Bitboys Delivers! ;)

[Nappe1]

at first, Launch information on Acceleon G32, G34 and G40
http://www.bitboys.com/news_events.php?action=read&id=10082004a

Next, NEC Licenses G34:
http://www.bitboys.com/news_events.php?action=read&id=10082004b


I am in a horry, but maybe someone else can make a shorterrned version out of the specs. on the first look, specs for G34 and G40 looks pretty good.

EDIT:
Comparison chart:
http://www.bitboys.com/comparison.php

Lots of cool stuff on G34 and G40 for example,
Internal Precision in G34 10-10-10-10 = (40 bits / pixel)
and in G40 is 16-16-16-16 = (64 Bits / pixel)

 

[sth]

The G40 almost looks like the total overkill for mobile applications ;-) I hope the power requirements are within certain limits.

I hope that OpenGL-ES compatible PDAs and/or Smartphones will enter the market soon. Can't wait to port some stuff over to a mobile platform

 

[Nappe1]

The G40 almost looks like the total overkill for mobile applications ;-) I hope the power requirements are within certain limits.

I hope that OpenGL-ES compatible PDAs and/or Smartphones will enter the market soon. Can't wait to port some stuff over to a mobile platform

It's like Glaze3D of the mobile cores, except now it's available.

 

[sth]

Hehe, exactly.
I hope that there will be some Palm PDAs using this chip (OpenGL-ES will be supported in PalmOS 6)

 

[Joe DeFuria]

I don't mean to be a party pooper, but...

From what I'm reading, an actual chip product does not yet exist...but NEC has licensed the core technology to possibly build into one of their products. Correct? (Or am I missing something.)

In any case...still great news for BB!

 

[Ailuros]

FLIPQUAD sounds suspiciously familiar

From what I'm reading, an actual chip product does not yet exist...but NEC has licensed the core technology to possibly build into one of their products. Correct? (Or am I missing something.)

Sounds like it; it took so far </= 2 years for those kind of processors to get on shelves. The G40 still seems to be the most technically advanced PDA/mobile, even if it isn't yet available in hardware.

I'd dare to guess for 800 MFLOPs/sec for the high end models.

 

[Nappe1]

I don't mean to be a party pooper, but...

From what I'm reading, an actual chip product does not yet exist...but NEC has licensed the core technology to possibly build into one of their products. Correct? (Or am I missing something.)

In any case...still great news for BB!

this is the case also on competitors solutions. (and it didn't prevent anyone here saying how the MBX is this and that, while the closest thing was a FPGA emulator.)

so, the point is, how do you define, when IP core / block actually exists? after all, we aren't talking about "real" stand alone chips in this market anymore.

 

[Ailuros]

this is the case also on competitors solutions. (and it didn't prevent anyone here saying how the MBX is this and that, while the closest thing was a FPGA emulator.)

With the only other difference that "this and that" was multitexturing and programmable VS support more than 2 years ago

 

[Simon F]

this is the case also on competitors solutions. (and it didn't prevent anyone here saying how the MBX is this and that

MBX variants have existed in silicon for about (IIRC) at least a year now.

 

[Nappe1]

this is the case also on competitors solutions. (and it didn't prevent anyone here saying how the MBX is this and that

MBX variants have existed in silicon for about (IIRC) at least a year now.

that's why I said "it didn't prevent..." instead of "it does not prevent..."

Also it's almost a year since NEC announced that they have implemented G10 on their TFT displays. (which happened a just few days before Kristof posted the announcement of MBX being implemented as silicon.)

if you don't believe it, use the search. I am happy to be prooven wrong. (it's the most powerful way to learn. )

 

[Joe DeFuria]

so, the point is, how do you define, when IP core / block actually exists? after all, we aren't talking about "real" stand alone chips in this market anymore.

Again, this is good news for bit-boys, but I am personally (as ImgTech supporters here know. ) product driven. Having licensed the technology is a great step for BitBoys, and I wish them the best, but I personally find it hard to be particularly excited until the tech is productized.

 

[Nappe1]

this is the case also on competitors solutions. (and it didn't prevent anyone here saying how the MBX is this and that, while the closest thing was a FPGA emulator.)

With the only other difference that "this and that" was multitexturing and programmable VS support more than 2 years ago

Still, the point was, does it really matter if it's not on silicon? if I understood Joe right, he thinks it does not.

(it seems that you agree with me that, it matters, if the core / block is available licensing. is it the same as chip on sale for card manufacturers in desktop market, is a different question then.)

 

[Simon F]

this is the case also on competitors solutions. (and it didn't prevent anyone here saying how the MBX is this and that

MBX variants have existed in silicon for about (IIRC) at least a year now.

that's why I said "it didn't prevent..." instead of "it does not prevent..."

I'm still not sure you've chosen the correct phrase. Anyway, it was the juxtaposition of that and the next sentence mentioning MBX that I was unhappy with.

Also it's almost a year since NEC announced that they have implemented G10 on their TFT displays.

I believe I saw it at last year's Siggraph. It seemed to have rather simple functionality but the the AA was quite reasonable.

 

[Nappe1]

so, the point is, how do you define, when IP core / block actually exists? after all, we aren't talking about "real" stand alone chips in this market anymore.

Again, this is good news for bit-boys, but I am personally (as ImgTech supporters here know. ) product driven. Having licensed the technology is a great step for BitBoys, and I wish them the best, but I personally find it hard to be particularly excited until the tech is productized.

don't be scared that I take this an insult or paresonal attack. Not at all. this is actually pretty interesting matter to discuss about, when everyone counts something to be a sellable product.


(besides, I have been sooo long quiet, that this all is like throwing a petrol on the fire! )

 

[Ailuros]

this is the case also on competitors solutions. (and it didn't prevent anyone here saying how the MBX is this and that, while the closest thing was a FPGA emulator.)

With the only other difference that "this and that" was multitexturing and programmable VS support more than 2 years ago

Still, the point was, does it really matter if it's not on silicon? if I understood Joe right, he thinks it does not.

(it seems that you agree with me that, it matters, if the core / block is available licensing. is it the same as chip on sale for card manufacturers in desktop market, is a different question then.)

Yes I do. I just wish (for Bitboys) that their approach would had been as aggressive with their first attempt.

Apart from that announcing that early has IMHO both positive and negative aspects. Positive because you get potential licensees to stick their heads out of interest and negative because the competition knows way too early where you're heading at.

But let's get to the technical point: 2*2 at 200MHz sounds to me like 90nm or smaller, correct?

 

[Nappe1]

this is the case also on competitors solutions. (and it didn't prevent anyone here saying how the MBX is this and that

MBX variants have existed in silicon for about (IIRC) at least a year now.

that's why I said "it didn't prevent..." instead of "it does not prevent..."

I'm still not sure you've chosen the correct phrase. Anyway, it was the juxtaposition of that and the next sentence mentioning MBX that I was unhappy with.

Also it's almost a year since NEC announced that they have implemented G10 on their TFT displays.

I believe I saw it at last year's Siggraph. It seemed to have rather simple functionality but the the AA was quite reasonable.

well, my point wasn't attack on MBX nor it being the first one, but more like that we need some sort of rules here, what's "available". if it is available for licensing then fine with me. Or if it's when product using it is out, also fine with me. Just make things somehow comparable.

MBX is available for licensing as well as soon available as a part of some devices, but that should not shimmer the fact that right now if someone is looking available public specs for licesing deals, G40 is VERY seriously taken competitor for MBX. no question about it. (power consumption numbers aren't available publically for Acceleons, so though I know something about them, I won't take them account here.) Also, situation can change quite fast, if other players introduce new cores in near future.

 

[Nappe1]

But let's get to the technical point: 2*2 at 200MHz sounds to me like 90nm or smaller, correct?

well, amount of pipelines isn't actually told.
(I have some off the record knowledge, but it's better wait if company themselves talks about this.)

quoted Max clock rate is in the comparison table said to be TSMC 0.13µm process. ( check the last line in the table. So that should give you the idea what's the deal with max textures per pixel (Single Pass number?) and also the fill rate.)



EDIT: it's TSMC, not TMSC

 

[Ailuros]

G40 is VERY seriously taken competitor for MBX. no question about it.

Yes there is a question about it. A licensee can build a product today with a high end MBX, while for G40 he'll have to wait for X amount of time.

The first Acceleons were competitors to MBX, while G40 is clearly aimed to compete to MBX2, Imageon2, AR20 (or whatever the competition will call their followup architectures).

Future success of any of those will depend most likely on how advanced each core and at the same time how low power consumption will be.

IMG gloats for some time now with effective fill-rates in the 750MPixels/sec ballpark and up to 3M Tris/sec and that even with a lower clockspeed than G40.

While I'm definitely impressed with what BB seems to be delivering there, I'll withhold judgement on next generation mobile hardware when I've seen them all.

 

[Ailuros]

But let's get to the technical point: 2*2 at 200MHz sounds to me like 90nm or smaller, correct?

well, amount of pipelines isn't actually told.
(I have some off the record knowledge, but it's better wait if company themselves talks about this.)

quoted Max clock rate is in the comparison table said to be TSMC 0.13µm process. ( check the last line in the table. So that should give you the idea what's the deal with max textures per pixel (Single Pass number?) and also the fill rate.)



EDIT: it's TSMC, not TMSC

I'm a layman, yet 130nm and 200MHz sounds quite a tad on the high power consumption side to me (or else someone correct me if I'm wrong). Sounds most likely that the core can yield that max frequency with 130nm, while it can go a fair bit higher and is actually meant for 90nm?

 

[Nappe1]

G40 is VERY seriously taken competitor for MBX. no question about it.

Yes there is a question about it. A licensee can build a product today with a high end MBX, while for G40 he'll have to wait for X amount of time.

The first Acceleons were competitors to MBX, while G40 is clearly aimed to compete to MBX2, Imageon2, AR20 (or whatever the competition will call their followup architectures).

Future success of any of those will depend most likely on how advanced each core and at the same time how low power consumption will be.

IMG gloats for some time now with effective fill-rates in the 750MPixels/sec ballpark and up to 3M Tris/sec and that even with a lower clockspeed than G40.

While I'm definitely impressed with what BB seems to be delivering there, I'll withhold judgement on next generation mobile hardware when I've seen them all.

so, you want compare G40 with MBX2?
okay, fine with me. G40 can be licensed right now, MBX2 not. though it will take X amount of time to get G40 based product out of the owen, it's still Y amount of less time (where Y is the time between G40 and MBX2 launches) than MBX2, if you look at the time from here to product availability.

and what's the schedule for MBX2 or Imageon2? this year? next year? it took exactly 12 months to come from G30 to G40, so I don't see a reason why following generation Acceleons wouldn't be available for licensing in August 2005. (this is a plain speculation of course.)