Swiss Retailer Stats Reveal Which GPU Brand Has the Highest Failure Rates

[LAwLz]

5 hours ago, Stahlmann said:

The graph shows duration after the part arrived in the service center, so they probably mean the one from the manufacturer/brand, not their own store warehouse.

Maybe I am dumb, but I don't understand what you mean.

Can you elaborate?

[Stahlmann]

37 minutes ago, LAwLz said:

Maybe I am dumb, but I don't understand what you mean.

Can you elaborate?

Your guess is that  they have a lot of one brand in stock. That means to me you're talking about digitec having them in stock for fast replacements.

 

The graph isn't talking about how fast the retailer can replace these cards, but how fast they're handled and sent back after arriving in the brands service/repair center.

[LAwLz]

9 hours ago, Stahlmann said:

Your guess is that  they have a lot of one brand in stock. That means to me you're talking about digitec having them in stock for fast replacements.

 

The graph isn't talking about how fast the retailer can replace these cards, but how fast they're handled and sent back after arriving in the brands service/repair center.

Yes, you nailed the part about my assumption but that I acknowledge doesn't make much sense.

 

I don't really get the second part though. Can you please explain in as much detail what the 2 days vs 17 days mean? I really feel dumb and lost right now so please try and use as clear and detailed an explanation as possible.

 

Are you saying that the time intervals shown on the graph is the time between the retailer sending the card to the manufacturer and then getting a replacement back? Because the graph specifically states that it is measuring the time it takes before the customer has their card back. Since it's the retailers that handle the returns in Czechia, I assume the flow is like this:

 

1) Card breaks.

2) Customer sends the card back to the retailer.

3) The retailer sends a new card out to the customer.

4) The customer gets a new card.

5) The retailer sends the broken card to the manufacturer.

6) The manufacturer inspects the card and repair/take a new card.

7) The manufacturer sends the card back to the retailer.

8) The retailer gets the new card and can sell that.

 

 

I see a few possibilities.

1) The chart measures the time between points 2 and 4. If this is the case, how fast the customer gets their card back should only be affected by which cards the store has in stock, not how the manufacturer behaves.

2) If point 3 never happens and instead the customer has to wait for the new card from the manufacturer, then I don't see how it is possible that AsRock does it in 2 days. The shipping time alone, since the card has to leave the country, is most likely more than 2 days.

3) If it's measuring between points 5 and 8 then I don't see how this is relevant to customers, and it doesn't make sense since the chart specifically says it's measuring how long it takes before the customer gets the card back.

[leadeater]

On 8/29/2023 at 9:32 AM, micha_vulpes said:

• The Only "lenovo" cards on Digitec are Passive Server compute cards, and 8-10 year old Quadros.
• The only HP cards on Digitec are NvidiaGrid cards of the Kepler Generation!

What were your filters? I get a lot of all cards and new ones, I'm sure you've done something with your filters to make it not show everything.

 

i.e. (sorted by high price)

 

Spoiler

 

Spoiler

 

HPE & HP

 

Spoiler

 

 

And FYI in regards to HPE, Lenovo etc when their products are under warranty, business and server, you deal directly with their support not the retailer you purchased it from so most likely why there is no warranty data for that aspect because they simply don't have that information.

[Stahlmann]

5 hours ago, LAwLz said:

Yes, you nailed the part about my assumption but that I acknowledge doesn't make much sense.

 

I don't really get the second part though. Can you please explain in as much detail what the 2 days vs 17 days mean? I really feel dumb and lost right now so please try and use as clear and detailed an explanation as possible.

 

Are you saying that the time intervals shown on the graph is the time between the retailer sending the card to the manufacturer and then getting a replacement back? Because the graph specifically states that it is measuring the time it takes before the customer has their card back. Since it's the retailers that handle the returns in Czechia, I assume the flow is like this:

 

1) Card breaks.

2) Customer sends the card back to the retailer.

3) The retailer sends a new card out to the customer.

4) The customer gets a new card.

5) The retailer sends the broken card to the manufacturer.

6) The manufacturer inspects the card and repair/take a new card.

7) The manufacturer sends the card back to the retailer.

8) The retailer gets the new card and can sell that.

 

 

I see a few possibilities.

1) The chart measures the time between points 2 and 4. If this is the case, how fast the customer gets their card back should only be affected by which cards the store has in stock, not how the manufacturer behaves.

2) If point 3 never happens and instead the customer has to wait for the new card from the manufacturer, then I don't see how it is possible that AsRock does it in 2 days. The shipping time alone, since the card has to leave the country, is most likely more than 2 days.

3) If it's measuring between points 5 and 8 then I don't see how this is relevant to customers, and it doesn't make sense since the chart specifically says it's measuring how long it takes before the customer gets the card back.

From seeing the chart description i'd assume the customer has to send the card back directly to the manufacturer's service center and also get the replacement directly from them. Brands like Asrock, Gigabyte, Asus etc. often have service centers in different countries so customers don't have to ship stuff to China, Taiwan etc.

 

So my assumption is that this chart shows how long it takes between the broken card arriving at said service center and a replacement arriving at the customer. It doesn't include the conversations between the customer and the retailer.

 

In that case it's possible for Asrock to be so fast because maybe they instantly ship a replacement, even before fully checking or repairing the card. And Gigabyte is so slow because the card goes through the whole repair process before they decide what to do. (maybe even going to a repair center in Asia)

 

I can also back up this assumption because that was exactly how Asus handled my broken monitor a year ago. I had to ship it directly to a service center. The service contractor (not Asus directly) instantly shipped a replacement unit, even before my broken one arrived or was checked. I never found out what happened to the old monitor after it arrived at the service contractor. It didn't really matter to me because i already had the new unit and the ticket was closed.

[leadeater]

6 hours ago, micha_vulpes said:

It appears to have been my absolute LACK of filtering that resulted in these... results. After clicking through the first few pages, and not seeing a single consumer card, I just incorrectly assumed their sales figures.

I have to say the website is horrible and I hate it haha 

[strajk-]

On 8/30/2023 at 4:23 PM, LAwLz said:

Yeah, but in that case shouldn't all manufacturers get the same result?

I don't see how one manufacturer can get an average return period of 2 days while another gets 17 days if in both cases it's the retailer that exchanges the product.

 

I guess one explanation is that if they have a lot of one brand in stock then the risk of having to wait for a new shipment is lower, but that isn't a good indicator of how brands handle returns or how reliable the manufacturer itself is.

I think the best way for you to get answers is to read what they have to say about their data:

https://www.digitec.ch/de/wiki/6167

 

Often you get a replacement instantly, sometimes depending on the supplier you might wait a while, I had one instance where I had to wait for 2 weeks to replace a UPS since those were handled by an affiliate in Germany that represented their main branch in Spain (Salicru), so it took a while to get processed.

 

Parsed it through ChatGPT (deepl asking for a sub lately for longer text):
 

Spoiler

How is the warranty claim rate determined and what does it indicate?

We count the warranty claims for a brand within a specific category (e.g., "TV") that need to be handled through our after-sales service.

f we sell 1000 TVs of a brand and 150 of them are sent in for warranty within two years, then we display a warranty claim rate of 15% for TVs of that brand. If we also have data for other brands (e.g., "LG," "Samsung," "Panasonic," ...), we enable cross-brand comparisons.

 

This allows you to see how likely it is for this product to have a warranty-related issue during the statutory warranty period of 2 years, and you can also compare different manufacturers of products within a category at a glance. Manufacturers that may prioritize sustainable construction are easier to identify. This helps you better assess the likelihood of trouble and resource waste due to a potential warranty issue.

 

Why don't I see a warranty claim rate for all products?

We only consider data from the last 24 months and display the rate only when there are 300 comparable sold products.

Additionally, for entirely new products, we wait one month after the sales launch before displaying data. Prior to that, the significance wouldn't be sufficient, and individual cases would distort the numbers too much.

 

What does the warranty claim duration indicate?

The "warranty claim duration" shows how many working days it takes to process a service case for a product from a specific brand, from its arrival at the service center until you receive it again.

Most purchased products are entitled to the statutory 2-year warranty from the seller. Therefore, Digitec Galaxus takes care of communication and the flow of goods with the manufacturers on your behalf within this period after you've reported a warranty case to us.

 

We organize and manage various repair and return processes with the respective manufacturers and service centers. This leads to different dependencies that influence the entire process of handling warranty claims.

 

What is the usefulness of the return rate and why do we show it?

The "return rate" indicates how often a product from this brand in the corresponding category (e.g., "Smartphone") has been returned within the last year. If we sell 100 smartphones of a brand and 15 of them are returned within one year, then we display a return rate of 15% for smartphones of that brand.

 

This includes products that didn't meet expectations upon receipt. Perhaps the manufacturer's description was too optimistic, the size was incorrect, the product images weren't accurate enough, or the quality was insufficient - to name a few reasons. The return rate aggregates products from a brand in a category where we have sufficient data. The brand represents a collection of products with the same promise of quality and reliability.

 

On the other hand, we strive to resell returned products in line with the principles of a circular economy. However, not everything is in a reusable condition and needs to be disposed of and recycled. With the information from the "return rate," you might be able to infer the quality of the manufacturer's product description and inquire with our community or the manufacturer if you have uncertainties. This can help mitigate effort on your part and unnecessary emissions in advance.

 

Why don't I see a return rate for all products?

We only consider data from the last 12 months and display the rate only when there are 150 comparable sold products. Before that, the significance wouldn't be strong enough, and individual cases would distort the numbers too much.

EDIT: Getting too late for me, I just realized that I replied to a very old post that probably already got addressed lol

[LAwLz]

9 hours ago, Stahlmann said:

From seeing the chart description i'd assume the customer has to send the card back directly to the manufacturer's service center and also get the replacement directly from them.

That's what I was thinking as well, but the evidence pointing against that are:

1) The extremely short times for AsRock and some other. Again, 2 days would barely be enough for shipping.

2) Someone earlier said in the thread that all RMAs go through the retailer, not directly with the manufacturer (that's why they are even able to gather these statistics to begin with).

3) I am pretty sure I saw on their website that they tell you to contact them, not the manufacturer if something goes wrong.

 

 

I just find the whole thing suspicious and would like to know more about what exactly it means. 

All it says on the website is:

Quote

The warranty case duration shows how many working days it takes to process a warranty claim for a product of a certain brand from the time it arrives at the service centre until it’s back with the client.

 

 

Could it be that the retailer handles all the contact and act as a middleman, but the return label goes directly to the manufacturer? That could explain some things. If the "countdown" also only starts once the product has reached the service center then AsRock's very short return times would make sense.

 

But in that case I would argue that it's a pretty useless metric. Because if the package is really slow to arrive at the service center, that provides the manufacturer with more time before the "clock starts" so to speak.

What I as a customer care about is how long time I am without a product. Here is an extreme example:

 

Let's say AsRock required you to send products back with a boat all the way to Taiwan. The shipping time might be 30 days.

Your card breaks, and you send it back. It will be 30 days before it reaches the service center and the clock starts ticking.

However, a few days before the card arrives, they send a replacement via air shipping. As a result, your replacement card arrived to you 31 days after you initiated the RMA, but it would show up as "1 day" on their stats page. Meanwhile, a company that got the card in 2 days, diagnosed and repaired the card and set it back to you in 5 days would result in a "score" of 5 days, even though you were without a card for merely 7 days.

 

The stat would show the manufacturer that left you without a card for 31 days as far better than the company that left you without a card for 7 days if that's how it counts.

[SpaceGhostC2C]

On 8/28/2023 at 4:33 PM, TVwazhere said:

Sadly this information doesnt have enough samples to have anything that people should even start looking into regarding longevity of a product.

 

On the contrary, their minimum sample is size is about 10 times what you need to start getting reasonable approximations to a simple mean.

 

On 8/28/2023 at 5:32 PM, micha_vulpes said:

The whole way this is presented bothers me for a few reasons

1. No mention of the number of cards corresponding to each percentage. This makes the percentage comparison effectively meaningless because we just don't know the sub-sample size and smaller samples will be more volatile and subject to wild swings from only a few reports.

They do tell you that all samples are bigger than 300. You don't really need the exact number, unless they also give you the sample variance so you can use both to construct a confidence interval. Instead, what they really should do is calculate the CIs themselves and report that together with the rates, then you wouldn't need anything else. It is entirely possible that you can't statistically reject all rates being identical. A lot of focus on the ranking, but the numbers are similar, and quite low. Providing such a standard measure of uncertainty around the point estimates would be enough.

 

Quote

1. It's not Aggregated in any way that makes sense to me. "HP, PNY, Lenovo" are all made by PNY. HP and Lenovo Dont make their own GPUs.

 

So, it was a PSU RMA rate, would you exclude brands like Corsair? Not manufacturing it doesn't mean that they have no control on the quality of what they ship. Linus doesn't sew backpacks in his house, but ultimately has a lot of bearing on the product that carries his brand.

[TVwazhere]

14 hours ago, SpaceGhostC2C said:

On the contrary, their minimum sample is size is about 10 times what you need to start getting reasonable approximations to a simple mean.

 

They do tell you that all samples are bigger than 300.

You only need 30-40 samples of something to be able to reasonably make purchasing decisions on upwards of $1000 products?

 

Respectfully, that's what some people will take away from this data. And while Lawlz did point out that one CPU/GPU purchase every year for 30-40 years results in only one defective part, I dont think most consumers are too confident if I showed them this data and said "Your Sapphire Card has a 3% change of having a defect, but dont worry!". From my experience helping people building PC's, they "would rather buy a PNY" even though they're probably cards used for reference board, minimum cooler design cards or pre built systems.

[SpaceGhostC2C]

8 hours ago, TVwazhere said:

You only need 30-40 samples of something to be able to reasonably make purchasing decisions on upwards of $1000 products?

Yes, you only need 30-40 observation in a random sample to start getting reasonable estimates of the average of a population characteristic and its variance. Needless to say, it's not ideal, but your estimated variance will reflect that. Not reporting an estimate for the variance or a confidence interval detracts a lot from its usefulness.

 

If the sample is not random, that's no longer the case, but more observations won't fix that either. You would need to correct for it, or readjust your question accordingly (for example, if all Lenovos are RTX Quadros and all Gigabytes are GT 1030s, it would be OK to estimate the failure rate of Gigabyte 1030s, but not the failure rate of Gigabyte in general, as you don't know which part is Gigabyte and which part is the 1030. With their data, they could perform multivariate analysis for that. What they report is barely scratching the surface of what their data allows).

 

8 hours ago, TVwazhere said:

Respectfully, that's what some people will take away from this data. And while Lawlz did point out that one CPU/GPU purchase every year for 30-40 years results in only one defective part, I dont think most consumers are too confident if I showed them this data and said "Your Sapphire Card has a 3% change of having a defect, but dont worry!". From my experience helping people building PC's, they "would rather buy a PNY" even though they're probably cards used for reference board, minimum cooler design cards or pre built systems.

I can't comment much on how potential readers will interpret the information provided. My comment was solely on statistical soundness, not on communication strategies.

[LAwLz]

13 hours ago, SpaceGhostC2C said:

Yes, you only need 30-40 observation in a random sample to start getting reasonable estimates of the average of a population characteristic and its variance. Needless to say, it's not ideal, but your estimated variance will reflect that. Not reporting an estimate for the variance or a confidence interval detracts a lot from its usefulness.

It's worth noting that while the failure rate data is based on at least 300 samples, which is pretty good (although more info would be better), the "return rate data" threshold is just 5 samples.