mystery solved?

[jjayb]

We were all pondering who Anand was talking about when He said:

In fact, one of the companies in today's article is stopping by our labs to show off something special later today. If all goes well we should also have a few guys from Epic over here to benchmark it as well, too bad you won't see the results of that for a little while

In his preview of the radeon 9700:

When ATI visited us in North Carolina just a couple of weeks ago, they mentioned that they were hoping for over 325MHz core and memory but we will see what happens over this coming month.

Was it Ati that he was talking about in the original quote or am I misreading this? Or was it nvidia with the nv30? hence, this quote:

Based on the current specs of NV30, believe it or not, it will be faster than the Radeon 9700.

Sounds pretty confident in his statement. Maybe it's more from having benched the card than going from the specs.


Okay maybe my topic isn't really very fitting. Maybe the mystery isn't solved?

 

[jjayb]

Just reread some of the old topic found here:

http://www.beyond3d.com/forum/viewtopic.php?t=1548&postdays=0&postorder=asc&start=75


and found this:

I can see it now...If we begin to see R300 previews in the near term, Anand tosses in a quote similar to the teaser found in the UT2k performance test..."As you can see, the R300 really does a number against the likes of the Ti4600; however, all is not lost for nVidia. Their next-generation part will also make mince-meat out of this benchmark" (or similar).

Maybe you weren't so far off the mark Typedef.

 

[Typedef Enum]

When we talked about that Anand meeting, I stated that I believed it was NV30.

Don't forget...Anand is located in Raleigh...So is Sweeney and Co...and let's not forget that nVidia does, indeed, have an office in Raleigh.

In addition, Anand did _not_ state in his R300 preview that "NV30 will have to do this-that in order to beat the R300..."

Nope, he stated...

Based on the current specs of NV30, believe it or not, it will be faster than the Radeon 9700.

You can go back-and-forth on this, but I'm simply of the opinion that he's seen it in action...otherwise, I think he would have stated something a little more generic.

 

[Althornin]

Based on the current specs of NV30, believe it or not, it will be faster than the Radeon 9700.

You can go back-and-forth on this, but I'm simply of the opinion that he's seen it in action...otherwise, I think he would have stated something a little more generic.

"based on current specs" just means based on the paper specs he has seen, it should be faster.
Well, we all know the worth of paper specs...

 

[Colourless]

It was ATI for sure. I say this for the simple reason as they would have gone to Anand to so he would be able to publish UT2003 benchmarks in his preview article. Anand's preview also had many more benchmarks than everyone else. It would seem to indicate that Anand actually had access to a card before the launch, unlike everyone else.

-Colourless

 

[Nagorak]

I'm almost 100% certain he was talking about the R300.

 

[KimB]

"based on current specs" just means based on the paper specs he has seen, it should be faster.
Well, we all know the worth of paper specs...

Not necessarily...don't forget that the clock speeds of both products might change before launch. He could have just been referring to that (For example, what if nVidia plans to use DDR2 with a 256-bit bus for the NV30, but they aren't able to get the DDR2 in time?).

 

[Ailuros]

(For example, what if the NV30 uses a 256-bit bus with DDR2 memory?)

I don´t think so.

 

[KimB]

I don't think it's all that likely, either...just a possibility...

 

[Ailuros]

That would be one hell of a last minute possibility then. I recall Kirk being quite clear about it in a recent interview. Bus width it not the be all end all; it depends how you use it.

 

[KimB]

Yes, he did say that he felt a 256-bit bus was "overkill," didn't he?

If nVidia manages to outperform the R300 on a 128-bit bus (even with DDR2), then nVidia will be in an even better financial position than I had argued about earlier.

Some other things to consider:
If nVidia is only working on a 128-bit bus, it would be very challenging for nVidia to realistically compete with the raw memory bandwidth of the R300. However, given nVidia's superior memory bandwidth usage with previous video cards (esp. GeForce3/4) compared to ATI, this might not be such a bad thing.

Anyway, given that Anand seems so absolutely certain that the NV30 will outperform the R300, you would think that the NV30 had a very significant advantage (Only a slight advantage would not warrant such confidence at this stage in development...).

 

[Ailuros]

Given the fact that he babbles about a 256bit bus I doubt he knows anything about specs.

That doesn´t exclude the possibility that he might have seen it already or tested it.

His confidence though could still derive from past track records.

 

[LittlePenny]

However, given nVidia's superior memory bandwidth usage with previous video cards.

I am certainly not doubting the usefulness of crossbar, but I was under the impression that the 8500 HyperZ blocked more pixels than NV's occlusion method in the GF3/4.

 

[Ailuros]

I´m not so sure that it´s actually the case. Someone could eventually dig up some comparative GL_EXTREME scores, but cards should have an equal amount of bandwidth f.e. 8500LE vs Ti4200.

 

[Pete]

I think it was the R300 that ATi brought specifically to test with UT2K3, as it seems Anand is the only one with the benchmark, and all other benchmarking was conducted at ATi's site (everyone flew to ATi, and in two weeks, the cards will fly from ATi to the reviewers ).

 

[KimB]

I am certainly not doubting the usefulness of crossbar, but I was under the impression that the 8500 HyperZ blocked more pixels than NV's occlusion method in the GF3/4.

Well, the 8500's HyperZ did do better when the scene was rendered back-to-front, meaning that the 8500 caches a fair bit of geometry before rendering, and the GeForce3/4 cards apparently don't. I'm not sure that this really helps the Radeon in real games, however, as any good engine will sort the scene from front-to-back as best as possible (they can't do it perfectly without sucking up tons of CPU power...but 80%-90% of the scene ordered front-to-back isn't out of the question).

 

[Hellbinder]

This gets so so old... every topic becomes a debate against the same two people.

Why would the UT guys, and the Nvidia guys all need to "stop by" anand's offices to show a piece of paper???? Um.. ever heard of Email or Fed ex??Its *crystal* clear from reading anand's comments regarding the Nv30 that it is just paper specs at this point. All the current information points to the Nv30 not being anywhere near ready for a few months still. Good grief If anyone got a real NV30 first it would be Epic.. not anand.

It is also clear that they only had a very short time with the R300, and that they spent a lot of time testing Ut 2003.. gee... I wonder what that could mean... No one but Id has seen it yet... ATi stops by... Epic shows up... a few days later we have a Preview of a card that was done primarily on an Epic game where they only had a few hours to test...

I dont understand.. would someone help me with this.. its soooo confusing..

 

[Hellbinder]

If nVidia manages to outperform the R300 on a 128-bit bus (even with DDR2), then nVidia will be in an even better financial position than I had argued about earlier.

Some other things to consider:
If nVidia is only working on a 128-bit bus, it would be very challenging for nVidia to realistically compete with the raw memory bandwidth of the R300. However, given nVidia's superior memory bandwidth usage with previous video cards (esp. GeForce3/4) compared to ATI, this might not be such a bad thing.

Ok this information is flat out incorrect. And i am pretty shocked that its even being presented.

The GF4/4 did not have *superior memory bandwidth usage* This has been discussed many times before. They had a better crossbar memory controler. This is common knowledge. Even JC pointed this out on more than one occation. 256bit memory is not overkill or a waste of time. This simply points out that Anand just does not know what the hell he is talking about, or that Nvidia propaganda has got to him. The New ATi memory controler operates in 4 independant 64 bit data paths. This is clearly going to give far superior performance than any 128bit controler.

I personally think its pretty clear that Nvidia's big advantage (on paper) is clock speed. They are aiming at 400mhz as sure as the world. If they think that 8 pipelines can effectively run on a 128bit memory controler.. then let them try.. Even clock speed wont save them.

 

[KimB]

]The GF4/4 did not have *superior memory bandwidth usage* This has been discussed many times before. They had a better crossbar memory controler. This is common knowledge. Even JC pointed this out on more than one occation. 256bit memory is not overkill or a waste of time. This simply points out that Anand just does not know what the hell he is talking about, or that Nvidia propaganda has got to him. The New ATi memory controler operates in 4 independant 64 bit data paths. This is clearly going to give far superior performance than any 128bit controler.

I don't disagree that it would be very hard for nVidia to outperform the R300 on a 128-bit bus. But that doesn't mean it's impossible. It probably would be impossible to do in all situations, but the NV30 could still come ahead in a few.

This is why it seems to me like nVidia is also going for the 256-bit bus (Actually, I kind of hope they don't...I hope they're going for eDRAM instead...much cheaper, if they can get the clock speed up there...).

Also, if you want to argue that a "128-bit memory controller simply couldn't pull enough data per clock" don't forget that nVidia's crossbar memory controller on the GeForce3/4 cards is actually a 256-bit controller (Well, four 64-bit controllers). It's just the bus width that's 128-bit. This allows nVidia's cards to fetch from memory at higher clock speeds than the core is running at. Similarly, a 256-bit bus width would require a 512-bit memory controller, and so would a 128-bit DDR2 bus.

The main problems facing nVidia if they go 128-bit and DDR2 would be supply of DDR2 and the available clock speeds (which would certainly not be effectively twice the clock speed of high-end DDR).

I personally think its pretty clear that Nvidia's big advantage (on paper) is clock speed. They are aiming at 400mhz as sure as the world. If they think that 8 pipelines can effectively run on a 128bit memory controler.. then let them try.. Even clock speed wont save them.

Well, given the confidence we seem to be seeing in the NV30, it probably has a 256-bit bus width.

I think it will be very interesting to see who brings DDR2 to the table first, though...

 

[LittlePenny]

This is why it seems to me like nVidia is also going for the 256-bit bus (Actually, I kind of hope they don't...I hope they're going for eDRAM instead...much cheaper, if they can get the clock speed up there...).

Let us all hope for Nappe's sake BitBoys is the first to bring eDRAM to the table. A guys dreams can only be crushed so many times.