Warlord wrote on 2021-04-08, 22:41:Well lets 1st address your comment on 2d performance. Now maybe my definition of performance is too high of a standard but its […]
Show full quote
ragefury32 wrote on 2021-04-08, 22:15:
Name a single series of chip that has its 2D performance at its power budget and mass pricing during its heyday.
Well lets 1st address your comment on 2d performance. Now maybe my definition of performance is too high of a standard but its pretty much most things that you would want performance in 2d in you would like it to work?
https://gona.mactar.hu/DOS_TESTS/
2nd to address your comment on its power budget. Key work here shouldn't be power but "budget" "mass pricing" which is synonymous with many other words but I'll leave it there. But let me sum it up for you theres a reason usually that something is cheap.
Red herrings aside Cirrus Logic chips come in mind as they were in plenty of low end hardware at the time and in laptops and don't have the same problems with "2d performance" Trident as well S3 made it into many laptops and integrated solutions and they also don't have the same kinda problems. IBM put a lot of rage chips in laptops back then again don't have as many problems and even will run wolf3d which apparently magicgraphs crash on. magic graph is so bad I would take a ATI rage over it anyday and I'm not in the market for a rage chip.
Lastly the notebook I was referring too the XKE isn't some cheap laptop at its time it was pretty much the most expensive thing you could buy so there would be no reason to put a crappy magic graph in it.
First of all, the mactar.hu DOS test list is not about performance but compatibility. You can have great compatibility but terrible performance, like what the Virge series showed in Direct3D 3. Furthermore, its also compatibility for a specific set of games - if you play Commander Keen, great, it's helpful. Or if you watch the warez greetz for prehistorik 2 or play that one Shareware Mario demo? Terrific. Otherwise it's only there to look at certain edge cases, or what hardware elitists use to beat each other over the head over silly minutiae. I mean, if you look at the graphs the Matrox are absolutely terrible for Commander Keen, the warez banners for Prehistorik 2 and Quake. Are you going to toss out a perfectly capable G400 simply because of this?
Second of all, there's only 1 entry for Chips and Technology (65554) and one for Neomagic, which is a Neomagic 256AV. There are 3 families from the PCI based NM2097/NM2160c (Gen 1), the AGP NM2200/2300 series (Gen 2), and the 2380 (Gen 3 with a very rudimentary 3D accelerator). Not all of them behave the same, and not all of them have the same DOS compatibility problems. Hell, I have the Mario Shareware in the 240 and I don't think the slow scroll is an issue in that game. Prehistorik 2 the game itself (not the warez greetz banner, which I skip over anyways) runs just fine, and Commander Keen, well, I don't play that much of it to let it bother me. It's just not my genre.
I am not even sure what you are talking about when you said that Wolf3D does not work on the MagicGraph - I have it running just fine in every level of the game. Hell, there are video footage of a Thinkpad 240 with the 128XD running various retro games with no issues. Duke3D, Quake, Doom, etc. You can probably ask someone here with a Latitude XPi how their retro games ran - they are on the original 128 (NM2097).
https://www.youtube.com/watch?v=SqmL0dJVfPE
I played Quake 2 + UT'99 on software mode, full screen in Win98, and it worked just fine. For regular VGA games like Duke Nukem 3D, Quake, Wing Commander 4, or TIE fighter, no issues there as well. So it sucks with certain games. So stop playing those games and play other games.
Third of all, the entire design of the laptop is not determinant on how well it performs in DOS gaming. It's based on providing a well rounded product within the bounds of a bill of material. The engineers are given a budget, a target size/dimension, a performance objective and a price target, and its their job to provide the most that they can given their limitations and they are not targeting it to run DOS well. It'll need to run Windows 95/98/2000 well since that's what the market had in mind.
Just how much do you think is a mobile GPU going to cost a vendor back in 96/97? In a production run of 10,000 chips, it's only going to cost roughly 35-50 USD per chip going into a laptop retailing north of 3k. When Neomagic first announced the NM2097 back in mid'95 they were mentioning that the chips will sell at 75 per unit. By 1997 they price would've dropped significantly, and considering that the Neomagic GPUs integrate both the GPU and the VRAM, it has a bit of an advantage when it was first launched since the engineers will not have to source separate VRAM chips (at least a few bucks in quantities of over 10,000) and the required interconnects.
https://techmonitor.ai/techonology/neomagic_l … cs_accelerators
Do you know what power budgeting means? It's not a red herring because it's important for laptops. The machine needs to consume only a certain amount of power within a certain runtime which has direct implications on how big the machine needs to be. It also determines how much battery it needs to carry to have a certain runtime. Sure, if your idea of a laptop is a 10 pound waffle iron that's always connected to an outlet and has multiple fans pushing air through the internals, sure, it doesn't matter. On most of those machines with the fast GPUs (like this Thinkpad T42p next to me) their runtimes are measured in minutes even when new . But if your machine is designed as a mobile warrior like the Thinkpad 600s, you want it to use as little power as possible so you can stretch that battery out, you want it to be reasonably light and small, and you want it to generate as little heat as possible so you don't need fans and heat pipes to deal with it - chances are, you are working on documents/spreadsheets or on an IDE coding up a storm. That's why it had a Neomagic. Why were the Neomagic chips were popular? Because they were highly integrated (where the VRAM is packaged on-die with the GPU), you have less traces/real estate on the laptop's motherboard (not the case with the Tridents, the C&Ts and the S3 Auroras). You don't need dedicated extra circuitry to run the external VRAM - I remember seeing specsheets that quote the power usage of a Neomagic at around 750 milliwatts worst case scenario, which is pretty good since it's VRAM+ASIC at the same time. To put it into perspective, a Celeron 300A mobile is around 8 when maxed out. When Neomagic owned 43% of the mobile graphics market, it didn't need to support 3D, and for a chip designed to run Win32 BitBLT, it was good enough.
Fourth of all, that Micro XKE's pricing is deceptive - if we are to believe the Micron Ad on page 1 of Byte Magazine's ad for the XKE, we'll see that:
https://archive.org/details/ByteNovember11199 … w=theater&q=XKE
It's ~5949, which comes with a 5 year warranty (around 250-300 upgraded from a 3 year warranty), and comes with Dragon Naturally Speaking (~150 to 200) and Office 97 Business (closer to 475 back then). So knock it down about 8-900, and it's closer to the Armada 7730DMT in terms of pricing (which has a 2MB S3 Aurora and goes for 4400).
If you take the Vaio PCG-707G (which is much slimmer and lighter and listed at 4500), add a 3 year warranty, it'll be closer to the original listing price of the Micron (5300). That's a Neomagic. If you wait 3 months for the first Dell Latitude CP to hit the market in Feb 1998, spec it out with a 13" LCD display (XT) and a 233MMX (or even an early Dixon P2 mobile), I can guarantee that you'll be paying north of 4300. And I bet there are computer magazine ads on archive.org that will back the assertion.
Also, the Neomagic (on the Versa 7060, the Transport XKE and the Vaio) has MPEG playback acceleration (YUV12 overlay/colorspace conversion) in a single chip. Both the Trident Cyber9397 (on the Gateway Solo 9100XL) and the S3 Aurora (Armada 7730DMT) requires external VRAM and real estate. That stuff takes up more power and costs money and board real estate, and in terms of Win32 acceleration, they are roughly comparable. So why wouldn't Micron use it on their design?
