VOGONS


First post, by Fallaxia

User metadata
Rank Newbie
Rank
Newbie

Hi,

I got a strage problem with my Aopen AK72 Slot A board.
An AMD Slot A 1000 Mhz Orion core CPU (the one with dedicated 512KB Cache modules in it´s cartridge) works just fine, no problems, while a Slot-A 1000 Mhz Thunderbird core (with 256 Kb cache on die) crashes (freeze), even in memtest86.

I don´t know why. Both CPUs are fine. I tested several Orion/Thunderbirds 1000Mhz CPUs, and only the Thunderbirds show these instability issues.
The power supply should be good too, it has 30A on 3.3V and 30A on 5V, it´s even a redundant one with several self-test features. Tested several others to be sure.

For info: There is no overclocking in place. The RAM is 3 sticks of 512MB each, 1.5GB in total of PC133 ECC SDRAM CL3, set in Bios with ECC active and CL3 accordingly.
GPU is a ELSA Virctory Erazor AGP
HDD (SSD) is a 60GB Transcent IDE Industrial SSD
Network card is an Intel Pro 1000S Dual (PCI-X card in PCI Slot)

What am I missing here? Does the thunderbird have any special needs to work properly. I had it running once, several years ago, so I know it works just fine with the AOpen AK72 board.
Any ideas?

Reply 2 of 15, by PC-Engineer

User metadata
Rank Member
Rank
Member

The Thunderbird SlotA doesn't work with VIA KX133.

With change from Orion to Thunderbird AMD changed something in the bus interface (from "open-drain" to "push-pull"). Chipsets which provide push-pull: VIA KT133, VIA KM133, AMD760.
There were some late hardware-modified (a workarround to emulate with open-drain a point to point behavior from push-pull - don't know anything about stability) boards with VIA KX133 and AMD750 chipsets produced, which can support Thunderbird, but very bad documented.

Epox 7KXA Slot A / Athlon 950MHz / Voodoo 5 5500 / PowerVR / 512 MB / AWE32 / SCSI - Windows 98SE

Reply 3 of 15, by Fallaxia

User metadata
Rank Newbie
Rank
Newbie

Yes, I am sure I ran the Thunderbird with the AK72 board.
I had it running with Debian Linux to do some testing, and the only thing that did crash the system was an openssl benchmark at the Camellia cipher. Everything else worked just fine.

Now I barely get to boot the system to this point where I can do anything. Sometimes it crashed after a few minutes, sometimes during boot, totally random.

@PC-Engineer: Would the change in the bus interface design affect stability or cause the system to not boot at all? I am asking becaus rthe Thunderbird works, just not stable.

Reply 4 of 15, by dionb

User metadata
Rank l33t
Rank
l33t
PC-Engineer wrote on 2020-01-05, 10:11:

The Thunderbird SlotA doesn't work with VIA KX133.

With change from Orion to Thunderbird AMD changed something in the bus interface (from "open-drain" to "push-pull"). Chipsets which provide push-pull: VIA KT133, VIA KM133, AMD760.
There were some late hardware-modified (a workarround to emulate with open-drain a point to point behavior from push-pull - don't know anything about stability) boards with VIA KX133 and AMD750 chipsets produced, which can support Thunderbird, but very bad documented.

Iirc AMD750 didn't officially support push-pull but it did work. That was the late Slot A dilemma: for best CPU performance (Thunderbird was miles ahead of Orion), you had to choose the older AMD750 chipset, with a feature set otherwise a lot more like a 1998 chipset (it's almost a copy-paste of i440BX), and slower clock-for-clock to boot. Alternative was KX133, which offered faster I/O, faster RAM (which mattered - unlike P3's single-pumped P6 bus, the double-pumped EV6 bus was starved with PC100's 800MB/s), but only stable with the older, slower, hotter Orion. That dilemma neatly explained why Athlon uptake didn't shoot off until SoA and the KT133 arrived.

Reply 5 of 15, by The Serpent Rider

User metadata
Rank l33t
Rank
l33t

If it's not stable, maybe there are some capacitors to blame. Also it's possible that CPU voltage drops too low under load, which cause instability. Try to increase voltage.

Thunderbird was miles ahead of Orion

Thunderbird is not miles away. L2 cache performance is only marginally better. Also it was never available in retail, only OEM. Which kinda shows how small the improvements were.

Last edited by The Serpent Rider on 2020-01-05, 15:26. Edited 1 time in total.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 6 of 15, by Fallaxia

User metadata
Rank Newbie
Rank
Newbie

@dionb: Thank you for the detailed info, anyway, dang, that´s rather sad to read. Got a nice Tbird 1000 and no board to really support it stable and well.

What board should I go for the Tbird 1000? Is there any good one at all?

Reply 8 of 15, by dionb

User metadata
Rank l33t
Rank
l33t
The Serpent Rider wrote on 2020-01-05, 15:15:

If it's not stable, maybe there are some capacitors to blame. Also it's possible that CPU voltage drops too low under load, which cause instability. Try to increase voltage.

Thunderbird was miles ahead of Orion

Thunderbird is not miles away. L2 cache performance is only marginally better. Also it was never available in retail, only OEM. Which kinda shows how small the improvements were.

The difference was bigger than that, about 11% in actual game performance:

win98benches_image010.gif
(yes, this is SoA T-Bird on KT133 vs SlA Orion on KX133, but those two chipsets were - apart from the push-pull support - pretty much identical. The only way you could compare like-for-like on SlA would have been with AMD750, and nobody in the day considered reviewing that)

Reply 9 of 15, by The Serpent Rider

User metadata
Rank l33t
Rank
l33t

about 11% in actual game performance:

Which is not "miles ahead". And that is in the most CPU intensive game of their time.

KT133 vs SlA Orion on KX133, but those two chipsets were - apart from the push-pull support - pretty much identical.

Yes, they are more or less identical, but the boards aren't. Some have bank interleaving enabled, some don't. Some are slightly faster or slower due to other reasons (BIOS tweaks, engineering, etc). With that in mind, I think the difference on the same board would be only 5-8% or less.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 10 of 15, by dionb

User metadata
Rank l33t
Rank
l33t
The Serpent Rider wrote on 2020-01-06, 10:20:

about 11% in actual game performance:

Which is not "miles away". And that is in the most CPU intensive game of their time.

11% is a pretty major difference between otherwise very similar CPUs, indeed it's bigger than the difference between Thunderbird and Coppermine, and the same order of difference as between the top of the line Core2Duo vs the top of the line Athlon64 X2 at time of the C2D launch, or the jump from Core2Duo to Nehalem Core i7. 11% is impressive in this business.

And of course you only see it when stressing the CPU, that's how you measure CPU performance. Unreal Tournament is also perfectly valid application for this CPU. Q3A shows a smaller (3.5%) difference, but Q3A was notoriously optimized for Coppermine, so no surprise it doesn't highlight the difference between the Athlons as well. General purpose benchmarks show results in between the two (9% in Sysmark2000).

KT133 vs SlA Orion on KX133, but those two chipsets were - apart from the push-pull support - pretty much identical.

Yes, they are more or less identical, but the boards aren't. Some have bank interleaving enabled, some don't. Some are slightly faster or slower due to other reasons (BIOS tweaks, engineering, etc). With that in mind, I think the difference on the same board would be only 5-8% or less.

Anand used a single single-sided PC133 DIMM for these tests, so 4x bank interleaving wouldn't affect results. Moreover the KX133 board was an Asus K7V, which supports interleaving and has it enabled by default. It was also the fastest Slot A board, bar none. You're not going to get better Slot A + Orion scores than this. The KT133 was a Via engineering board. Those things are rarely tweaked for performance, but even if it had been, it would at best have equalled the K7V.

Reply 11 of 15, by The Serpent Rider

User metadata
Rank l33t
Rank
l33t

Moreover the KX133 board was an Asus K7V

Here's other test with Asus K7V: https://www.ixbt.com/cpu/thunderbird-800.html

The difference is 3-5% (Sysmark2000 included).

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 12 of 15, by NostalgicAslinger

User metadata
Rank Member
Rank
Member

The only Slot A VIA KX133 which works fine with Thunderbirds until 1000 MHz, was the Asus K7V-T. I also had lots of problems with the 1GHz version on the Epox 7KXA Rev. 0.4 with freezes and lockups, thought the CPU has a defect... (the 750 Mhz T-Bird version works fine).

After searching for more infos here in the forum and other pages, I have found out that the KX133 itself is the problem and the K7V-T has a fix for it. The normal K7V and most of the other VIA KX133 boards should work fine with Thunderbirds until 800 MHz, but this is unofficially. Also recommended to use the latest mainboard revisions. For example the Rev. 0.3 of the Epox 7KXA is very unstable with all Thunderbirds. After all:

Reply 13 of 15, by The Serpent Rider

User metadata
Rank l33t
Rank
l33t

In any case, the difference between Orion and Thunderbird CPUs just not worth all that hassle.

Last edited by The Serpent Rider on 2020-01-06, 14:43. Edited 1 time in total.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 15 of 15, by Delerium

User metadata
Rank Newbie
Rank
Newbie

In some cases the old Orion core is faster than the Thunderbird core at the same frequency. May be due to the bigger L2 cache?

https://www.anandtech.com/show/557/23

Attachments