Hi, can you give links about how wirebond technology is more durable? Would love to read and understand more.

(I'm huge fan of am1 platform, apu is true SOC and board is more or less passive component without bga)

shfil wrote on 2025-01-22, 17:16:

Hi, can you give links about how wirebond technology is more durable? Would love to read and understand more.

(I'm huge fan of am1 platform, apu is true SOC and board is more or less passive component without bga)

I have no links about wirebond durability, but you can search articles about flip-chip disadvantages, for example, most popular NForce4 chipsets failures e.t.c.
Wirebond technology is more durable with molding, because welded wires are reliably sealed, without molding wires can break because of drop or vibrations. Cyclic heating (power off - power on -power off) cannot destroy contact of wirebond molded chip, because welding more reliable than flip-chip's lead-free soldering
Of course, today flip-chip technology can be more reliable than earlier, but it's always a lottery

I specially was looking for failure AM1 CPUs and found and found a couple with a description "it suddenly dead after next power-on", i tested them and beep codes was like there is no video adapter
AM4 CPUs also SoCs and also you can find dead CPUs, but i never seen dead 775,754,939 and other old CPUs

I would also like to say about dual-channel reliable VIA chipset PT880 (not PT880 PRO/ULTRA - it's a non-reliable flip-chip chipsets)

I have motherboard 775V88+ and it has strange CPU clock somewhere 194 MHz and it works correctly with 2 dual-channel slots DDR1 at full 400 MHz frequency, but if you use 4 slots dual-channel or maybe single-channel 1,2 or 3,4 slots, chipset reduces DDR1 clock somewhere 133 MHz (266 MHz)
With 478 board and this chipset the same is true, but I don't know how it works on SiS's dual-channel
So, this is true, for VIA 775 AGP chipsets 2 GB RAM will be better

While technically true that wirebond technology itself should be more reliable, I don't think it's the sole decisive factor for reliability. Temperatures also still matter.

Think about all of the dead/artifacting GeForce 4 TI video cards, for example - they all use wirebond technology, but still fail quite easily - especially the ones that use a stock cooler.
Likewise, the Southbridge on my AsRock 939Dual-Sata2 is also dying a slow death. I lost the audio first, then the LAN started being flaky, then the SATA ports started getting flaky... and now the PCI slots are also flaky and cause data corruption.
Then there's the VIA NB on my ASUS K8V-SE Deluxe, which also appears to have some issues now (I get graphical artifacts on just about every AGP video card on a cold boot... and all of my cards work fine on other systems.)

So I can't say that stuff based on wirebond technology will be more reliable. With proper cooling... perhaps. But then, even BGA stuff could be relatively safe if kept nice and cool.

In regards to Intel chipsets, I'm a big fan of the i865 and i965 (Q965 and P965) - these are pretty tough chipsets, considering how often they are left with poor/inadequate cooling, thus always running scorching hot... yet rarely have any issues. So I think the quality and complexity of the chip also play a role.

I like this discussion. As we are going further into reliability, imho we should clearly state that reliability is made from longevity and repairability of product. Different people will have different opinion what's more important.
(Often you need to make sacrifice. For example features vs reliability or longevity vs repairability)

For example normal customer cares only about longevity as he plans to use component max for 3 years. Here we would like to use component forever by always being able to fix it.

So longevity:
- lack of factory defects (check on used market, which you models you can see faulty)
- high quality components, overengineered design (for example more phases on VRM, or PGA instead of LGA)
- low stress: Chips with smaller die size tend to be more resistant than those with large. Choosing chips ported to smaller litography or chips without extra unneeded functionality can have better longevity.
- good working condition (for example oversized cooler)

Repairability:
- "overenginered design", for example putting extra resistors on traces will protect the worst thing to damage - PCB. Another thing is the smaller board or more components on board means worse condition to work with in case of failure.
- access to replacement parts, imho super important
- how easy it is repair (time is money, also for most people bga means unrepairable)
- less problematic components on board like north bridge the better.
- standard components (oems are terrible at this)
- access to schematics/boardview

Do I miss something?

Recommend to watch:
https://www.youtube.com/watch?v=3qKtS_uxdcU (newest info about xbox 360/TSMC problems)
https://www.youtube.com/watch?v=uMHUz16MuYA (Rambling about chip degradation)

shfil wrote on 2025-01-28, 22:26:

For example normal customer cares only about longevity as he plans to use component max for 3 years.

That's a good point.

This is actually referred to as the product life cycle by manufacturers.
Every device is designed with a specific life cycle. Typically, it's around the 5-year mark for most consumer stuff. That is, the manufacturer only tries to design with enough quality/reliability in mind where the product will last about 5 years *on average*. Anything less than that, and there's a chance the customer might still be in ownership and active use of the item and wishing for it to continue to work. If it stops working, that could leave the customer with a bad impression about the product/company. So (serious) manufacturers do have some interest in trying to create products that don't have too short of a life cycle. At the same time, most manufacturers know that *many* customers probably won't use an electronic product for much longer than 5 years on average... so they have no incentive to try to make something last forever... or as long as possible - that's just a waste of resources / unnecessary costs / money spending.

As such, most electronics made in the last 10-15 years have been designed with this 5-year minimum average life cycle in mind... which means, most will last at least 5 years with "regular" everyday use. Of course, there will always be outliers - some that don't last even half of that time or even a fraction of it (ones with factory defects, for example). And some will probably easily last 2x the minimum average life cycle. In the end, as with anything in manufacturing, there's always tolerance to be observed, which is why two seemingly identical products will never be exactly the same.

Anyways, what I want to say with all of this is that, for example, when you see manufacturers telling you that it's normal for CPU X or GPU Y to run at 70 or 80C... that's because they mean that with this 5-year average life cycle in mind. The truth is, if you want your hardware to last, you should always keep it much cooler than what the manufacturer has said is "normal" for the hardware.
And this is also why I always recommend to people to always keep their silicone cooled below 60C whenever possible.
In the case of nVidia bumpgate fiasco (and similar "sensitive" hardware from other manufacturers too), I'd say that temperature goal should be even lower: 50 - 55C max... which depending on the hardware, could become quite challenging to achieve. But if you want good reliability, sometimes you really have to go the extra mile to design/modify the design of the manufacturer to something better. In regards to old video cards, this means getting rid of their small crappy cooler with small crappy fans and actually using something appropriate for the TDP of the device (e.g. the stock Radeon 9700/9800 single-slot cooler - put something big on top of that from day one of owning the card, and it may have a much better chance of it lasting longer.)

momaka wrote on 2025-01-28, 20:56:

While technically true that wirebond technology itself should be more reliable, I don't think it's the sole decisive factor for […]
Show full quote

While technically true that wirebond technology itself should be more reliable, I don't think it's the sole decisive factor for reliability. Temperatures also still matter.

Think about all of the dead/artifacting GeForce 4 TI video cards, for example - they all use wirebond technology, but still fail quite easily - especially the ones that use a stock cooler.
Likewise, the Southbridge on my AsRock 939Dual-Sata2 is also dying a slow death. I lost the audio first, then the LAN started being flaky, then the SATA ports started getting flaky... and now the PCI slots are also flaky and cause data corruption.
Then there's the VIA NB on my ASUS K8V-SE Deluxe, which also appears to have some issues now (I get graphical artifacts on just about every AGP video card on a cold boot... and all of my cards work fine on other systems.)

It's very interesting
I know about GeForce 4 TI failures and also have Geforce 3 Ti 500 with graphical artifacts, but I don't know after what it happened. Maybe it is not a graphical chip, maybe it is RAM, maybe PCB board failure (some contacts in some inner layer was broken by heating), maybe capacitors, I don't research this problem, but maybe I will in future 😀
But about VIA NB - I bought a lot of motherboards (both AMD and Intel) on VIA, nForce3 and ATI chipsets based on wirebonding and all of them (!!!) worked 😀
I buy such motherboards without checking, knowing that they are definitely working. In your case VIA NB it should be researched better, I don't think that it is chipset's issue, maybe it is bad capacitors ?

I found out interesting thing: some motherboards on VIA P4M800 (not PRO) like P5V800-MX or Biostar P4M800-M7 if you set DDR1 frequency 400 MHz in BIOS, it sets 6:5 DRAM:CPU ratio that means not 400 (200 MHz), but 480 MHz (240 MHz) DDR1 clock. This feature overclocks these boards on P4M800PRO chipset's level.
I think that it was so because these boards was aimed to Celeron D and PC2700 DDR1 memory, but for example ASROCK 775VM800 have correct DDR1 400 MHz 1:1 DRAM:CPU ratio and with 800 MHz FSB it works slower, than on ASUS and BIOSTAR

Vendein_RaZoR wrote on 2025-01-28, 14:35:

I have motherboard 775V88+ and it has strange CPU clock somewhere 194 MHz and it works correctly with 2 dual-channel slots DDR1 at full 400 MHz frequency, but if you use 4 slots dual-channel or maybe single-channel 1,2 or 3,4 slots, chipset reduces DDR1 clock somewhere 133 MHz (266 MHz)

This maybe a consequence of PT800 signle-channel chipset legacy, because some motherboards with PT800 have 3 slots, but PC3200 was maximum on 2 slots

Vendein_RaZoR wrote on 2025-01-29, 07:47:

I found out interesting thing: some motherboards on VIA P4M800 (not PRO) like P5V800-MX or Biostar P4M800-M7 if you set DDR1 frequency 400 MHz in BIOS, it sets 6:5 DRAM:CPU ratio that means not 400 (200 MHz), but 480 MHz (240 MHz) DDR1 clock. This feature overclocks these boards on P4M800PRO chipset's level.
I think that it was so because these boards was aimed to Celeron D and PC2700 DDR1 memory, but for example ASROCK 775VM800 have correct DDR1 400 MHz 1:1 DRAM:CPU ratio and with 800 MHz FSB it works slower, than on ASUS and BIOSTAR

For P4M800PRO this is true too. With DDR2 533 MHz in BIOS ratio DRAM:CPU sets as 8:5 and with 533 MHz as a results we get 650 MHz (325 MHz) and with 210 MHz a little overclocking FSB we get 670 MHz (335 MHz) DDR2
But officially P4M800PRO supports only 533 MHz and in BIOS POST screen we can see 533, but in OS in CPU-Z or Everest we get overclocking 😀

Vendein_RaZoR wrote on 2025-01-29, 09:49:

For P4M800PRO this is true too. With DDR2 533 MHz in BIOS ratio DRAM:CPU sets as 8:5 and with 533 MHz as a results we get 650 MHz (325 MHz) and with 210 MHz a little overclocking FSB we get 670 MHz (335 MHz) DDR2
But officially P4M800PRO supports only 533 MHz and in BIOS POST screen we can see 533, but in OS in CPU-Z or Everest we get overclocking 😀

I found out, this is true only for 200 MHz CPU bus and true for P4M800 and P4M800PRO.
1:1 DRAM:CPU ratio is being installed only when used 266 MHz bus or when DDR frequency not set in BIOS in case of Pentium 4/D

Vendein_RaZoR wrote on 2025-01-21, 15:50:

----------------------------------------------AGP BOARDS--------------------------------------------- […]
Show full quote

----------------------------------------------AGP BOARDS---------------------------------------------

About AGP boards you can read here LGA 775 Motherboards with AGP Slots

(1?) P5VDC-MX V2.0 ----------------------------- VIA P4M800PRO + VT8251, it works with Core2 Quad, but unbalanced load on cores (maybe i find out it better)
(1?) ECS P4M800PRO-M2 ----------------------------- VIA P4M800PRO + VT8237R+, it works with Core2 Quad, but unbalanced load on cores (maybe i find out it better)

------------------------------------------------------------------------------------------------------

I also found out, P5VDC-MX V2.0 and ECS P4M800PRO-M2 works with Core2 Quad, but I get microcode error in BIOS (but not on ECS P4M800PRO-M2) and sometimes the system suddenly reboots on both boards
Someone thinks that it is because of insufficient power consumption, but I think it is for bad microcode, because this board uses RT8802A that used on P4M890 and P4M900 - all works fine, VID lines are the same
I also try QX6700 and QX6800 CPUs on P5VDC-MX V2.0 and with multiplier below default value (i try also decrease frequency) there is many glitches in Windows especially in graphics applications

Sorry, I recently found out that 775 VIA chipsets are unstable with a 1066 bus. This mainly concerns chipsets released before 2007 +/-.
I found out this on P5VDC-MX V2.0 motherboard with P4M800PRO chipset, that was released at the end of 2004 and it was modified version of P4M800 chipset that released in 2003.
If you set for example Core 2 Quad Q6600 (or maybe Core 2 Duo E6600) on P5VDC-MX V2.0, that supports Core 2 CPUs, an unpredictable reboot may occur at any moment. But possibility of this reload decreases with lower bus frequency and later chipset's release date (somewhere around 2007). This problem maybe comes from 2003, when Gallatin 1066 bus was supported by Intel for exapmle 865 chipset, but VIA not aimed on that frequency and this problem was fixed in later chipsets but not completely (mainly fixed by newer production technology)
P4M890 and P4M900 chipsets have the same problem, but I have never seen motherboard on these chipsets with changeable bus frequency to the downside. Reload problem with 1066 can may occur after weeks or months of use, but more often appears under load. This makes VIA chipsets on 1066 MHz unreliable.
So latest absolutely stable bus frequency on VIA is 800 (but I saw posts with ECS P4M900T-M2 mobo that suddenly stopped working with Core 2 Quad CPU's and only started with 533 bus, but this is not beacause of caps, power, memory, BIOS or something like that, thats very interesting)
On motherboards with SiS chipset like 671T-M i had reload only with sufficient overclock, but this chipset have 1066 with overclock (in comparison SIS671FX that have 1066 by default), I haven't seen yet SIS 671/672 chipsets unstable.
I also haven't seen yet AMD VIA's chipsets unstable, but will see how they will work.