VOGONS


Reply 42 of 63, by HangarAte2nds!

User metadata
Rank Newbie
Rank
Newbie
Gmlb256 wrote on 2021-09-23, 21:35:
HangarAte2nds! wrote on 2021-09-22, 05:29:
I just got an AGP 6326 today (which is a development of the S3 Virge - for the guy who thought this was different, lol). It is a […]
Show full quote

I just got an AGP 6326 today (which is a development of the S3 Virge - for the guy who thought this was different, 🤣). It is a late production model and appears to have 16MB onboard. There can be issues with these cards in DOS apparently but a later driver release fixed most of the issues. I also suspect that when I get to testing it, it will perform well. If they addressed the issues in DOS with later drivers, it stands to reason the later BIOS may just utilize the drivers better also.
I just ordered a Rage 128 Pro 16MB AGP and I have a TNT2 16MB AGP somewhere if I can find it. I am planning to start benchmarking old GPUs and I want to come up with a protocol that will cover 2D and 3D acceleration and basically give a coherent scale that will cover things from the beginning of 2D acceleration in the 1980s and run through 2007 or so when there was a paradigm shift in the way GPUs interacted with CPUs as GPUs started receiving their own native floating point capabilities. In the old days an 8:1 ratio of system RAM to VRAM was appropriate. But today, GPUs have more raw processing power than the CPUs which aid them and a ratio of 2:1 or so is desirable.
I have the ability to pair AGP cards with a Core 2 Duo (albeit running on one core) and DDR2 memory which will make for an excellent test rig for AGP, PCI and and PCIe cards of the vintage persuasion in DOS, 95, 98, 2000, XP and Vista as there will be no chance of bottlenecking any of the cards I am testing. In fact, being based on the Pentium Pro architecture, and lacking later, problematic instruction sets, I cannot think of a better CPU than a Core 2 for empirical testing of such a broad range of video capabilities. I want to get a database going, like Passmark but for potatoes.
I would not hold out much hope for the 6326 because one article I read said it was 60% of an ATi Rage (not 128) Pro and one third of an NVidia Riva 128. The Rage 128 Pro is considered similar to the Riva 128.
Problematically, the shady computing industry released many cards with similar names but rather different performance. People complain about modern components but it is much better. Even in the late 2000s, this obvious scammery with similar names was still going on. I have 2 7800GT cards and one 7800GTX. You would think the GTX would be more powerful but no. There were times a GTS card might be rather good or really crap. They tell you the GTX 1650 is based on the Turing architecture. But you have to take a deep dive before you realize they lopped off the RTX and AI cores from a defective RTX 20 series. Great cards for the price but you are literally buying their garbage. That is something relatively new and is a product of small process manufacturing.

The only real development based on the S3 ViRGE were the Trio3D cards, I don't know where are you getting this nonsense.

Also the GTX 1650 is indeed based on the Turing architecture but without the raytracing and tensor cores, nothing defective there. This is just market segmentation and low-end stuff were usually aimed for casual or less serious gamers.

Maybe I got my wires crossed on the S3 thing. I tried to find the reference but gave up after 3 minutes because it is easier to just delete a line in my post. They literally take RTX dies which have defects in either the RT and/or Tensor cores but have enough good cuda cores to make a 1650 and just lock out everything else. That's how they were able to sell them for $150 before the shortage. Did you really think they can just stamp out chip after chip and have them all be perfect and identical? That isn't how it works. It is like when they make a Celeron. Do you really think they just go and set out to make budget chips? It was supposed to be something better but had too many defects. So they disable some cores or reduce the cache or something like that, depending on where the defect is. Now I may not have direct proof Nvidia does this to make the 1650 specifically but this is generally how the industry works so I am just going to assume that is what it is because I am a cynic and don't really need proof that people are cheap, greedy and selfish. That is one of the few things I am willing to take on faith.

Reply 43 of 63, by zapbuzz

User metadata
Rank Oldbie
Rank
Oldbie

I used my 6326 to play many dos games a great generic level of dos fun. Sure theres many many fancier gpu's but this one did catch up a bit in its later shelf life with a boot loaded tsr released that injected a later bios. No 6326 has this later injected bios.
Worth looking for.
For those with the ability of updating the gpu bios chip you can with the tsr payload (not the tsr itself) or put it into a system modular bios and remove the vga card bios chip effectively making system bios responsible for the vga vard bios functions (A legit award bios oem feature to save chip dollars) I don't know if that works on agp mobo's so i recommend only pci without professional experience bios modders help.
Windows xp would not allow it to dual support without a signed driver so i managed to find a hacked one. It inadvertently overclocked the 6326 to keep up with an intel chipset graphics controller that worked fine just the 6326 was a bit hotter.
when i went gaming i used dual monitor that streching was potentually dangerous te 6326 kept me informed of system vitals and a good spot for dos prompt message apps media players etc

Reply 44 of 63, by Doornkaat

User metadata
Rank l33t
Rank
l33t
HangarAte2nds! wrote on 2021-09-25, 08:26:

Maybe I got my wires crossed on the S3 thing. I tried to find the reference but gave up after 3 minutes because it is easier to just delete a line in my post. They literally take RTX dies which have defects in either the RT and/or Tensor cores but have enough good cuda cores to make a 1650 and just lock out everything else. That's how they were able to sell them for $150 before the shortage. Did you really think they can just stamp out chip after chip and have them all be perfect and identical? That isn't how it works. It is like when they make a Celeron. Do you really think they just go and set out to make budget chips? It was supposed to be something better but had too many defects. So they disable some cores or reduce the cache or something like that, depending on where the defect is. Now I may not have direct proof Nvidia does this to make the 1650 specifically but this is generally how the industry works so I am just going to assume that is what it is because I am a cynic and don't really need proof that people are cheap, greedy and selfish. That is one of the few things I am willing to take on faith.

If you look at die shots of the Turing GPUs you'll realise the TU116/117 GPUs are not TU102/104/106 dies with parts disabled. They're their own thing based on the same architecture.
What you describe is common practice in the industry but it is not what's happening here.

Btw. This practice is not recycling of rejected silicon ('garbage' as you call it) but rather an intentionally implemented design option that allows dies with certain parts not functional to still be put to use. It is planned during design of the chips that during production different dies will qualify as a different product.
The logic is not "We accidentally ended up with partially defective chips that we now sell at lower price with the defective areas disabled to make a quick buck." but rather "How can we efficiently design our chips for production so that certain common defects during production do not make big parts of a wafer completely useless?"

Reply 45 of 63, by HangarAte2nds!

User metadata
Rank Newbie
Rank
Newbie
Doornkaat wrote on 2021-09-25, 09:24:
If you look at die shots of the Turing GPUs you'll realise the TU116/117 GPUs are not TU102/104/106 dies with parts disabled. Th […]
Show full quote
HangarAte2nds! wrote on 2021-09-25, 08:26:

Maybe I got my wires crossed on the S3 thing. I tried to find the reference but gave up after 3 minutes because it is easier to just delete a line in my post. They literally take RTX dies which have defects in either the RT and/or Tensor cores but have enough good cuda cores to make a 1650 and just lock out everything else. That's how they were able to sell them for $150 before the shortage. Did you really think they can just stamp out chip after chip and have them all be perfect and identical? That isn't how it works. It is like when they make a Celeron. Do you really think they just go and set out to make budget chips? It was supposed to be something better but had too many defects. So they disable some cores or reduce the cache or something like that, depending on where the defect is. Now I may not have direct proof Nvidia does this to make the 1650 specifically but this is generally how the industry works so I am just going to assume that is what it is because I am a cynic and don't really need proof that people are cheap, greedy and selfish. That is one of the few things I am willing to take on faith.

If you look at die shots of the Turing GPUs you'll realise the TU116/117 GPUs are not TU102/104/106 dies with parts disabled. They're their own thing based on the same architecture.
What you describe is common practice in the industry but it is not what's happening here.

Btw. This practice is not recycling of rejected silicon ('garbage' as you call it) but rather an intentionally implemented design option that allows dies with certain parts not functional to still be put to use. It is planned during design of the chips that during production different dies will qualify as a different product.
The logic is not "We accidentally ended up with partially defective chips that we now sell at lower price with the defective areas disabled to make a quick buck." but rather "How can we efficiently design our chips for production so that certain common defects during production do not make big parts of a wafer completely useless?"

More like "we are really stretching our capabilities here to the absolute max of human productivity and we just can't do any better so we'll say it is by design" 🤣 GM gets cracks when they cast an engine block but they don't JB weld it and sell it at a discount! Probably because it is a lot rarer. We have been making them a lot longer. I would bet they only do it because marketing says so. If you left it up to the engineers, that would never happen.
So technically the 1650 is a 1660 super that didn't make it. Same difference.

Reply 46 of 63, by Doornkaat

User metadata
Rank l33t
Rank
l33t
HangarAte2nds! wrote on 2021-09-25, 10:32:

More like "we are really stretching our capabilities here to the absolute max of human productivity and we just can't do any better so we'll say it is by design" 🤣

You may suspect otherwise but this is an intentional decision in the design process and there's nothing wrong with it. There are physical limitations to semiconductor production, yes. That's why designing your chips cleverly in a way that mitigates the negative effects of those limitatios is a good thing. It brings down the cost of the high end chips because this way production cost of a single wafer does not only create a few high end chips but also cheaper chips for customers that need less performance. It just makes sense from an engineering as well as an economical standpoint. I don't see what you think would be wrong with it, especially since at the same time manufacturers are constantly working on refining their production process to reduce the amount of defective chips.

HangarAte2nds! wrote on 2021-09-25, 10:32:

GM gets cracks when they cast an engine block but they don't JB weld it and sell it at a discount! Probably because it is a lot rarer. We have been making them a lot longer.

This analogy does not fit this discussion. First from a technical perspective the process of casting engine blocks can hardly be compared to the process of creating complex semiconductors.
Secondly if cracked engine blocks could be welded and if that would be an economical process that would create a safe product you can be sure they would do that. They don't because they can't.
Lastly even processors with parts disabled are not repaired. They have certain defective parts disabled. The working parts have always worked as intended.
A more fitting analogy would be an engine with cylinders disabled due to production defects.
But if you have an engine that has a cylinder not running you get several relevant issues apart from reduced performance. Even if you use a different cylinder head, crankshaft, ignition timing, fuel injection and so on to get the engine running smoothly on fewer cylinders you still end up with significant excess weight that does not contribute to performance which is not acceptable in a car engine. At the same time the technical drawbacks of extra silicon are mostly irrelevant in a processor die.

HangarAte2nds! wrote on 2021-09-25, 10:32:

I would bet they only do it because marketing says so. If you left it up to the engineers, that would never happen.

Engineers are not only concerned with creating the fastest processors but also with efficiently utilising production capacities. There's a lot of effort in designing processors so that defects in certain areas do not mean the loss of a complete die. This means a manufacturer can sell their top of the line product cheaper and tap into a lower market segment in the same production run by creating less waste. This is an engineering feat, not some scummy marketing malpractice.

Reply 47 of 63, by matze79

User metadata
Rank l33t
Rank
l33t

Well for a 50$ Card what would you expect,
The 6326 is designed as lowcost solution.

If one had to decide between SiS, trident, Virge, rage2 I always would Take latest rev SiS 6326 card
It has proper rendering.
A i740 however will beat them all or permedia 2.

https://www.retrokits.de - blog, retro projects, hdd clicker, diy soundcards etc
https://www.retroianer.de - german retro computer board

Reply 48 of 63, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

I've got an i740, and trying it out on a P3 system back in the day, I'd have put it between an S3 Virge Gx and a Rage II, was not impressed.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 49 of 63, by canthearu

User metadata
Rank Oldbie
Rank
Oldbie

Hmmm, I pulled my 6326 out today ... drivers say chip revision C5. AGP version, 8meg ram.

3dMark99 and Final reality both looked fine, but of course were quite slow. Could not choose 800 x 600 resolution for 3dMark99.

So maybe C5 vs H0 isn't a big deal. Don't have a C3 version to test though.

Reply 50 of 63, by Putas

User metadata
Rank Oldbie
Rank
Oldbie
BitWrangler wrote on 2021-09-26, 20:33:

I've got an i740, and trying it out on a P3 system back in the day, I'd have put it between an S3 Virge Gx and a Rage II, was not impressed.

Problem of that particular system, most probably.

Reply 52 of 63, by canthearu

User metadata
Rank Oldbie
Rank
Oldbie
Ydee wrote on 2021-09-27, 15:12:
canthearu wrote on 2021-09-27, 04:17:

Hmmm, I pulled my 6326 out today ... drivers say chip revision C5. AGP version, 8meg ram.

Where can I see the core revision in the drivers? Thank!

right click on desktop,
choose properties.
choose the settings tab
press the advanced button,
choose adaptor tab.

It will be listed there what revision the chip identifies as.

Reply 53 of 63, by matze79

User metadata
Rank l33t
Rank
l33t
BitWrangler wrote on 2021-09-26, 20:33:

I've got an i740, and trying it out on a P3 system back in the day, I'd have put it between an S3 Virge Gx and a Rage II, was not impressed.

Properly misconfigured.
It was a decent lowcost card.
I had around 50+ FPS on glquake at 640x480. (P2 350)
Try that with virge or rage 2

https://www.retrokits.de - blog, retro projects, hdd clicker, diy soundcards etc
https://www.retroianer.de - german retro computer board

Reply 54 of 63, by leileilol

User metadata
Rank l33t++
Rank
l33t++

glquake's not that good of a benchmark on a 3d card's compatibility (given one blending mode other than alpha is just for the lightmap). You know what also was fast on that? PowerVR PCX2. Does it mean it's on the level of Voodoo2? No.

apsosig.png
long live PCem

Reply 55 of 63, by pentiumspeed

User metadata
Rank l33t
Rank
l33t

Ceramic as in dishes, tiles, mugs etc, glass, etc are made from basic minerals but the basic materials is not renewable sources. This needs to be looked into.

Ditto to plastic, etc.

Concrete too but at least they can be ground up and recycled into new concrete but feedstock minerals needed. Better way is to deconstruct anything into basic minerals in many ways that we can use again.

Same with metal but we have this processes in place to de-alloy and purify to purity standards is easily done.

Cheers,

Great Northern aka Canada.

Reply 56 of 63, by HangarAte2nds!

User metadata
Rank Newbie
Rank
Newbie
Ydee wrote on 2021-09-24, 13:00:
leileilol wrote on 2021-09-24, 00:45:
HangarAte2nds! wrote on 2021-09-22, 05:29:

The Rage 128 Pro is considered similar to the Riva 128.

is it the number???????????

😀))) Yes, number is same...

You guys must not have much to do with your time. That was a typo and I was actually thinking of the RIVA TNT2 M64 32MB. But I guess you didn't take the time to think that through and realize I obviously meant the contemporary card to the Rage 128 Pro.
However, if you want to get technical, the Rage 128 Pro and the RIVA 128 are similar in that they both use a 128 bit data bus.

Reply 58 of 63, by BSA Starfire

User metadata
Rank Oldbie
Rank
Oldbie
sis 305 16mb.jpg
Filename
sis 305 16mb.jpg
File size
28.49 KiB
Views
1053 views
File license
Public domain
Diamond sis 6326.jpg
Filename
Diamond sis 6326.jpg
File size
475.48 KiB
Views
1053 views
File license
Public domain

I have just purchased a pair of SiS AGP cards, so the journey will continue soon, sadly not on the old Cyrix box however as that does not have a AGP slot. For these two we will be testing with either a ASUS CUV/4X & Pentium III 933 or a DFi VIA KT133 and AMD Duron 650.
If anyone has prefernces between the two boxes, do say, they are the last 2 AGP machines I have, so either is fine, either one is as easy to get out of storage as the other!
The first card is a Diamond Speedstar a50 AGP, SiS 6326 8 MB ( I beleive this is the same card as tested by Putas on vintage3d.org, 100 MHz RAM clock and "should" be a HO revision, so I hoping this is the best of the best of the 6326 cards).

Diamond sis 6326.jpg
Filename
Diamond sis 6326.jpg
File size
475.48 KiB
Views
1053 views
File license
Public domain

The next is far more vague, the listing wasn't overly descriptive, but it's the only one at a fair price I've seen recently so we will see what we get, but anyway, apart from the blurry picture we we have a SiS 305 with 16 MB.

sis 305 16mb.jpg
Filename
sis 305 16mb.jpg
File size
28.49 KiB
Views
1053 views
File license
Public domain

Looking at the pic the card looks heavily cut down with 6 empty sets of pads for RAM, so I'm going to guess it'll be "non optimal", but heck if, we were looking for optimal of the era, we wouldn't be looking at SiS anyway!
matze79 said:

Grab SiS 305 Card!

The Laguna is very bad.. i also own it, there is almost nothing running as it should. Expect a lot of drawing errors 😁

Well the Cirrus Laguna wasn't that bad...it did cope with a few things that the C3 SiS 6326 couldn't cope with so it'll be interesting to see how the 305 does.

more when(if?) the cards arrive.

286 20MHz,1MB RAM,Trident 8900B 1MB, Conner CFA-170A.SB 1350B
386SX 33MHz,ULSI 387,4MB Ram,OAK OTI077 1MB. Seagate ST1144A, MS WSS audio
Amstrad PC 9486i, DX/2 66, 16 MB RAM, Cirrus SVGA,Win 95,SB 16
Cyrix MII 333,128MB,SiS 6326 H0 rev,ESS 1869,Win ME

Reply 59 of 63, by BSA Starfire

User metadata
Rank Oldbie
Rank
Oldbie

I also grabbed this card as it was only a fiver, another vanilla H0 6326 AGP.

Attachments

286 20MHz,1MB RAM,Trident 8900B 1MB, Conner CFA-170A.SB 1350B
386SX 33MHz,ULSI 387,4MB Ram,OAK OTI077 1MB. Seagate ST1144A, MS WSS audio
Amstrad PC 9486i, DX/2 66, 16 MB RAM, Cirrus SVGA,Win 95,SB 16
Cyrix MII 333,128MB,SiS 6326 H0 rev,ESS 1869,Win ME