I'm just curious, have you actually used an NV 8800 card under Vista?

A 5800 is:

1. Late
2. Loud
3. A lousy performer vs the competition.

So far it's only numero uno right? Two more to go. :)

What's up with Kyle anyway? Did DAAMIT "forget" to invite him to tech day? :oops:

Considering that a Core 2 Duo at 2.4 GHZ (E6600) beats or it's almost equal to a 3 GHz Athlon 64 in almost every gaming benchmark and that AMD claims much faster FP performance of Barcelona (and a little higher integer performance, too) over Core 2 architecture on a clock-to clock comparison, I would not understimate the possibility. Barcelona, moreover, is rumored to come out at 2.7 GHz, too (and IMHO it could achieve such speed in this stage with a little "help" in the cooling department)

AMD's claims of much higher FP performance over Core2 are debatable. In peak numbers, both architectures likely trade blows, with some scenarios favoring one chip over the other.

The few relatively hard numbers given by AMD are for SpecFP rate, which is a measure of scalability. This is essentially pointless for the desktop.

Claims of superior per-clock integer performance over Core2 have not been substantiated. It has not been indicated that the per-clock gain is enough to offset a 600 MHz clock deficit.

What is known is that there will continue to be a sizeable clock advantage for Intel until at least 3Q 2007. At that point, it will be reduced to merely a clock advantage, assuming Intel doesn't just up the clock speed grades it is likely holding back.

The only thing I've seen discussed is a dual-core variant, Kuma, that is set to release Q3 2007, a quarter after Barcelona.

Don't worry guys I checked Kyle's 5800Ultra preview and then distiled his conclusions and thus have the first X2900XTX preview:



http://enthusiast.hardocp.com/articl...hlbnRodXNpYXN0

Probably, but on the paper there are advantages in Barcelona's FP implementation over Core 2's one. AMD claim it's maybe tto bold (40%) but it's reasonable to think that there will be probably an advantage. How much this advantage is, is indeed debatable.

I mean integer performance over Core 2 duo, not A64. Note that Core 2 E6600 is almost as fast as A64 6000+, with a 600 MHz disadvantage. Barcelona will be not only a "beefed up" A64.

I mean an overclocked CPU, anyway a 2.7 GHz Barcelona is on the roadmap, not for June, this is sure. But what I want to point out is that Barcelona will be probably at least on par of Core 2 architecture, which is way faster in games than K8 architecture. This could lead to suppose that even a 2.4GHz (or more if oveclocked) Barcelona could be indeed faster in games than a 3 GHz A64.

With this, I don't want to say that R600 is late because Barcelona, but that because it was anyway late, and being the Barcelona launch not too far away, they could have indeed thought to showcase R600 in the tech day with Barcelona (which is already working as seen in the "teraflop" demonstration).

Don't be silly.

Jawed

Just because Barcelona will be improved over A64 doesn't mean it will beat Core 2.
So far, details on Barcelona show it is not as aggressive as Core 2 at extracting ILP. Its prefetchers are less aggressive, it's memory reordering is less aggressive, and its clock will be significantly less.

The cache hierarchy is also more favorable for desktop loads with Core 2.

Little of what is currently disclosed indicates this to be the case for desktop workloads.

Barcelona's biggest advantage is its overall platform, something that doesn't matter with the limited threading of desktop games and apps.

Barcelona is not expected to exceed Core 2's per-core and per-clock performance by an amount that exceeds its clock deficit, this is more important than the number of cores for the desktop.

If AMD intends to overclock a Barcelona chip, it may as well disable half the cores and vapochill the thing. Its results would be as relevant or helpful as using a Core 2 chip.

I take it that you don't even have an 8800 card then? So much for practical first-hand experience to support a theory :D

What did you mean when you said "their Vista drivers are so shit"? What does that mean?

It's not a theory. Does NVidia pay you to post on forums, money or free hardware, anything like that?

It means they're shit. I'm sure there'll be a press release announcing that they're all working dandy at some point and there's no need for the Vista Problems page:

http://www.nvidia.com/object/vistaqualityassurance.html

Jawed

No they don't. Does AMD/ATI pay you to post on forums, money or free hardware?


Mod Edit: Personal remark removed.

It has several advantages in fetching instructions with respect to core 2 (32 bits against 16), has OOO loads and some OOO store, doubles the bandwidth between L1 and L2 with respect to A64, can execute 2 128 bit loads per cycle, etc. etc. . There are no real details on cache prefetching, which was said to be "better" than A64's
So it's probably going to be significantly better that A64 on a clock per clock basis, and if it is still only on par with Core2, it will be anyway better than any present AMD CPU.

This is pointless. Barcelona is the first incarnation of an architecture that should cover any application from desktop (and gaming) to server and workstation market. AMD does not want to make only server chips. If it is not at least on par or anyway very close with Intel's offering, then we can say "bye bye AMD". Reducing prices because a performance disadvantage means making no money.



We really don't know at the moment what Barcelona is capable to do, I anyway agree that probably Intel will keep the absolute performance crown for a while, because process and clock advantages.

Yes, but hey, it's marketing! Intel refused for years to make a confrontation between Opterons and Xeons. And anyway it could be a dual core prototype with higher clocks than Barcelona.

No :smile:

Jawed

While I agree with jimmy in principle, because some of the opinions right now are a tad exagerated (101.41 isn't that bad, imo) - I'll have to point out that my current experience with the 8800GTX in Vista massively indicates how system-dependent the problems are.

On my previous PC, the 8800GTX and Vista under 101.41 worked all fine and dandy. Switch to card to my new PC, and tadam, 3D applications crash within 10 mins and then I need a reboot to get them to work again! Rather ridiculous, to say the least. However, by disabling Aero, everything works fine again. So, excluding that single problem (which is huge, obviously), my experience in Vista with 101.41 has been positive, and even older games worked for me.

But as it is, unless NVIDIA fixes that problem with driver resets, which at least for me was system dependent and only happened with Aero, I'll have to categorize their current G8X driver efforts as "complete and utter shit". Given I've got only a single problem left now though, that could change drastically for me by the next driver release. As it is, though, I'd certainly prefer running a R600 than a G80 under Vista, bah! Would have been nice if it had been released in January for that reason alone in my case... :( Let alone to add some pressure on NV and give an alternative if needed.

That standard would cause this thread to go up in smoke right quick! :lol: (said the 8800GTX/Vista owner)

Now THAT is a great answer, Arun, and exactly what I was interested in hearing about! :)

I could understand if NV is holding back on higher performing drivers until R600 comes out. However, there is no reason why they should hold back on higher stability drivers one would think. They must have been challenged in a big way to not only create super fast and stable drivers for brand new G8x cards, but also for new OS in addition to supporting older cards.

With respect to R6xx cards, it seems that at least publically AMD is very happy with the driver development so far. However, I'm sure that they anticipate many of the same issues. The delay in launch is a good thing for their drivers, even if it keeps the driver team working overtime for a couple months.

I like the fact that consumers have a much better idea of when to receive driver updates [monthly] from AMD/ATI, as opposed to NV's strategy of randomly providing updates.

It will be interesting to see how the drivers compare between AMD and NV around summer time, when hopefully many of the stability issues will be worked out.

Barcelona's L2 caches are smaller than A64's, it will clock 600 MHz less, and it will be in early production silicon at best.
Most of its cores will not be utilized.
Core 2 was a vast improvement over Netburst. K8 was a superior core to Netburst per-clock.
The amount of improvement will not be as much for Barcelona as Core 2 had over Netburst.

For the sake of demonstrating a superior gaming platform, it is likely closer to a tie to the current top-end gaming FXs.

Its desktop equivalents are a quarter past Barcelona's release, and that release is still months after the likely launch of R600.

Barcelona is a quad-core server processor. The idea of using it to demonstrate a superior platform for gaming and for R600 is what is being debated.

I am only discussing Barcelona because that name only applies to that particular chip.

I've already brought up that possibility, which is a better one than saying Barcelona is a superior gaming chip.

Hence "Agena FX". :wink:

Agena isn't expected until a release with Kuma in Q3.

It still clocks lower than Kuma (a better choice, since if they have Agena around, it should be around), but would do better than a Barcelona at 2.4 GHz.

Maybe i'm confusing codenames (probably :D), but isn't there a clear distinction between "Agena", "Agena FX" and "Kuma" ?

From what I've seen, Agena and AgenaFX differ only on the socket they use, and the non-FX version will have more speed grades below the top speed.

Kuma is the dual-core variant of Barcelona, and it has the higher clocks because of the reduced number of cores.

But currently there are three FX speed grades (FX 70; FX 72; FX 74), so why are they going back to a single "halo" CPU again ?

Makes little sense to me, certainly even less when the only difference between an "Agena" and an "Agena FX" will likely be a step up in the speed bin and an unlocked multiplier (or so we hope), and all of them will have to be used with niche motherboards (LGA 1207 will take care of carving it, price-wise).


The last time AMD used a server/workstation socket exclusively for a server/workstation CPU and/or a top desktop part (socket 940) clearly didn't work, or else they wouldn't have devised FX processors for the socket 939. They would've just kept them in s940.

1Mbyte vs 512Kbyte of cache leads to an 3-5% improvement in games on A64 at the same clock. Barcelona caches will have more bandwidth and there will be the other improvements. Core 2 was a major improvement versus Core architecture (Yonah) but Yonah was on the same league of A64 on a performance per clock basis. If we see the same Core-Core2 jump between A64 and Barcelona...

There is a misunderstanding here: Barcelona will of course not be the "AMD gaming superior chip", whay I'm saying is that even if it's a server chip it's probably going to be better (maybe not so much, but definitely better) than current AMD flagship chips in gaming, even if it will end almost surely slower than Core 2 or core Quad Extreme chips. Moreover, I see no great differences between Barcelona and Agena FX chips. What I want to say, is that the showcase IMHO could use a CPU with "Barcelona architecture", even if it's not an Opteron chip.


See above, I refer to Barcelona as an architecture instead of an Opteron chip. Anyway, the dual core desktop parts are surely a quarter late, but as I said before the FX line (gaming chips, accordingo to AMD) of "Barcelona-related" AMD CPUs will be 4-cores, and there is really almost no difference between a 4 core Barcelona and a 4 core Agena FX (except maybe clock speed), according to what we know: both will have coherent HTT links, and same socket: it makes no sense from a production point of view to have a different die. IMHO there are more differences between Agena and Agena FX than between Barcelona and Agena FX. I think that Barcelona and Agena FX are indeed the very same silicon, and the reason of later launch of desktop line is simply that ramping up the production will take time and hence the number of CPU produced in the first months are not compatible with the numbers required for the desktop market. It was the same with Opteron/A64 in the Clawhammer times.


What I'm saying is that probably the "barcelona-like" chip being used for the showcase will be a superior gaming chip with respect to AMD's current line of CPU, not being necessarily superior to Intel's flagship chips.

Agena FX might be meant for the Quad FX platform.

The socket it uses, AM3, also has support for DDR3, and independent power planes, so there are some benefits to the Agena FX core, though they don't look like they'll really change too much for peak performance.

I'm not sure that Kyle's statement could be true in an absolute sense. Sure, there are probably similarities that could be drawn. Both were late and came many months after the world's first DX9/DX10 cards unveiled by the competition, and had some physical aspects that were not as desireable as the competition (such as higher power consumption and/or louder fan).

However, the 5800 never did very well in the DX9 environment that it supported. On the other hand, AMD seems very confident about R600's ability in DX10, even if performance on current games may not be very superior to the competition.

One interesting thing to note is that the delayed 5800 was soon replaced by the 5900. In the last generation, the delayed X1800 was soon replaced by the X1900. This may very well happen with the delayed X2900 as well.