AMD Ryzen 9 5900X + Ryzen 9 5950X Dominate On Linux

[ms178]

With respect, we were already well into cash shortages and layoffs by 2007. The K10 architecture was just launching in 2007 and we lost ramp-up time via the TLB HW bug. AMD was deep into survival mode by the time Bulldozer development started.

If you want to say that Phenom and Barcelona were crap and they were developed when AMD was doing well that would be fair, but I don't hear that said much.

Mr. Bridgman, that it is exactly what I said in more generic terms. I don't remember the layoffs, but according to IDC's numbers AMD had a server market share of 14.6 % and total market share of 23.1% by Q4/2007. That is comparable to now, in fact it is still a tad better. My point still stands: All that financial success in that era didn't lead to great products later on (I could be more specific which errors, but people can read up that part of tech history themselves). I still remember the problems with AMD's execution in that era which not only lead to long product delays but also cost your company a lot of good will and trust with the OEMs which made the pitch with 1st Gen EPYC even harder.

I also don't see any cash shortages in the revenue numbers for 2007, there were some negative quarters but overall the financial performance was still far better than in the 2008/2009 time frame which you can see over here: https://www.macrotrends.net/stocks/c...MD/amd/revenue

It might be worth reading up more about Zen3 architecture - the CPU core was a bottom-to-top revision with non-trivial development costs. Saying that "the design process is comparatively cheap when compared to overall manufacturing costs" is closer to wishful thinking than fact.

I am not saying that no effort went into the design, but it is still a fact that Zen 3 doesn't include major new ISA features, uses the same process node which is very mature for a long time now, and that the manufacturing costs for the wafers are the dominent factor for the overall costs of the product over its life span, while the design costs are distributed over every server and client chip you sell due to the chiplet approach. Hence the proportion of the design costs fade over time while manufacturing costs dominate. With all due respect, that is not wishful thinking these are basic principles of economics.

I'd be glad if you could pass on the information to whom it may concern that at least some people like birdie and me don't buy your marketing messages and still value price/performance above all.

[bridgman]

Mr. Bridgman, that it is exactly what I said in more generic terms. I don't remember the layoffs, but according to IDC's numbers AMD had a server market share of 14.6 % and total market share of 23.1% by Q4/2007. That is comparable to now, in fact it is still a tad better. My point still stands: All that financial success in that era didn't lead to great products later on (I could be more specific which errors, but people can read up that part of tech history themselves). I still remember the problems with AMD's execution in that era which not only lead to long product delays but also cost your company a lot of good will and trust with the OEMs which made the pitch with 1st Gen EPYC even harder.

I also don't see any cash shortages in the revenue numbers for 2007, there were some negative quarters but overall the financial performance was still far better than in the 2008/2009 time frame which you can see over here: https://www.macrotrends.net/stocks/c...MD/amd/revenue

You might not have seen them but layoffs were already happening in 2007. You can find some of them with a quick search.

Agree that you would not see cash shortages in revenue numbers - that is not where they would appear - but if you click on Operating Income or Net Income tabs you'll get a better picture of what was happening. We lost over $2B in 2007 alone, which is crippling for a <$6B company. Agree that things got even worse in 2008/2009 but they were already disastrous.

I am not saying that no effort went into the design, but it is still a fact that Zen 3 doesn't include major new ISA features, uses the same process node which is very mature for a long time now, and that the manufacturing costs for the wafers are the dominent factor for the overall costs of the product over its life span, while the design costs are distributed over every server and client chip you sell due to the chiplet approach. Hence the proportion of the design costs fade over time while manufacturing costs dominate. With all due respect, that is not wishful thinking these are basic principles of economics.

I'd be glad if you could pass on the information to whom it may concern that at least some people like birdie and me don't buy your marketing messages and still value price/performance above all.

Again, I don't understand why you think development cost correlates with ISA features. It primarily correlates with changes to the execution pipeline, which was substantially changed in Zen3. Adding instructions is cheap by comparison.

Finally I don't understand why you insist on comparing lower bin 3xxx parts with high bin 5xxx parts and pretending the comparison is meaningful, particularly if you "value price/performance above all".

[smitty3268]

Just some food for thought for people who think along your lines: If your competitior sells overpriced shit for way more than you charge for your shit it is still shit these two would be selling. Makes that your shit any better?

What evidence do you have that the price of the CPU is too high? It seems like you just wish it was lower, which... Well, everyone always wishes everything was cheaper. I wish it was free, but that doesn't make it a fair price.

It's cheaper than the competition. It's clearly cheaper than what people are willing to pay for, because stock is flying off shelves right now.

It's not free, so it's more than what I'd prefer to pay for it.

How are you determining what a fair price is, if it's not one of the above criteria?

both are positioned as entry-level six core offerings

Maybe that's where you are going wrong. The 5600X is NOT positioned as an entry level 6 core part. It's positioned as a premium 6 core part that delivers huge increases to performance.

AMD isn't currently offering any entry level Zen 3 parts, but they are expected to come out next year as they get more TSMC production capability.

For consumers it is usually the price/performance metric which should be the basis of any buying decisions. At least that would be the rational choice. Hence I conclude from this launch that both companies negelct the value oriented buyers and AMD just had its Turing moment.

Price/performance has improved, though. It's just that they are only selling the high end parts to start the launch. It's well-known that you should wait for the non-X parts if you care about value and that was just as true with Zen 2 as it is with 3.

[Raka555]

This will be the final post to show how AMD "extorted" money from us back then, and were "saved" by Intel:


See https://www.anandtech.com/show/2051/7 for performance comparisson.

So back then Intel released their MUCH faster CPUs at 40% less than AMD was charging.

Now AMD launched their slightly faster CPUs at a 30%+ increase in price.

Go figure ...

[ms178]

What evidence do you have that the price of the CPU is too high? It seems like you just wish it was lower, which... Well, everyone always wishes everything was cheaper. I wish it was free, but that doesn't make it a fair price. It's cheaper than the competition. It's clearly cheaper than what people are willing to pay for, because stock is flying off shelves right now. How are you determining what a fair price is, if it's not one of the above criteria?

It is not what I wish for, this is purely analyzing and commenting on their pricing strategy. People like you still don't seem to comprehend the concept of price in relation to performance. For people like you, there is thankfully this graph from Hardware Unboxed which provides you with that data to see all the calculations in a consumer friendly way, and as you see you get exactly the same performance for your money in the much PR-touted gaming scenario (which is the best case scenario for Zen 3, multi-core performance hasn't improved that much) with the 5900X/3900X or nearly the same with the 5950X/3950X: https://static.techspot.com/articles...st_Average.png

For the 5900X that means STAGNATION (for the 5600X this means even a major REGRESSION when compared to the 3600) on the value front! How many times do I need to explain that to people until they realize this fact? As a customer I expect a significant improvement of at least 30% here with each new generation which would traditionally translate into better performance for the same price but again, we get exactly (or nearly) the SAME when setting performance into relation with the price for the mentioned CPU pairs and we get even LESS when comparing the 3600 to the 5600X. The single letter means nothing to me, it only signals that it is supposed to be a higher positioned SKU. The 3600 and 3600X could be clocked to the same level, you don't need an MBA to guess which one sold better, it was not even close.

Price/performance has improved, though. It's just that they are only selling the high end parts to start the launch. It's well-known that you should wait for the non-X parts if you care about value and that was just as true with Zen 2 as it is with 3.

That graph above and me reading the facts out loud from this graph proves that this is - strictly speaking - only true for the 5950X but not by much. Of course these numbers will change with the street prices over time but without major changes the margins should stay roughly the same.

[extremesquared]

I'm just hoping the cash influx from popular cpus makes it into the pockets of some additional driver developers on the gpu side of things. We all should be. As it stands, the only option for using consumer graphics cards for non-gaming tasks on linux has closed-source drivers, and strong cpu sales could potentially change that.

[smitty3268]

It is not what I wish for, this is purely analyzing and commenting on their pricing strategy.

You're doing so very poorly then.

People like you still don't seem to comprehend the concept of price in relation to performance.

Sorry, but the facts don't fit the narrative you are trying to spin here.

For people like you, there is thankfully this graph from Hardware Unboxed which provides you with that data to see all the calculations in a consumer friendly way, and as you see you get exactly the same performance for your money in the much PR-touted gaming scenario (which is the best case scenario for Zen 3, multi-core performance hasn't improved that much) with the 5900X/3900X or nearly the same with the 5950X/3950X: https://static.techspot.com/articles...st_Average.png

Wait, so now you're talking about the 5900X in only Windows gaming scenarios? I thought we were talking about the 5600X? And why not just use the performance tests Michael has run on this site?

Anyway, sure. Let's go with that chart you linked: The 5900X has identical price/performance ratio with both the 3900X and the 10900K. Ok, it's 1 penny off, but that's well within any margin of error and easily changed by testing different games. It's identical, not dramatically worse as you seem to be claiming. Since when is identical value something worthy of derision or complaint? Further, it's moved into a performance leadership spot vs the old part that was way behind - which typically carries a premium, since it's now the fastest part available rather than needing to compete on value.

And the 5950X is even worse for your argument - your own chart very clearly shows it's performing better per dollar than the older 3950X was.

(for the 5600X this means even a major REGRESSION when compared to the 3600) on the value front! How many times do I need to explain that to people until they realize this fact?

How many times do we need to explain to you that the 3600 doesn't compete with the 5600X, and that those are 2 completely different parts? Just compare the 3600 with the 3600X and you will similarly see that the 3600X is a terrible part based on value. Yet I don't see you complaining about how AMD has dared to release a 3600X part.

Your whole point of view seems to boil down to the fact that you wish AMD's CPUs sucked more so that they had to price them cheaper. I'm happy they don't suck.

As a customer I expect a significant improvement of at least 30% here with each new generation

You are, of course, free to expect whatever you want. But how many times throughout history has a 30% performance uplift per cpu generation occurred? Ryzen 1 only happened because AMD was incredibly far behind to start with. Maybe a couple parts per company in history? I think you're going to be deeply disappointed in pretty much every launch, ever.

[ms178]

You might not have seen them but layoffs were already happening in 2007. You can find some of them with a quick search.

Agree that you would not see cash shortages in revenue numbers - that is not where they would appear - but if you click on Operating Income or Net Income tabs you'll get a better picture of what was happening. We lost over $2B in 2007 alone, which is crippling for a <$6B company. Agree that things got even worse in 2008/2009 but they were already disastrous.

I fail to see your point here, relatively speaking the financially good times for AMD were in the 2004 - 2007 era, products conceived in that time frame didn't perfom that well in the market later. The point which the other user made is therefore countered with happenings in your company's past. It proves that the current success is not set in stone nor does past success dictates the success of tomorrow. The industry is simply too fast moving. To illustrate this point further I applauded the takeover of Xilinx a couple of days ago (hopefully the deal will close succesfully in a year) - when looking at the data center, AMD's current IP portfolio missed some offerings, e.g. leading networking IP, FPGAs, AI solutions and other parts your competitiors could integrate in the future to get a competitive advantage.

Again, I don't understand why you think development cost correlates with ISA features. It primarily correlates with changes to the execution pipeline, which was substantially changed in Zen3. Adding instructions is cheap by comparison.

I might be off here with that assumption, after all I am not a CPU engineer and for me it looks to be a much bigger effort to integrate AVX-512 and design everything around to feed these vector enginges properly. But my major talking point still stands that even if the Zen 3 core were an all-new effort and did cost a lot to design, the 7nm node is mature enough and yielding well from what I've read so far and the IO die is still the same which is made by GF, and you haven't come up with a convincing counter argument to my manufacturing cost dominates the overall product costs over time argument. I implied with that notion that Zen 3 is about cutting costs (as in amortizing at least some past investments which had to be made already with Zen 2) due to using the same node (no yield headaches with going with a new node) and with using the same IO die (no new design costs for these), which are major contributing factors to the overall manufacturing costs which are the dominant costs over the overall product life cycle, not the pure design costs. The chiplet design allows you to spread the design costs among server and client products which - again - helps to decreases these design costs significantly over time. We customers don't get to see the benefit of these cost saving measures in form of better prices though, at least not yet.

I am just pointing out loud that this launch seems to me to be about optimizing your margins first on two fronts, with a price increase and manufacturing cost savings.

Finally I don't understand why you insist on comparing lower bin 3xxx parts with high bin 5xxx parts and pretending the comparison is meaningful, particularly if you "value price/performance above all".

I just told another user which I would think is a crystal clear point: A look at the price/performance graphs of Hardware Unboxed provides you with that data and as you see you get exactly the same performance for your money in the much PR-touted gaming scenario (which is the best case scenario for Zen 3, multi-core performance hasn't improved that much) with the 5900X/3900X or nearly the same with the 5950X/3950X: https://static.techspot.com/articles...st_Average.png

These are not comparisons betwen lower bin and higher bin parts - these are comparisions between direct successors!

For the 5900X these numbers mean STAGNATION, for the 5600X this means even a major REGRESSION in that metric when compared to the 3600 when assuming similar performance levels! As a customer I expect a significant improvement of at least 30% here with each new generation which would traditionally translate into 30% better performance for the same price but again, we get exactly (or nearly) the SAME when setting performance into relation with the price for the mentioned CPU pairs and we get even LESS when comparing the 3600 to the 5600X. The single X letter means nothing to me, it only signals that it is supposed to be a higher binned SKU. The reality with the 3600 and 3600X was that you could clock both to the same level, you don't need an MBA to guess which one sold better, a hint: it was not even close.
Of course these numbers will change with the street prices over time but without major price changes between the two Zen generations the margins should stay roughly the same. I hope I have made my points more clear, English is not my first language.

[ms178]

You're doing so very poorly then.


Sorry, but the facts don't fit the narrative you are trying to spin here.


Wait, so now you're talking about the 5900X in only Windows gaming scenarios? I thought we were talking about the 5600X? And why not just use the performance tests Michael has run on this site?

Anyway, sure. Let's go with that chart you linked: The 5900X has identical price/performance ratio with both the 3900X and the 10900K. Ok, it's 1 penny off, but that's well within any margin of error and easily changed by testing different games. It's identical, not dramatically worse as you seem to be claiming. Since when is identical value something worthy of derision or complaint? Further, it's moved into a performance leadership spot vs the old part that was way behind - which typically carries a premium, since it's now the fastest part available rather than needing to compete on value.

And the 5950X is even worse for your argument - your own chart very clearly shows it's performing better per dollar than the older 3950X was.


How many times do we need to explain to you that the 3600 doesn't compete with the 5600X, and that those are 2 completely different parts? Just compare the 3600 with the 3600X and you will similarly see that the 3600X is a terrible part based on value. Yet I don't see you complaining about how AMD has dared to release a 3600X part.

Your whole point of view seems to boil down to the fact that you wish AMD's CPUs sucked more so that they had to price them cheaper. I'm happy they don't suck.


You are, of course, free to expect whatever you want. But how many times throughout history has a 30% performance uplift per cpu generation occurred? Ryzen 1 only happened because AMD was incredibly far behind to start with. Maybe a couple parts per company in history? I think you're going to be deeply disappointed in pretty much every launch, ever.

Sorry, I am not wasting my time with you any longer, birdie and me pointed out the obvious to anyone here. Hardware Unboxed told you the numbers, if you sill praise the Zen 3 launch you fail to take pricing into account. It is just that simple.

[pal666]

Intel wins in the AV1 decoding test and is therefore the fastest! /s

all intel wins were in single core tests, probably with manual assembler code targeting intel cpu