It looks like you haven’t enabled AGP 2X in the BIOS. Also, if you lower the resolution, you’ll be able to correctly benchmark the CPU, as it won’t be bottlenecked by the GPU.

@PCRetroTech

3 жыл бұрын

Well that would certainly explain some things. I don't know how I overlooked that, or why it isn't the default on this board. I'm certainly going to have to do a followup video now aren't I. :-)

@Dxceor2486

3 жыл бұрын

​@@PCRetroTech I was going to say that too once I reached the end of the video ^^ I think it's disabled by default because the AGP implementation in these boards is rather poor and can have lots of issues, and this is one of the way to fix them. I have a Asus ALi motherboard and a Shuttle and a DFI via motherboards so I could do some benchmarks if you want to.

I love that board!!!! I had one for two years on a (sadly) lackluster 200MMX. Thankfully the owner after me took it to AMD K6 II 500 mhz.

Now I realize, how back them evolution of hardware was insane. 96 - Voodoo 1, 98 - Voodoo 2, 99 - Gforce 256, 00 - Gforce 2, 01 - Gforce 3. What was that year High End, next one almost couldn't run new games.

@billhart3814

2 жыл бұрын

Yeah, those were crazy times. The companies were also trying to outdo one another on price as well, so there were all sorts of el cheapo options hanging around at the same times as the latest parts as well.

@petertorda5487

2 жыл бұрын

@@billhart3814 Right, for around 600$ you have got absolute newest high-end card (2001 GF3), not like today some mainstreamish version.

@JohnSmith-iu8cj

Жыл бұрын

Talk about a 14k DM 1995 PC that was worth 2k DM in 1997

I'm sure that one of the via of the big trace is not connected to the back and is just connecting through the board meaning you can actually solder it from the back (I've personally never seen a board with caps sitting flush against the PCB being made differently). What may have caused trouble however is the fact that these planes are really big, making it harder to solder things to if you don't have the right tip/soldering iron/heat. One of the things pointing to that idea are the burn marks on these pads.

@PCRetroTech

3 жыл бұрын

Yes, I think the via goes all the way through, but there's not much real estate to solder to. It was much easier to solder to the other side. I was also a bit worried about introducing a short by soldering the back.

@Dxceor2486

3 жыл бұрын

@@PCRetroTech the trace might have been damaged by the previous owner by heating it for too long while not being able to melt the solder. Now unless the copper have been exposed I doubt you'd have had a short there.

Ah memories, I had a similar ALI board with a K6/2 450 back in the day with a Banshee. The AGP on the board was flakey but it was a good performer

@PCRetroTech

3 жыл бұрын

Yes, I've read the AGP implementation on ALI boards is flakey. I did have some hangs in the PCPlayer benchmark (after it finished), but otherwise haven't actually experienced any problems so far.

i've always had issues with VIA chipset AGP support under windows 95 and 98, not very stable in combination with some implementations. ALI and SIS was generally pretty good, stable, and compatible in my experience. VIA got much better later on, and i had a couple years of wonderful experience with windows XP and the Slot A chipsets (KT133, KT400, et c). i think i still have a perfectly running Soyo KT333 Dragon Ultra Platinum that's an absolute beast.

I have almost the exact same motherboard. Cool video!

Still have this same Gigsbyte board with an k6-2+ @550 for years now last time I used it still fine…

It's the VIA AGP GART driver included in the VIA 4in1 drivers causing your performance issues. I tore out all my hair on this 20 years ago with a MVP3 board.

The CPU is 100% limiting the graphics card. I get 7300 in 3DMark99 with a Pentium III with my GeForce 2 GTS. Thanks for the great video. I just picked up a GA-5AA last week.

You could try out the utility TweakBios 1.53b. This allows you more tweak settings for the chipset, which you can't see in the bios.

@PCRetroTech

3 жыл бұрын

Cool!

Those boards go over a wave solder machine in production, there's no way a cap lead would not be soldered on the underside. Not all caps bridge across *visible* voltage planes. Sometimes, one leg connects on the opposite side or internally to its plane. It certainly wouldn't be soldered from the top side of the board. Most motherboards of that time would be at least 6 layers, some as many as 8. I you look around the edge of the board, you may find a legend which is a series of boxes with numbers on them running along the edge for about an inch, each one being deeper and deeper in the board. A 6-layer board will have layers 1-5 marked. If you look down in the corner where the CPU socket is, next to the retention clips of the DIMM sockets, you might find it at the edge of the board there.

@PCRetroTech

Жыл бұрын

You may be correct, but for example I've seen AGP sockets soldered on both sides. I don't know how this would have been done from a manufacturing point of view. It sure is a pain to remove them though.

@Fredjoe5

Жыл бұрын

@@PCRetroTech That would be a weird-looking socket, are there any images of that online? I'd be interested in seeing one.

@PCRetroTech

Жыл бұрын

@@Fredjoe5 I doubt you'd see it without a closeup. I couldn't tell if the solder had gone through the vias. Maybe that's how they did it.

@SianaGearz

Жыл бұрын

It's so funny that we have gone backwards. My mainboard today a B450 is only 4 layers, i was almost knocked off my chair when counting them.

@Fredjoe5

Жыл бұрын

@@SianaGearz Right, it's the higher integration of modern boards, which minimize and lot of interconnecting, leaving just the high-speed serial busses. RAM is really the only parallel bus left these days.

a video card and ram upgrade would be interesting to see. without using a voodoo card. most of us cant afford a voodoo card. going 8 to 32, 64, and 128mb with max sdram.

Nice board, socket 7 with agp

Hello, the Ali chipset only supports AGP 1x . That's why I switched from this mainboard to a DFI P5BV3+ Rev.C with Via chipset, much better results with a Voodoo 3 2000 (although the Voodoo doesn't support Agp) Greetings

@SianaGearz

Жыл бұрын

Voodoo supports AGP, it only doesn't support the AGP GART address remapping feature, so it cannot hot-swap data from main RAM on demand. The interface connection uses AGP signalling.

Set the multiplier to 2x for that K6-2+ and it'll run at 600Mhz ;) For Super Socket 7 the maximum you'll want is a Voodoo Graphics or Riva 128 from '97 or at the very most Riva TnT, ATI Rage Fury or single Voodoo 2 from '98. The CPU will bottleneck anything faster. a Gefoirce 256 will probably squeeze a bit more performance out of it, but they are stupidly overpriced these days, grab a Geforce mx400, which performs virtually identically. That 3DMark2000 score is incredibly high for an AMD K6, absolutely nothing to scoff at. When we host the Retro LAN's here in Sweden, our rental PC's have a goal of 2000 points in 3DMark 2000, which is roughly what you'll need to have an enjoyable experience in most games from 1999 and earlier.

@PCRetroTech

3 жыл бұрын

Awesome. I'm looking forward to a followup and doing some tweaking and overclocking.

@BadManiac

2 жыл бұрын

@ch282 if you want to attend one of our retro LAN parties but don't have your own retro PC, you can rent one from us, we have 50+ 1999 era PCs ready to go.

Super socket 7 with a k6-2. A computer that I never had but perhaps could have if had known a thing or 2 about computers back in 1998. Of course, it being 98, the GPU would have been a much slower / older than that. A voodoo1 initially, most likely, possibly upgraded to voodoo3 in 99.

Maybe the "Pentium" board is just not lying There is no 1.5x multiplier on Super Socket 7 boards, as it is exactly same as 3.5x and it depends on CPU, not board

i have that same chip set on an ATX ver.. run that intel 233 at 100Mhz FSB at 2.5x running the intel chip at 100Mhz FSB makes a huge Difference.

For the speed improvements I heard that the ALI chipset had a better AGP implementation in some areas compared to the VIA one. PhilsComputerLab's benchmark suite has also some programs that might interest you which are supposed to up the speed massively the video throughput. You might want to use CPU only benchmarks to check this out and also try a PCI video card in both boards. As for the "hiccups" caused by the HDD, these shouldn't happen I think. Have you checked the "dma" mode of the hdd in windows 98 ? This should prevent this from happening.

This gigabyte GA-5AA board uses an AMI BIOS, however, my GA-5AX (which is basically the ATX version of that board with an extra pair of PCI slots) has a Award BIOS. Would be interesting to see which BIOS is faster and has more options on the chipset feature page. Something for me to do when I get my workshop set up.

@PCRetroTech

3 жыл бұрын

Yes, note that the AMI BIOS on this one is quite similar to the look and feel of the Award BIOS. They eventually got the message that their interface sucked and switched, right around when this board came out.

@MegaDoraibu

3 жыл бұрын

@@PCRetroTech Yeah ,I hated the AMI WinBIOS, had to put up with it on my Supermicro socket 7 boards. I also have a no-name clone 486 ALi chipset board with it, so I flashed an Award BIOS from another ALi board (Terminator SMT) to avoid it and got even better performance too!

Hi - Good video Yes strange results, i find Phil's kit quite good, i did use those 2 test programs ( the bouncing HAL and the chess board and little tank ) on 286/386's back in the day, if anything PC player benchmark runs ok and gives reliable results but the 640*480 on some machines / video cards throws video sync corruption - there is a set of vesa tools to try and install and re-run programs, some can have a extra 20-30% improvement Bad cache or different levels of L! & L2 makes a very big difference as does the video card used, Check with Phils webpage not youtube he has a spreadsheet on submitted results from viewers, you can download it and browse / filter it. The variety of bios options can make or brake anything, memory branding, speed, memory timings, wait states, they all add up, what work in one board does not always work in another. With Win98 benchmarks, i actually turn off virtual memory as the read/write can make things sluggish as you experienced with one of those benchmarks that was stuttering for no reason - actually i would run then all with no virtual memory, the do another pass with fixed extra virtual memory say x4 of total ram in system - but never go past the unofficial 512meg limit To eliminate the hard drive issues, try a compact flash card say or 4 gig CF card, you should have no hard drive slow down, as these format under dos 6.22 to the max 2Gig very fast - they run at almost the IDE full bus speed Regards George

@PCRetroTech

3 жыл бұрын

Thanks for the suggestions.

RAM controller having that large impact. For reference, test an AMD XP 2000+ on a normal VIA single channel motherboard then test it on a NVidia NForce (possibly NForce2) with dual Channel RAM controller, the difference in performance is akin to getting a CPU upgrade and you only changed the motherboard. Of course one board price was around $70 the other $250. It was still be best performance impact from a single upgrade not expected to have such a large impact (was planning for a later CPU upgrade and needed the board support, ended up delaying the upgrade to the next generation as it was no longer needed. Yes that much of a performance difference)

@30:15 - You're surprised at the wrong moment. That 40-something fps is 3dbench was very-very low for a Pentium MMX. Something was off with the first setup, really. 3DMark scores aren't _that_ different, if you look this way. And if you enable byte merge in BIOS you'll get even better with the ALI. @36:54 - I'm not sure anymore, your K6-2+ 3DMark score is already very nice. @40:00 - So the Windows drivers messed up DOS performance. Nice one. It almost looks like they tried to actively badmouth DOS in favor of Win98 :D I don't understand the PCP score, btw. What video mode was this? Ah, mode 101. OK. It must be very video bandwidth limited but it needs another bad setting for sure. Just try that byte merge and K6WC (software) too on the K6-2+. At 500MHz it should do much better. Near the end - wait, did cachechk just measure the 128k L2 of the K6-2+ the same speed as the onboard L3??? What the...

@PCRetroTech

3 жыл бұрын

Yeah I will definitely give that option a go along with the software. There'll be a followup video for sure. :-)

Great video! That GF2 GTS is severely CPU bottlenecked in both cases - Which is great for this video, because it shows clear CPU scaling. But GF2 is really for an Athlon 700+ or a PIII close to 1Ghz, and it scales a easily into P4 systems. If you like to keep it period correct, something like a Banshee/Voodoo2/TNT for the P233 and a Voodoo2-3/TNT2 for the K6 would be the way to go...

Lucky Goldstar changed or actually shortened their name...perhaps you've heard of them....LG 😉

@PCRetroTech

3 жыл бұрын

Unless it was a typo, this was Lucky Star Technology Co Ltd, not Lucky Goldstar.

@rasz

3 жыл бұрын

@@PCRetroTech Taiwanese company, I think they went out of business around 2003

@SianaGearz

Жыл бұрын

No it's not the same company. We had dealt with lots and lots of Lucky Star boards in the 90s, because they were the cheapest. They were terrible and flakey, the very absolute bottom of the barrel.

With all that quèche talk, my timings are getting hungry...

So wait this board has both AT and ATX power connectors? Weird

@PCRetroTech

3 жыл бұрын

Yeah it was common enough for a certain small period of time.

@mindphaserxy

3 жыл бұрын

@@PCRetroTech I don't remember it but my PCs from the 1990s era were pre-built

@Fredjoe5

Жыл бұрын

Many of the later transitional AT-style boards had them, although they usually didn't have the soft power on feature of the later ATX boards.

@6:40 it might be even 6 layers. 386 boards were already 4 layer. Proper late 1999 early 2000 poorspec, or maybe even foolspec. K6/2 400 was more expensive ($200) than Celeron 300A ($150), while slower even before overclocking the Celeron. Celeron overclocked to 450MHz using nothing more than a snippet of electric tape to block 4 pins absolutely destroyed any K6 2/3, and as a bonus you could use 440BX board with actually usable AGP. Those Super Socket 7 motherboards werent even cheap back then, often sold as a premium upgrade option and more expensive than rock stable 440BX/ZX models like Aopen AX6BX or Zida ZX98-AT. Celeron 300@450 gets 3800-3950 points (depending on ram speed settings) in 3dmark99MAX using TNT, Voodoo3 2000, TNT2, or even TNT2 Ultra. Guess its all CPU limited. Best AMD result on Vogons was with K6-3+ 6x100 at 3766 using Voodoo3 2000.

@PCRetroTech

3 жыл бұрын

Wow, I had no idea they were already so many layers. Interesting indeed.

@rasz

3 жыл бұрын

@@PCRetroTech Every time you see PCI bus there are at least 4 layers to match impedance specification (65ohm). Then you have huge bga chips requiring more signal layers to fan out. Today good motherboards are 10 layers, graphic cards up to 12.

@Dxceor2486

3 жыл бұрын

I've never been able to run a celeron 300A at 450MHz and get it stable. I have one that seems stable but isn't and the other that doesn't even post iirc. I don't know if I'm just unlucky or if there's more to it. My celerons are for OEMs so that may be why ?

@rasz

3 жыл бұрын

@@Dxceor2486 I on the other hand never had 300A that wasnt stable at 450 :) Very few were stable at stock 2V, hence tape on 4 pins, one on B21 for 100MHz plus three others (B121? A119? dont remember, its been 23 years :P) to bump up Vcore to 2.2V. Celerons 333 were difficult, I had no luck with them above 83MHz, I guess the first batch was topping out at 450MHz. Then Intel must of done something in the foundry because new models released ~half year later bumped up the upper barrier and 366 FPGA Celerons again started to OC with no problem to 566MHz, often without changing voltage, while all later FPGA Mendocino models were a bust topping somewhere around 550MHz so best you could hope for would be 433 at 83MHz fsb. Pretty much all Coppermines overclocked to 100MHz FSB bus all the way to Celeron 766 model, was just a matter of bumping voltage (up to 2V), but at that point AMD released Durons with the cheapest ones once again overclocking all the way to 950-1000MHz with only 4 pencil strokes. I started out working for bulk importer of PC parts (official European Asus, CTX, Fujitsu, Primax and bleh PCchips distributor) in 1998, but quickly went independent and just build/sold computers for a living for a couple of years. I rarely sold non overclocked systems. It was so easy and lucrative back then, making $150 part perform like $500-600 one. Nothing like it left on the market right now, its all gamer focused blue led backlit premium sucker niche.

@Dxceor2486

3 жыл бұрын

@@rasz I tried to up the voltage too ! Unfortunately even at 2.6V it didn't work so I just had really bad samples of celerons I guess

Please don't use Windows 98 Gold. It's unstable and some drivers don't work properly. Always use Windows 98 Second Edition when you want to use Win98.

@pazsion

Жыл бұрын

Windows nt… should still be actively supported and faster and better

But will it run DOOM? I mean, Doom 3? I believe K6-2 and GeForce 256 is absolute minimum for that, not even sure it will run on this CPU.

Did you have any trouble getting 2.0v for the K6-2 CPU on the Gigabyte board? I have a very similar Gigabyte board with the same chipset and the manual lists a jumper setting for 2.0v but there's a defect in the way it's wired and it actually produces 0v on that setting. You could run 2.1v instead, but my K6-2+ 500 doesn't love that so I soldered a jumper wire across two pins of the regulator to allow access to the "lower half" of the voltage table which includes a working setting for 2.0v.

@PCRetroTech

3 жыл бұрын

No, I didn't have any problems with that as far as I recall. I wonder how many boards they made like that! Is yours also the GA 5AA? I have a late revision of that board from memory. So maybe they fixed that. You may have a very early revision. The board was lauded for its K6 II support when it came out, and also later when as yet unreleased K6's just worked on the board.

@angieandretti

3 жыл бұрын

@@PCRetroTech I went back n looked at my documentation. My board is actually Asus P5A, but similar appearance and same chipset.

@PCRetroTech

3 жыл бұрын

@@angieandretti Cool, thanks for looking it up.

You actually tried an amd in a intel board 😂

@pazsion

Жыл бұрын

Says it’s supported but I’d never trust it… intel did shady things

@pazsion

Жыл бұрын

No idea which test was amd or intel

Ali is not as good as vía. Some Ali exactly M1541 doesnt work with 100 MHz fsb on AMD and a Nvidia TNT 2

@PCRetroTech

3 жыл бұрын

Interesting. Mine did work, luckily and the VIA didn't. But it probably depends on the precise chipset and motherboard manufacturer.

@asanjuas

3 жыл бұрын

@@PCRetroTech mine doesnt work unless fsb las 66 with 100 MHz only works with 3DFx voodoos

@PCRetroTech

3 жыл бұрын

@@asanjuas Interesting. What revision board do you have?

@asanjuas

3 жыл бұрын

@@PCRetroTech It has a freetech a long time ago

@asanjuas

3 жыл бұрын

And this was my experience with this chipset