News Intel and AMD forge x86 ecosystem advisory group that aims to ensure a unified ISA moving forward

[stuff and nonesense]

But the price is directly derived by the size of the chip.

It is, and that is related to the yields. Physically smaller chips allow more chance of getting higher yields for a given size of wafer.

I’d suggest you have a look at how an IC is made. It’s fascinating. As for the number of transistors being a metric, I’ve never seen it other than being a boast by the manufacturers.. Red, green and blue.

 

[Kamen Rider Blade]

Well sure, reviews aren't directly talking about transistor counts cause the important part for a product is the price. But the price is directly derived by the size of the chip. inadvertently everyone is talking about transistor counts, since that's what it comes down to.

Let's say you buy a "Lego Set", do you care about the Brick Count, or what model you can make from the set?

99.99999% care about what Model / Diorama / PlaySet you can make.

Almost NoBody cares for the "Brick Count".

The same is with CPU's / GPU's.

No Body cares how many Transistors are in each Core or the total CPU, they care about how each Core can perform individually, & as a set within the CPU.

It's way more evident for example with GPUs. When the 4080 12gb was announced everyone was mad at nvidia cause it is such a small die at such a high price point. Cause obviously, a smaller die leads to less performance.

That's because nVIDIA has a history of trying to sell you less VRAM & Smaller SM Count GPU's for more money.

It's been a trend line for several generations.

They want to sell you less, for more $.

And then they justify the performance with DLSS & other BS to say "Here's your performance, take it or leave it".

 

[TheHerald]

Let's say you buy a "Lego Set", do you care about the Brick Count, or what model you can make from the set?

99.99999% care about what Model / Diorama / PlaySet you can make.

Almost NoBody cares for the "Brick Count".

The same is with CPU's / GPU's.

No Body cares how many Transistors are in each Core or the total CPU, they care about how each Core can perform individually, & as a set within the CPU.

This lego set can make a castle with 10.000 pieces. This other lego set can make the exact same castle with 1.000 pieces. If prices are the same, sure I don't care, but most likely the 10.000 pieces lego set will be more expensive since it costs more to make. And since the end result is the exact same castle, I absolutely care about buying the cheapest lego set that produces the same castle.

 

[Kamen Rider Blade]

This lego set can make a castle with 10.000 pieces. This other lego set can make the exact same castle with 1.000 pieces. If prices are the same, sure I don't care, but most likely the 10.000 pieces lego set will be more expensive since it costs more to make. And since the end result is the exact same castle, I absolutely care about buying the cheapest lego set that produces the same castle.

But what if the 10,000 piece Castle is more Epic, Grander, Larger?

And the 1,000 piece is smaller, less extravagant, quaint?

Obviously the 10,000 piece Castle set costs more.

Assuming you can afford both, but have to spend more, which one would you want?

How into Medieval Lego Castle sets were you?

I was always a Space Lego Kid myself.

 

[TheHerald]

But what if the 10,000 piece Castle is more Epic, Grander, Larger?

But it''s not. It's the exact same castle. 0 difference.

In this analogy, the pieces are the transistors and the castle is the performance. It's obvious that whoever designed the 10.000 pieces lego set and ended up at the same results as the 1000 pieces set, he did a damn bad job.

 

[stuff and nonesense]

This lego set can make a castle with 10.000 pieces. This other lego set can make the exact same castle with 1.000 pieces. If prices are the same, sure I don't care, but most likely the 10.000 pieces lego set will be more expensive since it costs more to make. And since the end result is the exact same castle, I absolutely care about buying the cheapest lego set that produces the same castle.

If it made the same size castle with a tenth of the pieces… a defect on a piece would take out 1 piece but that one piece is the same as 10 pieces in the other castle.

If you have a wafer with space for 200 chips, your size 2000 chips with the smaller ics…. 10 “identical” manufacturing defects would give a 95% yield on your big chips, 99.5% yield on the little chips and as I believe chips are sold by the wafer (happy to be corrected) you get a lot more chips for your money.

 

[Kamen Rider Blade]

But it''s not. It's the exact same castle. 0 difference.

In this analogy, the pieces are the transistors and the castle is the performance. It's obvious that whoever designed the 10.000 pieces lego set and ended up at the same results as the 1000 pieces set, he did a damn bad job.

But in this case, it's not the exact same castle.

Each one is built differently, functions differently.

Has different Pros / Cons.

 

[Vigilant256]

See, now we are getting somewhere. You know you could have posted that in page 1 instead of arguing about this useless disgustingly bad m3 chip.

So sure, if that chip is ~75B it seems to be competitive with the 128 epyc that uses 80b transistors. But the epyc was faster, wasn't it?

And uses way more power.

Nobody uses performance per transistor as a metric.

 

[Vigilant256]

The Herald here is trying so hard to cherry pick metrics to make apple m3 look bad.

Performance per watt
TheHerald:- Lets not talk about that

Battery Life
TheHerald:- Lets not talk about that

P-core area
The Herald:- Lets not talk about that

Alderlake has less transistors than M3
The Herald:- Yes lets talk about this, M3 sucks having more transistors than Alderlake and only comparable performance. Alderlake absolutely "crush" M3.

 

[TheHerald]

The Herald here is trying so hard to cherry pick metrics to make apple m3 look bad.

Performance per watt
TheHerald:- Lets not talk about that

Battery Life
TheHerald:- Lets not talk about that

P-core area
The Herald:- Lets not talk about that

Alderlake has less transistors than M3
The Herald:- Yes lets talk about this, M3 sucks having more transistors than Alderlake and only comparable performance. Alderlake absolutely "crush" M3.

Yeah, that's exactly it. I never talked about performance per watt.

 

[bit_user]

Transistors are the building blocks, if one ISA needs 10 times the materials to produce the same result as another ISA then yeah, it's going to be more expensive.

LOL, the irony of this statement is the ISAs that intrinsically need more transistors are CISC. In this case, x86 is the lone holdout. It's much more expensive to decode, which is (for the most part) all that ISA really has to do with implementation complexity.

Another way of saying it is that "if you held everything else about an x86 and an ARM CPU equal, the x86 one would be bigger". Now, people can debate how much bigger, but there's absolutely no question x86 requires bigger and more complex decoders.

 

[bit_user]

Yes, to get the performance of the 7950x you need twice the transistors of the 7700x and therefore it's going to cost twice as much.

It's not linear scaling, for a lot of reasons. There's only one IO die, interposer, package, contact grid, and heatspreader. The time & labor involved are also not scaling linear with the number of CCDs. As weight is virtually the same, shipping, warehousing, shelf space, and the box will be virtually unchanged.

The above are just the unit costs. You also have hardware & software engineering, sales & marketing, support costs, and warranty coverage, all of which are pretty much the same. So, while the dies are the most expensive single component, there's a lot else that goes into the cost of these products.

Saying nobody cares about it is insane.

What ultimately matters to users is the price they pay (and what they get for it). Exactly what goes into that price isn't something we have much visibility into or something we should worry too much about.

inadvertently everyone is talking about transistor counts, since that's what it comes down to.

No, it doesn't. A transistor's cost is affected by the node it's on. That's why Lunar Lake can have like 2.5 times as many transistors as an Alder Lake-P die and not cost 2.5 times as much.

 

[bit_user]

And uses way more power.

You're saying Nvidia Grace uses more power? No. 2x Grace CPUs + 2x 512 GB of DRAM use 500W. So, each Grace CPU + 512 GB of LPDDR5 memory use about 250W. The EPIC CPUs it was being compared with each used around 360 W, excluding DRAM.

 

[TheHerald]

No, it doesn't. A transistor's cost is affected by the node it's on. That's why Lunar Lake can have like 2.5 times as many transistors as an Alder Lake-P die and not cost 2.5 times as much.

It's been 7 pages. You still don't get what ceteris paribus is? Isolated test conditions? Come on now...

LOL, the irony of this statement is the ISAs that intrinsically need more transistors are CISC. In this case, x86 is the lone holdout. It's much more expensive to decode, which is (for the most part) all that ISA really has to do with implementation complexity.

Another way of saying it is that "if you held everything else about an x86 and an ARM CPU equal, the x86 one would be bigger". Now, people can debate how much bigger, but there's absolutely no question x86 requires bigger and more complex decoders.

As i've said, i really have no idea about the theory. Im just going by the results. Maybe m3 is just a terrible chip and not representative of arm, in which case sure, the comparison is invalid. I just used the m3 (and all its variations) because a lot of people marveled it, when in reality it's just a slow power hog. Are the new snapdragon X better than m3?

It's not linear scaling, for a lot of reasons. There's only one IO die, interposer, package, contact grid, and heatspreader. The time & labor involved are also not scaling linear with the number of CCDs. As weight is virtually the same, shipping, warehousing, shelf space, and the box will be virtually unchanged.

The above are just the unit costs. You also have hardware & software engineering, sales & marketing, support costs, and warranty coverage, all of which are pretty much the same. So, while the dies are the most expensive single component, there's a lot else that goes into the cost of these products.

You are arguing just for the sake of it my man. If the dies are the most expensive single component then why are you even arguing here? We agree, the majority of the cost is the actual die itself. So an ISA that can extract more performance with a lower transistor count is inherently superior simply because it's going to be cheaper to make and cheaper to sell. Especially since the only one that has the "luxury" of offering big dies for cheap is / was Intel due to them having their own fabs.

 

[palladin9479]

People, play nice and stop the back and forth.

 

[bit_user]

If the dies are the most expensive single component then why are you even arguing here?

Because there are lots of other factors that go into the price of these products. In the end, the price is what matters, not the transistor count. That's why nobody else is talking about it, virtually anywhere on the internet, except you. It's a misplaced concern.

We agree, the majority of the cost is the actual die itself.

That's not what I said. I said it's the biggest single component. It's like how the first ingredient in your favorite snack food might be sugar, but that doesn't mean the item is more than 50% sugar.