Still waiting to upgrade my gaming system

[Alexey]

Because Intel didn't want you to be able to use all your PCI lanes unless you buy their $1,200 processor?

I wonder is that pure marketing or is there some technical reason?

 

[phaZed]

I wonder is that pure marketing or is there some technical reason?

IIRC it was part of the silicon "binning".. "bad wafer chip" CPU's had their defective cores disabled in production, hence, cutting the available PCIe lanes, and giving us the different classes/models/Mhz etc.

 

[sapphirescales]

I wonder is that pure marketing or is there some technical reason?

Seeing as this was never an issue before, I call BS.

IIRC it was part of the silicon "binning".. "bad wafer chip" CPU's had their defective cores disabled in production, hence, cutting the available PCIe lanes, and giving us the different classes/models/Mhz etc.

If this is true then why was this never a problem before? I never remember having to worry about being able to use all the resources of my motherboard because I didn't buy an expensive enough processor before.

 

[phaZed]

If this is true then why was this never a problem before? I never remember having to worry about being able to use all the resources of my motherboard because I didn't buy an expensive enough processor before.

I agree with you on the PCIe lane "upsell" BS.
For right or for wrong, PCIe is a "costly" item to implement in both size on the die and cache/buffer/transport etc and really only makes sense to implement it per-processor, up to the 24 physical lanes or whatnot IF you want the full performance of the CPU-PCIe - so, that cuts out the previous generation tech of Northbridge/southbridge.. which simply isn't going to do these 2500-3000MB/s types of speeds on standard PCB interconnects.

That being said, there is no reason that there COULD be a switchable emulator or bridge implementation for all the SATA stuff and x1/x4 PCIe - so, I'm sure if they wanted it enough, at the cost of some PCIe speed, they could do something. It's just convenient not to.

But ya, this is BS:
Opinion: Why Counting ‘Platform’ PCIe Lanes (and using it in Marketing) Is Absurd

In Intel’s documentation, it explicitly lists what is available from the processor via the PCIe root complexes: here 44 lanes come from two lots of sixteen and one twelve lane complex. The DMI3 link to the chipset is in all but name a PCIe 3.0 x4 link, but is not included in this total.

On the other hand, PCIe and "things" that can actually take advantage of this technology are fairly new. NVMe SSDs, Graphics cards that are pushing the boundaries of their interconnects... It was just 3 or 4 years (?) since NVMe for consumers. I remember in the not too distant past.. pretty much every motherboard with PCIe 1.0 you got was 1 x16 slot, then an x8 or two, with some little ones sprinkled in and/or shared. M.2 What?

So, maybe why you don't remember it being a problem before is because we have never been so dependent on it as we are now?

 

[Alexey]

Well, dual CPU systems had it for ages, to use all memory slots and/or all PCI-Es, both CPUs must be installed. With one, you only get half the memory slots and also some PCI slots are lost. If this applies to multiple CPU sockets, I see this can apply to cores (which are to an extent independent CPUs on a single chip).

 

[sapphirescales]

So, maybe why you don't remember it being a problem before is because we have never been so dependent on it as we are now?

It's not about the specific technology. I've been building computers since the Windows 95 days and I've NEVER had to worry whether or not I could use all the ports and features of my motherboard before. The ports and features were motherboard dependent. It didn't matter what CPU you had. You could have the cheapest CPU or the fastest, most expensive CPU and all the features of the motherboard would work regardless. Crippling the features of your motherboard unless you buy a super expensive CPU is idiotic. AMD isn't pulling that sh*t. That's like buying a car and limiting the size of the gas tank based on whether you bought the leather seats or not. "Oh, you want more than that tiny little 10 gallon gas tank? Well then you have to upgrade to our luxury interior. No, we don't sell the upgraded gas tank separately. Oh - and the luxury interior doubles the price of the car. Sound like a good deal?" F*ck no.

 

[Alexey]

The ports and features were motherboard dependent. It didn't matter what CPU you had.

It often depended on how many CPUs you had, and it still does, for dual CPU motherboards (and I guess for quads but I never worked with those).

 

[sapphirescales]

It often depended on how many CPUs you had, and it still does, for dual CPU motherboards (and I guess for quads but I never worked with those).

I don't know about that. I can count the number of times I've worked with dual CPU motherboards on one hand.

 

[NJW]

I can count the number of times I've worked with dual CPU motherboards on one hand.

But that's @Alexey's point (I think ...) – a modern multi-core CPU is essentially multiple processors in a single package and you're working with dual (or more) CPU motherboards. With increasing complexity in a single package, yield almost certainly goes down, so there is a huge incentive to recover something marketable from the waste stream, hence selecting 'lower spec' product from what they've produced.

It has always been thus, especially with budget CPUs. Earlier generations were selected according to their speed capability, now it's how much of the silicon actually works to spec. Some of us remember when you had to pay extra for a high-gain variant of a transistor, or a 5% tolerance resistor.

 

[sapphirescales]

But that's @Alexey's point (I think ...) – a modern multi-core CPU is essentially multiple processors in a single package and you're working with dual (or more) CPU motherboards. With increasing complexity in a single package, yield almost certainly goes down, so there is a huge incentive to recover something marketable from the waste stream, hence selecting 'lower spec' product from what they've produced.

It has always been thus, especially with budget CPUs. Earlier generations were selected according to their speed capability, now it's how much of the silicon actually works to spec. Some of us remember when you had to pay extra for a high-gain variant of a transistor, or a 5% tolerance resistor.

All this is great - in theory. But then why doesn't AMD try to pull this BS then and they're still able to provide their processors for less?

 

[Sky-Knight]

All this is great - in theory. But then why doesn't AMD try to pull this BS then and they're still able to provide their processors for less?

Because they do, all of this is fluff.

A mainboard has resources, a CPU has resources, a mainboard that can handle a range of CPUs might have resources that don't get used unless you have a larger CPU. And ALL desktop mainboards are short on PCIe lanes. I get people trying to use crap to run Untangle and then they wonder why they bottleneck on throughput... It's almost ALWAYS the PCI bus.

If you want more lanes you need a server board, or a high end desktop board. So in almost all cases a modern mainboard doesn't have enough lanes to support a modern CPU at the desktop level. The only place I've seen a CPU have more lanes than the mainboard can support is on the AMD side because they insist on this asinine backwards socket compatibility game that doesn't do anything but make working with firmware a nightmare.

This is only an issue for gaming rigs thanks to NVME, it's the only thing that pushes your needs into a place where some congestion on the bus actually happens. If you're running one of those huge boards with 4x GPUs in it that'd do it too... but if you're doing that you should be forking over big $$ for a huge mainboard AND CPU for other reasons.

 

[phaZed]

All this is great - in theory. But then why doesn't AMD try to pull this BS then and they're still able to provide their processors for less?

Well, the short answer is: AMD put themselves in that position, and were able to, seeing as they didn't have to do anything for a long time. So, good for them.. a decision in the right direction, but also an easy one to make at the design phase.

Intel, on the other hand, has an architecture already. Remember, each PCIe "lane" is two physical wires to the CPU.. two pins on the CPU.. 24 lanes = 48 pins on the CPU. So, there are some growing pains at Intel that were essentially turned into cash cows with different sockets and chipsets... ya, it sucks... but there is a logical real-world limitation/reason for this that is going to require Intel to take serious time and money at the design phase to rectify. Like Fight club and the "Insurance equation", if X+Y is more than the return.. we don't do it! You have to "think corporate" when looking at Intel. Ya, that sucks.

IMO, AMD's quality control is lackluster and/or their chips fail too often - AKA cost-savings! This has been AMD's problem for a while too. AMD's sockets are cheaper to make (Pins instead of Pads), the cooler mounts are inferior to Intel's, "Infinity Fabric" is for the 2 or 4 die layout abilities (Re Intel: 'Glued Together') - all cost-cutting measures that do have performance and longevity implications.

Serendipitous picture of the day - my 2nd Ryzen embedded graphics failure (VEGA 11) this month (Ryzen 5 2400g this time), this one in a CyberPowerPC:



I can't remember the last Intel CPU I replaced.. since the Pentium/Core2Duo days at least. I've replaced 4 or 5 Ryzens overall.

 

[sapphirescales]

but there is a logical real-world limitation/reason for this that is going to require Intel to take serious time and money at the design phase to rectify.

Yes, I agree. But they never should have released these processors until they fixed this problem. You don't release a car that won't fire on all cylinders. You shouldn't release a CPU with bottlenecks that will prevent people from being able to use all the features of their motherboard.

Serendipitous picture of the day - my 2nd Ryzen embedded graphics failure (VEGA 11) this month (Ryzen 5 2400g this time), this one in a CyberPowerPC:

I can't remember the last Intel CPU I replaced.. since the Pentium/Core2Duo days at least. I've replaced 4 or 5 Ryzens overall.

It's all anecdotal at this point, but I've seen a lot more CPU failures lately across the board. I've seen 3x i7-6700k failures in the last few months. I wouldn't be surprised if there was an actual problem with that CPU. I've also seen a few dead AMD's. I think quality is definitely getting worse nowadays. CPU failures used to be unheard of.

 

[phaZed]

Yes, I agree. But they never should have released these processors until they fixed this problem. You don't release a car that won't fire on all cylinders. You shouldn't release a CPU with bottlenecks that will prevent people from being able to use all the features of their motherboard.

But in there lies the problem. The technology is growing, it didn't just "BAM!" show up one day. The limitations are not "problems" as in defects, they are problems only recently as the technology has started moving in that direction, with previous iterations already in place. It's the same 'ol story, all over again.. this is how it works...

eg. IF NVIDIA RTX's Ray-tracing is eclipsed in 4 or 5 years.. imagine everything is going to real time RT... would you then look back at the first gen RTX cards and say "Well how stupid was that that it only did X or Y!?"... Well, right now, it's a hell of a thing. So damned if you do, damned if you don't?


Now, to be fair, if you are buying a motherboard and CPU as a consumer - Intel doesn't hide these facts and the Motherboard manufacturers have their "abilities" listed... so the onus is on the consumer to make the correct decision of CPU/MB they require.

In your car analogy, you get to buy all the options.

So, your analogy could be looked at differently:

You can buy the 4 cylinder or a 6 cylinder or an 8, maybe turbo's.. It's the same chassis platform underneath but they put in a tiny engine for the lowest cost model. So, for example, should we blame Dodge for using a 400-700HP capable chassis for a 290HP 6 cylinder engine?





"Well, look at that! The same car for twice the price!? And I have to buy the one that's 60% MORE to have the things I want!? WTF"

It's capitalism, bud. Welcome.

 

[NJW]

You don't release a car that won't fire on all cylinders. You shouldn't release a CPU with bottlenecks that will prevent people from being able to use all the features of their motherboard.

Modern cars frequently have exactly the same engine in several models, the only difference being the ECU firmware, which dictates engine power and torque curves, rev. limit, etc. You want more performance? You pay for that. The incremental cost to the manufacturer is ~zero.

 

[Thomas B. Fournier]

I think it's going too far guys... lol.

AMD has never been a true alternative since the Athlon/Phenom days, they are coming back with some competitive offerings, why so quick to put it down?

Intel has been holding back on certain features behind huge paywalls, you may think it's "OK" but personally I am glad AMD is coming back to push shoulders again.

There is a huge market for mid-range computer parts and AMD is definitely on the right track (If not already more competitive than Intel in the mid-range side of things) if you ask me.

Any competition is great and should be welcomed with open arms.

 

[Sky-Knight]

AMD isn't Intel's competition though... That's the problem. They've carved up the market into the respective bits they want, and they exist to compete with each other only in enough ways to ensure anti-trust doesn't happen.