R360 and RV360

[Joe DeFuria]

"R360" vs. NV40 is the interesting one, IMO. Dammit... to know more about NV40! Might have to go and do some morally questionable things again...

Please do!

Yes, that is the interesting one. This is how I see things playing out (at least in the high-end).

1) Currently NV35 and R350 cores are very competitive, though if the Doom3 benchmarks hold true, the NV35 will be seen as the superior part.

2) It's reasonable to expect the R360 to take back the crown without much question.

3) One would have to assume that the NV40 would be clearly superior to the R360, if it meets whatever specifications nVidia currently has slated for it, and it ships on time.. Contrary to popular belief though, nVidia has a relatively poor track record when introducing brand new cores. They are either late (NV30, GeForce3), or they underperform to original expectations (TNT-1, GeForce256, GeForce3, NV30 in any reasonably shippable form). It usually takes nVidia's first chip respin to get the new core "right". (TNT-2, GeForce2 GTS, GeForce4, NV35).

4) With 0.13 apparently getting more mature, and with ATI apparently basing LOCI off of primarily R3xx logic, it should be a much easier chip to get to market than the NV40. On the other hand, it stands to reason that it won't be as "revoluitionary" as NV40.

So right now, Loci vs. NV40 will be a very interesting match-up to watch indeed. The fact that these companies are off-cycle from each other by 3 months makes for some tough decisions:

* Ati might be able to get Loci out for the fall to compete with the expected release of NV40....but then that would give R360 a very short life span.

* ATI might wager that nVidia will fumble NV40, and not have it in "competitive form" vs. the R360 in the fall. This will give ATI more time to polish and tweak Loci for a winter '04 launch. However, the obvious risk is if nVidia executes properly on NV40...and right in time for the release of Doom3.

This are the things that keep management awake at night.

 

[Bjorn]

This are the things that keep management awake at night.

And keep these forums going

Things have been heating up lately (to say the least ) and it seems it's only getting better with respins here and refreshes there being anounced as close to 1-2 months after the latest releases has hit the market.

 

[Mummy]

How many card makers like Visiontek and so on you guys think will go out of business because of this 3 month cycle mess?

 

[KimB]

2) It's reasonable to expect the R360 to take back the crown without much question.

If the R360 is just now taping out, it is reasonable to expect that it will not be available for about three months. nVidia will be releasing their own fall version of the NV3x core around that time.

 

[MuFu]

2) It's reasonable to expect the R360 to take back the crown without much question.

If the R360 is just now taping out, it is reasonable to expect that it will not be available for about three months.

I don't see why that should be the case - it's obviously not a completely new ASIC.

nVidia will be releasing their own fall version of the NV3x core around that time.

Perhaps.

MuFu.

 

[Mariner]

If the R360 is just now taping out, it is reasonable to expect that it will not be available for about three months. nVidia will be releasing their own fall version of the NV3x core around that time.

I thought that NVidia's next chip was supposed to be NV40 some time towards the end of the year?

It's getting ridiculous the number of 'mini-refreshes' of chips everyone is talking about. It's like waiting for the bus: you wait months for one and then three arrive at once...

Disclaimer: I'm talking about the buses in the UK; I'm sure the bus services are decent throughout the rest of the world!

 

[Joe DeFuria]

If the R360 is just now taping out, it is reasonable to expect that it will not be available for about three months.

Um, right. "Summer '03". Did I say anything different? I expect "launch" in 2 months, and shipping about a month later.

My point is, R360 is a core refresh, (actually, a core refresh of a core refresh) not a brand new core, so there shouldn't be any major issues post tape-out that delay the introduction of the part.

nVidia will be releasing their own fall version of the NV3x core around that time.

I doubt it. If anything, it would come perhaps 3 months after the the R360 is introduced. nVidia is set to ship the 256 MB NV35 in a "month or two" followed by a 128 MB version, and some "value" version of NV35 product.

Are you saying you expect the NV35 refresh chip to launch a month or two after NV35 ships?

nVidia's next "higher-end part than NV35" won't be launched for 6 months. Whether it's a refresh of the NV35, or it's the NV40, is the question. A refresh of the NV35 should be competitive or a bit better than the R360 as one would expect, but that would give ATI until '04 to get Loki out the door to compete with NV40.

 

[stevem]

The only NV3x part mentioned is NV36 which is rumored to slot in between 5900 & 5600 series once 5800 is gone. 5700...?

 

[MuFu]

The only NV3x part mentioned is NV36 which is rumored to slot in between 5900 & 5600 series once 5800 is gone. 5700...?

Yes, that makes sense and has been widely predicted. Could well be logically the best part of a NV35 with a 128-bit memory interface, IMO.

MuFu.

 

[Tagrineth]

Wow.

Isn't it funny how the companies tried to tell us how they were going to do "12-month" and "18-month" cycles, when suddenly BOTH ATi AND nVidia are falling into nV's old 6-month cycle?

 

[MuFu]

Well tech cycles are still going to be 12-24 months. The OEM market needs regular refreshes, even if the retail sector doesn't.

MuFu.

 

[stevem]

Architecture & marketing considerations may allow more product derivatives for each longer product cycle.

Of interest for NV36 is the RAM interface & type. The spin indicates performance closer to NV30 than NV31. Low power may have been mentioned, too.

 

[Nite_Hawk]

I don't think "cycles" really describes the situation right now. It feels more like both of them are in a mad rush trying to beat each other out.

ATI I think made a rather stupied mistake in that the 9800pro should have either been released earlier or been more of an improvement over the 9700pro. If it was released earlier, they'd be at about the right time to announce a second refresh to counter the NV35. If they'd have made the 9800pro more impressive, Nvidia would either have to face stiffer competition in the reviews, or make the NV35 perform even faster (requiring more extravagent cooling, possibly DDR2 memory, etc).

As it stands now, ATI made a pretty weak showing with the 9800pro, and left a nice window open for nvidia to launch the NV35. ATI is going to have to decide if they want to eat being on par (or even slightly behind) nvidia for a while, or quickly phase out the 9800pro in favor of a new refresh. It's too bad, because with some proper planning ATI could have avoided this.

Nite_Hawk

 

[Joe DeFuria]

I I think made a rather stupied mistake in that the 9800pro should have either been released earlier or been more of an improvement over the 9700pro.

Well, the problem with that, is ATI is is not just competing with nVidia, but with themseleves. The faster ATI releases the 9800 and the more powerful it is, that means less profit / shorter life-span on the 9700 = not happy OEM customers. Both ATI and nVidia (and their OEM partners) are between a rock and a hard place. You need a better part to go up against the competition...but the better the new part, the lower return on investment you'll get on the previous part.

There is no clear-cut or right answer. It's more of a balancing act. You don't necessarily want to deliver a part that oblitterates both your competition and your previous part. In fact, to be honest, if ATI could do it all over again, I bet they would have toned down the performance of the 9700! It could have been much lower speced, and still wiped the floor with the Ti 4600...and that would make the 9800 in it's current form a bigger "splash" than it actually had.

Ideally, you want to come up with a product that is "just good enough" so that it is deemed to be accepted as the "better" product than your competition. The problem, of course, is that if you mis-claculate..."just good enough", can easily turn out to be "not quite good enough." Then you have problems.

It's too bad, because with some proper planning ATI could have avoided this.

It's not even all about proper planning. Some of it is pure technical ability to execute, or technical demands from your customers (OEMs.)

ATI seems to be catering to OEM's wishes for generally less obtrusive and power hungry products. No FXFlow, no two-tier cooling, no power-plugs on the mid-range product, etc. nVidia on the other hand, is being forced to go against OEM wishes in that respect in order to be performance competitive. nVidia seems to be betting that end user demand for their product / brand will "force" oems on board despite the less OEM friendliness of their product.

The strategy that ultimately wins out remains to be seen.

 

[Ante P]

"R360" vs. NV40 is the interesting one, IMO. Dammit... to know more about NV40! Might have to go and do some morally questionable things again...

MuFu.

Hmm, a refresh R350 challanging the NV40, doesn't sound very reasonable?

 

[Joe DeFuria]

Hmm, a refresh R350 challanging the NV40, doesn't sound very reasonable?

Right, but the NV40 won't be for at least 3 months after R360. A *die shrunk* R3xx variant (Loci) could be more than a reasonable challenge for NV40...but that might not be for 3 months after the NV40...and the cycle continues.

As far as I'm concerned, I welcome a 3 month "refresh" cycle between the two companies. I'm sure they and the OEMs don't particularly appreciate it though!

 

[MuFu]

"R360" vs. NV40 is the interesting one, IMO. Dammit... to know more about NV40! Might have to go and do some morally questionable things again...

MuFu.

Hmm, a refresh R350 challanging the NV40, doesn't sound very reasonable?

No, of course not - that's why it's interesting.

MuFu.

 

[Seiko]

R360 wont improve in too much in clock speed

As it stands now, ATI made a pretty weak showing with the 9800pro, and left a nice window open for nvidia to launch the NV35. ATI is going to have to decide if they want to eat being on par (or even slightly behind) NVIDIA for a while, or quickly phase out the 9800pro in favor of a new refresh. It's too bad, because with some proper planning ATI could have avoided this.

Nite_Hawk

Yep, I think anyone expecting the R360 to be a far higher clocked part is going to be in for a disappointment. I don't believe for one moment ATI set the R350 at the speeds they did if they where capable of higher clocks. Sure this refresh may add a stable 20mhz more but ultimately it's not going to be much faster on a .15 process. I don't think ATI have anything to be ashamed about as the R3xx has been a breath of fresh air to the industry and forced NVIDIA to up the ante yet again but to expect .15 to continue to deliver is daydreaming imho. If ATI are to compete they need to get there flagship chip on a .13 process and the clock speeds that would hopefully come from that. My only fear is ATI aren't going to be there until R400 and that looks like its delayed until next year and as NVIDIA continue to allow their flagship products to take advantage of .13 maturity ATI are facing a loosing battle. I think the fact that we see the 9800Pro 256 with DDR2 still not pushing 400 is a genuine indication that the .15 process is pretty much at its limit. Unless of course they too opt for a monstrous cooler!

EDIT = typos

 

[Joe DeFuria]

Re: R360 wont improve in too much in clock speed

Yep, I think anyone expecting the R360 to be a far higher clocked part is going to be in for a disappointment. I don't believe for one moment ATI set the R350 at the speeds they did if they where capable of higher clocks. Sure this refresh may add a stable 20mhz more but ultimately it's not going to be much faster on a .15 process.

I disagree.

I see no point at all for refreshing and taping out a new core for something like only 20 Mhz or so more. There's a difference between being "capable" of higher clocks, and being "capble of higher clocks, while meeting OEM power consumption guidelines for the board level product."

380 core/680 DDR-I might simply be the point where they are at the power-consumption limit, even if the R350 core can be pushed a bit higher with only minor impact on yield.

What will be interesting to see, is if the R360 comes with DDR-I, or DDR-II, and how much memory it will be available with. By using "only" 128 MB DDR-II, they may be able to significantly raise core and memory clocks, and still keep within power consumption limits of the current 128 MB DDR-I 9800, and the 256 MB DDR-II 9800.

Personally, I expecting something like 450 Mhz core, and 128 MB of 400 Mhz DDR-II.

I would guess, that if ATI offers a 256 MB version of the R360, (or faster than 400 Mhz DDR-II memort) that card might indeed need some exotic cooling and generally be some "OEM Unfriendly" part...or even a part that perhaps ony ATI sells directly themselves. (Basically just a PR piece for retaining the technology leadership crown.)

...but to expect .15 to continue to deliver is daydreaming imho.

That seems to be a recurring concern with the last 2 0.15u high-end products ATI made....and ATI has delivered both times.

I think the fact that we see the 9800Pro 256 with DDR2 still not pushing 400 is a genuine indication that the .15 process is pretty much at its limit.

Again, disagree...considering that the board likely meets some OEM guidelines for power consumption, and the board as 256 MB of memory on it.

Unless of course they too opt for a monstrous cooler!

My prediction (again): we might indeed see it, but only if there is a 256 MB version of this refresh product.

 

[demalion]

Seiko,
Typical overclock with stock cooling and voltages for the R350 appears to be 420-450+ MHz.

It is apparent that target clock speeds were around 400 MHz for the R350. It is apparent that lower clock speeds from this target allowed it to remain competitive, while retaining more 9700 inventory value, and that they had incentive to be conservative with clock speed and not simply aim for the highest target they could.


Before the R350 shipped, it was assumed that the 325 MHz of the R300 was pushing the limits of 0.15. However, it is clear that ATI has invested significant effort in this direction and has demonstrated impressive abilities in this regard already. This is consistent with their mobility part track record, and further supported by discussion of consultation with Intel with regards to low level design tweaking in association with the R3xx cores.

A further tweaked core reaching 425-450 stock is not at all unreasonable, and proposing the 0.15 process as a reason that this could not be achieved is just a repetition of what was being said about the R300 last year, with no apparent support in the above indications. Please note that the R350 at 380MHz is not showing the signs of being at the limit of the process.

The issue for the next card, IMO, is not the core clock speed, but the memory clock speed.
It is the bandwidth hurdle that presents the challenge for ATI (in achieving a profitable board design), and represents the significant achievement of nVidia once the NV35 ships. I also think that the NV35's bandwidth is why the 256MB 9800 Pro were released at the same clock speeds: to focus resources on the next part to compete well with the NV35 for the important buying "inflection point" that is Doom 3.

Venturing further into rumors and away from observations: There is apparently a 0.13 micron part based on the R3xx core design for the end of the year. Search term: "Loci", for threads discussing it, as your consideration doesn't seem to take it into account.