, while we have been putting words in the horses' mouth.

Intel was making all sorts of false claims about netburst. And it sold alot of cpu's since the public didn't know any better. But, AMD hasn't really been doing that(keep in mind, didn't double check, could be wrong).

Turpit made some good points, and I think he was right on.

wes

I think their success was due to many reasons, and to tell the truth, partly due to the mis-steps of Intel:

1. The Opteron platform, with its cache-coherent serial point-to-point links and independent IMCs, was no less than phenomenal compared to Intel's Xeon fully-shared FSB setup. This was especially apparent in MP setups.

2. The x86_64 ISA ended up catching Intel off-guard as Intel didn't think it would catch on very well with software vendors. Intel also wanted to push Itaniums down into the smaller-server space instead of undermining it with some other (less expensive) 64-bit CPU like a 64-bit Xeon.

3. On the desktop, the K8 was a nice boost over the K7, but it really performed as well relative to Intel's Pentium 4s as it did because the K8 could scale its speed and the Pentium 4 Prescott could not. If Intel could have gotten the Prescott to the 5 GHz that they'd initially planned, the K8 would have been competitive, but not outstanding. Much like how the K7s compared to the Northwoods.

4. Cool 'n Quiet was an excellent addition to the cores that helped out servers quite a bit. Intel's SpeedStep on desktop CPUs *still* hasn't caught up to Cool 'n Quiet in terms of energy savings or scaling options.(It is just full speed and 6x multiplier, IIRC, two speeds- just like 5-year-old PIII-Ms.) When a higher-clocked CPU on a larger process node uses less power at idle than a lower-clocked CPU on a smaller node...that says something. This is an area that Intel would do well to improve upon. It should not be that hard as the mobile chips have an excellent SpeedStep with lots of different graduations and lower idle clocks than their desktop equivalents.

5. AMD was and is generally willing to price their CPUs to match or beat Intel on a price per performance basis. Intel only seemed willing to do that once the Core 2 Duos came out. They wanted a bunch for Pentium 4s and Pentium Ds that performed much worse than an equivalently-priced Athlon 64 or X2- and this was all across the price range, too- not just at the high end where yields and binning might have a CPU be sold a little higher than what its price/performance would otherwise be.

I would love to hear your reasoning about this. I'm not pro-Oracle for everything, but a blanket statement like that has to be justified.

Oracle works for me and has scaled just fine. People need to remember you can't just take an application running against one RDBMS system and drop it on another RDBMS and expect the same results. All database systems run differently.

Why is that when one man (Baron) stood up to say that he believes Barcelona's going to be a beast, a list of others start personal attacking, framing, and shooting out unrelated arguments?

We don't know how Barcelona will actually perform, period. But it doesn't mean the microarchitecture doesn't look like a beast. It does. The shared L3 design is brilliant, plus cc-NUMA make Barcelona a very scalable chip. You don't need to believe what AMD says, but it's not unreasonable to agree with their estimates, either.

And BTW I don't recall AMD ever made a claim that 4x4 would outperform C2Q. AMD claims 4x4 to be better for workstations that perform multiple mega-tasks and/or need high graphics and IO capability, and IMO it's also not unreasonable simply due to the NUMA nature (suppose, of course, the OS & software fully support NUMA). You can argue on and on that nobody here would actually use that much power - well, then 4x4 isn't for those people anyway, and AMD never said it is.

Actually, the architectural improvement on paper establishes the maximum improvement that can be expected. The realistic improvement often is much lower. AMD's estimates are in between theoretical and realistic. They designed the chip, but their production is not final, so they have to extrapolate on paper what their design optimization will result in. And of multiple extrapolations, the marketing department is likely to pick an above-average estimate, if not the highest.

Good & fair points.

All your examples except Prescott do not apply to Barcelona over K8.

1) P4 is little similar to P3. CA AQA by H&P 3rd ed. has a good comparison of the two different microarchitectures.

2) Nothing's screwed up in Brisbane over Windsor. The L2 cache has 2 cycles longer delay, which is probably needed for a. the lower power envelop, b. the .5x multiplier, c. testing future (Barcelona) changes. Brisbane is a nice shrink. AMD promised none other than the lower TDP, and they meet the promised.

3) 4x4 and dual Opterons are in different markets with different price tags. 4x4 stresses on high-bandwidth IO and graphics performance much more than scalability and stability.

Absolutely. That's why I commented to him earlier that it was "charming".

Absolutely. That's why I commented to him earlier that it was "charming".

Wish I had such self-restraint

Yeah, this sounds definitely like an application problem. I've partitioned things many different ways in Oracle, but most of the time, the app presents data as it wants the users to see.

Because he starts posts with phrases such as 'you idiots'.

And I suppose that gives you the right to personal attack him, frame him, shoot out unrelated arguments to him, and spread FUDs against 4x4 and Brisbane? Or does the 'you idiots' phrase make Barcelona worse than it would've been, or make AMD more a liar?

IMHO, he shouldn't have used the word 'idiots'. You can bash him on that however you wish without spreading FUD on 4x4/Brisbane or playing denial to the specs of Barcelona; yet I see no one did the former, but everyone was doing the later.

OTOH, one would react to such a phrase so much probably because it was right on the spot for him. 4x4 and C2Q are different animals, for different purposes. AMD showcased 4x4 by running 4 heavy weight tasks on 4 screens smootly at the same time; can C2Q do so? Can C2Q even do two?

Because he starts posts with phrases such as 'you idiots'.

And I suppose that gives you the right to personal attack him, frame him, shoot out unrelated arguments to him, and spread FUDs against 4x4 and Brisbane? Or does the 'you idiots' phrase make Barcelona worse than it would've been, or make AMD more a liar?

IMHO, he shouldn't have used the word 'idiots'. You can bash him on that however you wish without spreading FUD on 4x4/Brisbane or playing denial to the specs of Barcelona; yet I see no one did the former, but everyone was doing the later.

OTOH, one would react to such a phrase so much probably because it was right on the spot for him. 4x4 and C2Q are different animals, for different purposes. AMD showcased 4x4 by running 4 heavy weight tasks on 4 screens smootly at the same time; can C2Q do so? Can C2Q even do two?

I did bash him on that without spreading FUD.... but I agree it's no reason to slag off Brisbane, in fact I'm really looking forward to it as I'm sure most here are.

But 4x4 is reactionary shite.

Based on the reviews using "mega-tasking", the C2Q can and will complete all 4 tasks faster, or have higher total throughput.

I guess those reviews were performed with 4 graphics cards performing 2+ graphically intensive tasks? Ouch, aren't they?

Or do they simply think like most people on this forum did, that hacking (C2D and C2Q) around for SLI doesn't worth one's efforts? Last time I checked quad-SLI isn't even possible on any Core 2 platform (please correct me if I'm wrong).

Of course you will argue how useless quad-SLI is. I guess people have very different meaning for the term "mega" in in mega-tasking, then. But even under dual-SLI with multiple graphical apps my guess is QFX would also slightly outperform C2Q.

You can't do meaningful quad-SLI on a Quad FX, seeing as you can't physically fit 4 dual-slot video cards.

The only purpose of SLI is for games. Most people aren't going to run multiple games simultaneously. Games are latency dependent, which is why the Quad FX loses to the single socket FXes in games, let alone the Core 2s.

Assymmetric SLI has yet to be demonstrated.

Video encodings won't even touch the video card. CAD perhaps, but unlikely to require much power. And there aren't many scenarios that people will pay for 4 expensive video cards.

Snooping, every memory transaction requires the CPU to check the other CPU first. NUMA doesn't solve this "requirement".

Based on what we see in this thread above... I can hardly say I agree with you about the "most here" part.[/quote]

Yes, you're right, I should have said 'most sensible'

Assymmetric SLI has yet to be demonstrated.

Video encodings won't even touch the video card. CAD perhaps, but unlikely to require much power. And there aren't many scenarios that people will pay for 4 expensive video cards.

Snooping, every memory transaction requires the CPU to check the other CPU first. NUMA doesn't solve this "requirement".

if some people buy 2-3 quadcore processors, phase change coolers, super huge towers, hardccore watercooling for all the parts.. they will buy the quadsli ( its a ego issue and too much money )

http://images.dailytech.com/nimage [...] y-list.png
It seems that it'll take AMD 3 quarters to squeeze out 100MHz to 2.6GHz According to DT, 2.5GHz quads (seems HKEPC is right ) will be the top bin. Unless AMD goes nuts and makes 150W parts.

Question, is quad 8800 possible? And isn't dual 8800 better than quad 7900? And does it really matter how many GPU's are used, if 2 are faster than 4?

I think the major problem is the power envelope......

Indeed.

We have not seen any details about the microarchitecture yet, so we can't conclude.

How? Have you seen it? Do you know how it performs?

cc-NUMA has nothing with the scalability of Barcelona! It has with the scalability of its platform.

I don't believe to their undefined bold claims, until they support them with benchmarks.

They are estimating too much, but they haven't provided any evidence or explanations about what they are estimating and how. For example I think that it is absolutely unreasonable to expect Barcelona to outperform Clovertown by 40%.

1. lower performance/clock
2. lower frequencies

They promised Brisbane to be available in 2006, but they didn't kept their promise.

Yes, QuadFX(not 4x4) is a platform, C2Q is a CPU.

Nope, both are for same purposes desktop/workstation PC's.

So?

Yes.

Yes.

We haven't seen a review with 4 GPUs yet. Have you? What's the big deal about the 4 graphics cards, when only two can be uses in SLI ?

If you mean graphics and video editing, encoding and rendering, then yes. If you mean two 3D games running on two monitors, then no because no one can play two 3D games at same time.

What do you mean by hacking around for SLI?

I think that you haven't checked and you don't know what is Quad-SLI. Quad-SLI was introduced on Intel platform using Pentium EE CPU, the Dell XPS 600. And yes it is possible on a Core2 platform with 680i/650i chipset. So, you may go check. Here is a reputable link: http://www.nvidia.com/object/IO_28569.html

Are you sure?

Oh, please enlight us with your definitions for mega-tasking. We just came from Mars and we just saw a PC for the first time.

Damn! You have no idea what are you talking about.....
1. QFX does not outperform Core2 Quad.
2. For gaming Core2 Quad platforms are wiping the floor with QFX using one graphics card, two graphics cards in SLI and two graphics cards in quad-SLI. Oh, BTW, QFX can have 4 graphics cards but only two can work in SLI. And QFX SLI is weaker then the SLI on the 680i chipset Intel mainboards. On the 680a the one PCIe is x16 and the other is x8. On the 680i both are x16!



Time to wake up. Some Quad FX reviews:
AbcNews
Anandtech
DailyTech
EliteBastards
ExtremTech
FiringSquad
HardwareSecrets
HardOCP
Hexus
Hexus2
HotHardware
HWupgrade
LegitReviews (overclocked FX-74)
PC.Watch
PCPerspective
TechReport
THG
XbitLabs

According to these reviews, Core2 Quad compared to QuadFX is:
1. faster
2. cheaper
3. more overclockable
4. more energy efficient
5. cooler
6. smaller
7. more supported

BTW, If I missed any on-line review, please let me know.

I think that you are consuming too much AMD marketing. That has nothing with the real life.

Because you have no idea how is NUMA working.

Yes I can. But you can also learn by your self. Here is a nice paper:http://www.research.rutgers.edu/~w [...] ulz.pdf.gz.

Is the number ~70% pulled out of your ass?
100% *(1333 / 1067) - 100% = 24.93% != ~70%

Indeed.

We have not seen any details about the microarchitecture yet, so we can't conclude.


How much more details do you know about Core 2?

Let me just ask you, does it use score boarding or reservation stations? How many entries of TLB does it have, with what replacement algorithm? Does it tag caches with virtual or physical addresses, and in what format does it store instructions in its L1 cache?

All you think you know about Core 2 are those marketing-oriented high-level specs - not much different from what we know about Barcelona.

How? Have you seen it? Do you know how it performs?

No, I haven't seen it, much like you haven't seen the shared L2 smart cache of Core 2. Oh yeah, do you think the latter a good design? Don't you?

BTW, Barcelona's L3 is a large shared victim cache. If you still don't know what I mean, then you're not yet qualified for the answer; sorry. :-p

cc-NUMA has nothing with the scalability of Barcelona! It has with the scalability of its platform.


And Barcelona works and supports such a platform, unlike any Intel x86 chips.

I don't believe to their undefined bold claims, until they support them with benchmarks.

Good, I already said you don't need to. Go get your candy now.

They are estimating too much, but they haven't provided any evidence or explanations about what they are estimating and how. For example I think that it is absolutely unreasonable to expect Barcelona to outperform Clovertown by 40%.

We will see whether you think is right or wrong when it's out. You better eat your own post when there are some apps that Barcelona outperforms Clovertown by 40%.

1. lower performance/clock
2. lower frequencies


1. With lower price and much lower power consumption, a 2% less performance/clock is nothing.
2. Be sure it will pick up shortly, and go higher than Windsor.

They promised Brisbane to be available in 2006, but they didn't kept their promise.

It is available, to the OEMs and the channel, but not to the end users. Did AMD call you and say it'll be available to you and your neighbors?

Yes, QuadFX(not 4x4) is a platform, C2Q is a CPU.

So you know, don't you?

Nope, both are for same purposes desktop/workstation PC's.

Still you claim the platform and the processor are for the same purpose? Do you know the difference between a platform and a processor?

So?

So what?

Yes.show us.

Yes.wow, something for you to brag about..

We haven't seen a review with 4 GPUs yet. Have you? What's the big deal about the 4 graphics cards, when only two can be uses in SLI ?
What's not the big deal to have 4 different displays from one machine?
Ah... I know, you really love to buy two or more C2D for the same purpose.

If you mean graphics and video editing, encoding and rendering, then yes. If you mean two 3D games running on two monitors, then no because no one can play two 3D games at same time.

Who plays 2 games at the same time?

What do you mean by hacking around for SLI?

I don't know. I usually find the word "hack" associated with C2D+SLI on the Internet.

I think that you haven't checked and you don't know what is Quad-SLI. Quad-SLI was introduced on Intel platform using Pentium EE CPU, the Dell XPS 600. And yes it is possible on a Core2 platform with 680i/650i chipset. So, you may go check. Here is a reputable link: http://www.nvidia.com/object/IO_28569.html

And that link refers to a Core 2 system? Are you sober?

Are you sure?

Do I have to be?

Oh, please enlight us with your definitions for mega-tasking. We just came from Mars and we just saw a PC for the first time.

How unfortunately, because if so you wouldn't understand. (Read: stop replying like a nutcase.)

Damn! You have no idea what are you talking about.....
1. QFX does not outperform Core2 Quad.

You are wrong. It depends on the application.

Are you sure the structure of 680i is stronger? Because if so, then you are again wrong.

I guess if that makes QFX (the platform) a failure compared to C2Q (the processor), then the whole Intel's Pentium-4 line is a failure, at its initial release, compared to both Athlon XP and Pentium-3.

I think that you are consuming too much AMD marketing. That has nothing with the real life.

I guess your understanding of real life is just too limited.

Because you have no idea how is NUMA working.

No, because I know what is NUMA properly supported, whereas you apparently don't.

Yes I can. But you can also learn by your self. Here is a nice paper:http://www.research.rutgers.edu/~w [...] ulz.pdf.gz.


Why don't you just admit that you can't answer? That paper does not compare memory access latency in multiprocessor NUMA and UMA at all.

If NUMA is properly supported, then remote memory access should be very rare, and mostly from local cache. The latency will be lower than a highly congested FSB.

Is the number ~70% pulled out of your ass?
100% *(1333 / 1067) - 100% = 24.93% != ~70%

You are just ignorant. In the first order, power consumption is proportional to square of frequency. 1.3*1.3 = 1.69.