News AMD deep-dives Zen 5 — Ryzen 9000 and AI 300 benchmarks, Zen 5, RDNA 3.5 GPU, and XDNA 2 microarchitectures

[TheHerald]

Trying to group products by market segment. I maintain that it doesn't make sense to compare products in different market segments, which TheHerald has been doing in the past.

IMO, the comparison that's actually useful to people is to look at performance and efficiency of products in the same/adjacent market segments, as a user would typically configure them. Comparing anything else is a bit silly, as is looking only at efficiency, in isolation of things like price and performance..

But isn't the 13700kf (266$) at the same segment as the 7700x (246$) ? Like how much closer do the segments need to be ? They are almost identical in price.

Isn't the 7800x 3d (385$) in a higher price segment than the 13700k? It's still losing - by a lot - in MT efficiency.

But market segments are dictated by MSRP and naming schemes. A company dropping prices doesn't make their products any more efficient. I mean prime example, the 10900 (10850)k. It was supposed to compete against the 3900x / 5900x. It was losing by both in ISO efficiency. Then Intel slashed the price down to 370€ , similar price to the 5600x back then. Does that suddenly mean that the 10850k was an insanely efficient cpu just because it was more efficient than the 5600x? Of course not.

 

[stuff and nonesense]

The 7900 doesn't draw 65w either, TDP is NOT PPT. Just because amd is fooling you with those numbers doesn't mean it fools everyone else. According to the gnexus review, the 7900 draws 86 watts, not 65.

Computerbase has tested on every possible configuration btw, here is the data

The 13900k is the most efficient chip they've ever tested at 45w btw. You can also clearly see the i5 beating the r5 handily and the i7 beating the R7 at most configurations.

So if efficiency is your TOP priority, you buy an Intel. Period. Both ST and MT efficiency they have the lead.

https://www.tomshardware.com/reviews/intel-core-i9-13900k-i5-13600k-cpu-review/4

Nuff said.

 

[bit_user]

But isn't the 13700kf (266$) at the same segment as the 7700x (246$) ? Like how much closer do the segments need to be ? They are almost identical in price.

The further down the price scale you go, the more the prices seem to bunch up. They should be compared proportionately, rather than in absolute terms.

However, at that price, I'll agree that the i7-13700KF is definitely a better value. Here's where PCPartPicker's price history becomes relevant. It lets us see that it only dropped below $295 a little over a week ago! That price was only reached in June. Before that, it was generally selling in the low $300's.

Compare that to the i7-14700KF, which has been selling for $350 to $375, and it seems that Intel is moving aggressively to clear out the stock of the old Gen 13 product.

Isn't the 7800x 3d (385$) in a higher price segment than the 13700k? It's still losing - by a lot - in MT efficiency.

I've said before: that's a nonsense comparison, because the main reason to buy a 7800X3D is for gaming and the i7-13700K can't touch it, in gaming performance.

market segments are dictated by MSRP and naming schemes.

Nope. A consumer only cares about what price and performance something is now. They don't care about names, when considering which CPU to buy, and they look at street prices instead of MSRP.

 

[thestryker]

I knows the fact that AMD CPU are hard to cool so that works at high temperatures but this does not means that is more difficult to dissipate 1000W than 500W and more difficult to dissipate 500W than 100W and so on.
What is not clear of this ?
So, in the end, Intel CPUs that producing more heat are also more difficult to cool.
In simple terms you need more cooling.

What's not clear is how you can somehow not understand that heat transfer matters. If a cooler on Intel can dissipate 250W of heat while the CPU is running 82.5C, but on AMD that same cooler can only dissipate 239.7W with the CPU throttling that means it's easier to cool the Intel part. The coolers have more thermal capacity than AMD can utilize due to poor transfer. This is really simple and something that you seem to absolutely not care about understanding as there have been multiple posts explaining just that.

 

[usertests]

I was hoping maybe we see a refresh of the current Radeon discreete GPUs.... Perhaps a Radeon RX 7950 XTX w/ RDNA 3.5 .. since the 7950 was never released... perhaps it could be with the RDNA 3.5 enhancements?

Appears RDNA 4 (RX 8000 series cards) won't release anything that is faster faster than the current RX 7900 XTS (RNDA 3). A faster RX 7950 XTX w/ RDNA 3.5 tech would be a good holdover, no?

It's not a refresh if they use RDNA3.5, it's a whole new die.

It has long been rumored that top RDNA4 will not beat the 7900 XTX (raster), but maybe expectations are too low. What it should do is bring 7900 XT-like performance to a lower price point. That seems like a good enough plan to me.

Only having 16 GB VRAM would be a concern for some, but they could fix that by allowing partners to include 32 GB.

 

[TheHerald]

The further down the price scale you go, the more the prices seem to bunch up. They should be compared proportionately, rather than in absolute terms.

However, at that price, I'll agree that the i7-13700KF is definitely a better value. Here's where PCPartPicker's price history becomes relevant. It lets us see that it only dropped below $295 a little over a week ago! That price was only reached in June. Before that, it was generally selling in the low $300's.

Compare that to the i7-14700KF, which has been selling for $350 to $375, and it seems that Intel is moving aggressively to clear out the stock of the old Gen 13 product.


I've said before: that's a nonsense comparison, because the main reason to buy a 7800X3D is for gaming and the i7-13700K can't touch it, in gaming performance.


Nope. A consumer only cares about what price and performance something is now. They don't care about names, when considering which CPU to buy, and they look at street prices instead of MSRP.

Even percentage wise the difference between the 2 chips is less than 10%. Of course the 13700kf is a better value but that wasn't the point, the point was ISO efficiency. Apparently AMD does not have the lead in that. It does have a lead on one particular segment, the high end, where the 7950x is currently untouchable (in most power limits, but not all). The absolute efficiency king is Intel but I'd argue 45w or lower isn't a very realistic power to run such a CPU at.

At 1440p with a 4090 (lol) the 13700k gets 94% of the gaming performance.

What consumers do is irrelevant here. Your argument is basically that now that Intel dropped prices, the efficiency of their cpus has increased. Which is just silly.

 

[bit_user]

What consumers do is irrelevant here.

It's very relevant. It's arguably the only thing that matters.

Your argument is basically that now that Intel dropped prices, the efficiency of their cpus has increased. Which is just silly.

No. The way I look at it is that within that market segment, there's now a more efficient option. That also happens as new products are launched. Markets do not remain frozen in time.

 

[thestryker]

I was hoping maybe we see a refresh of the current Radeon discreete GPUs.... Perhaps a Radeon RX 7950 XTX w/ RDNA 3.5 .. since the 7950 was never released... perhaps it could be with the RDNA 3.5 enhancements?

Appears RDNA 4 (RX 8000 series cards) won't release anything that is faster faster than the current RX 7900 XTS (RNDA 3). A faster RX 7950 XTX w/ RDNA 3.5 tech would be a good holdover, no?

Everything AMD has been talking about with RDNA 3.5 is efficiency related. This likely explains why they called it a half improvement. To utilize this in discrete form they would likely have to enter into a new production contract with TSMC. The problem here being that the chips would still be the same size and may very well cost more to produce and not perform better. Thus AMD would potentially be paying more to get 7000 series GPUs that were only more efficient than the existing ones. This wouldn't make for a sound business decision.

 

[NinoPino]

What's not clear is how you can somehow not understand that heat transfer matters. If a

Yes, I understand and not negate the importance of heat transfer efficiency.
As I wrote in other posts before.

cooler on Intel can dissipate 250W of heat while the CPU is running 82.5C, but on AMD that same cooler can only dissipate 239.7W with the CPU throttling that means it's easier to cool the Intel part.

Here is the point.
The AMD CPU at 239W is arrived and give you 100% of expected performance, while for the Intel CPU need an extra 50/100/150W.
Ok if you say it is easier to cool ISO watt, but considering the total dissipation needed to obtain the nominal performance I'm saying that it's more difficult to cool the more hungry CPU.

The coolers have more thermal capacity than AMD can utilize due to poor transfer. This is really simple and something that you seem to absolutely not care about understanding as there have been multiple posts explaining just that.

This imply that, to maintain the AMD CPU at full speed, we need a more performant cooler that what should be necessary.
And with this I totally agree.
But how much bigger is the problem ? Bigger than Intel ?
Sorry but without a proof I cannot believe that to cool a 240W CPU we need a bigger cooler that to cool a 350/400W one.
To clarify, with "easier to cool" I intend that we need a lesser performant cooling system.

 

[thestryker]

Here is the point.
The AMD CPU at 239W is arrived and give you 100% of expected performance, while for the Intel CPU need an extra 50/100/150W.
Ok if you say it is easier to cool ISO watt, but considering the total dissipation needed to obtain the nominal performance I'm saying that it's more difficult to cool the more hungry CPU.

Your "point" is utterly false and easily proven so:

Highest power consumption in their stock 13900K testing is blender at 283W then with high speed memory 287W and this can be cooled by a Thermalright Peerless Assassin:
https://www.techpowerup.com/review/intel-core-i9-13900k/22.html

Highest power consumption in their stock 7950X testing is also blender which comes in at 235W then 247W with high speed memory. The former can be cooled by a Thermalright Peerless Assassin, but the latter cannot:
https://www.techpowerup.com/review/amd-ryzen-9-7950x/24.html

Both of these CPUs are giving "expected" performance yet one needs a better cooler than the other, and it happens to be the AMD one using less power.

 

[-Fran-]

Trying to group products by market segment. I maintain that it doesn't make sense to compare products in different market segments, which TheHerald has been doing in the past.

IMO, the comparison that's actually useful to people is to look at performance and efficiency of products in the same/adjacent market segments, as a user would typically configure them. Comparing anything else is a bit silly. Also silly: looking only at efficiency, in isolation of things like price and performance, because nobody who would seriously consider a 7950X3D would just change their mind and buy a i5-13400F, instead. Probably not even a i9-14900T, either.

Ah, ok; that makes sense.

It's a slippery slope as prices are moving all the time and what was meant to be within a certain segment could slide to another rather easily.

There's 5 (edited, lol) metrics being discussed at any given time, so the ping pong is rather weird to me:
- Power
- Performance
- Price
- Intended/Perceived Segment
- Operating Temperature (edited)

All I know is that AMD is more power efficient in the ranges that matter.

Regards.

 

[TheHerald]

Your "point" is utterly false and easily proven so:

Highest power consumption in their stock 13900K testing is blender at 283W then with high speed memory 287W and this can be cooled by a Thermalright Peerless Assassin:
https://www.techpowerup.com/review/intel-core-i9-13900k/22.html

Highest power consumption in their stock 7950X testing is also blender which comes in at 235W then 247W with high speed memory. The former can be cooled by a Thermalright Peerless Assassin, but the latter cannot:
https://www.techpowerup.com/review/amd-ryzen-9-7950x/24.html

Both of these CPUs are giving "expected" performance yet one needs a better cooler than the other, and it happens to be the AMD one using less power.

Not to mention there is no reason to strictly compare 13900k to 7950x. Both Intel and AMD make other cpus. I'd argue the 13700, a 65w chip, has way better dissipation and a lot less power to cool compared to the 7700x.

I did a test with a friend btw, his 7700x at 110w was hitting 86-88C on a be quiet loop2. My 12900k (similar cpu to the 13700) on a single tower air cooler at the same wattage was below 60. The difference in cooling requirements immense. No idea why people pretend otherwise.

 

[TheHerald]

Ah, ok; that makes sense.

It's a slippery slope as prices are moving all the time and what was meant to be within a certain segment could slide to another rather easily.

There's 5 (edited, lol) metrics being discussed at any given time, so the ping pong is rather weird to me:
- Power
- Performance
- Price
- Intended/Perceived Segment
- Operating Temperature (edited)

All I know is that AMD is more power efficient in the ranges that matter.

Regards.

You keep saying it but it's not true. The 13700 is more efficient than the 7700x in every mt workload at all the ranges that matter for example. Why are people like you claiming otherwise I don't get...

 

[bit_user]

The difference in cooling requirements immense. No idea why people pretend otherwise.

Don't confuse cooling requirements with cooling efficacy. The R7 7700X can be air-cooled without throttling, if only barely.

Part of the reason for its lower temps was that the Intel CPU simply had lower thermal density across the die. Like way lower, in the case of that particular CPU! This is the unpleasant reality Intel is going to face, when it moves to smaller nodes:

Source:

https://twitter.com/x/status/1564409602933792770

View: https://x.com/harukaze5719/status/1564409602933792770

​

 

[TheHerald]

Don't confuse cooling requirements with cooling efficacy. The R7 7700X can be air-cooled without throttling, if only barely.

Part of the reason for its lower temps was that the Intel CPU simply had lower thermal density across the die. Like way lower, in the case of that particular CPU! This is the unpleasant reality Intel is going to face, when it moves to smaller nodes:

​

​

Source:​

https://twitter.com/x/status/1564409602933792770

View: https://x.com/harukaze5719/status/1564409602933792770​

​

​

Does it matter why? The end result is an Intel cpu at the same segment is faster, more efficient, runs considerably cooler with an entry level budget cooler. The 20$ extra that the 13700 costs over the 7700x will be completely covered by the extra cooling requirements the 7700x needs.

 

[bit_user]

Does it matter why?

Yes, because this article is about Zen 5, not Zen 4. Intel will soon also be moving beyond Raptor Lake. The issue of thermal density is one they will both struggle with, in future. In AMD's case Zen 5 is on a lower-density node than Arrow Lake, which might somewhat reverse the situation between Ryzen 9000 and Arrow Lake.

I even wondered if AMD's decision to use TSMC N4P for Zen 5, while they're using N3 for Zen 5C, was based partly on thermal density.

 

[TheHerald]

Yes, because this article is about Zen 5, not Zen 4. Intel will soon also be moving beyond Raptor Lake. The issue of thermal density is one they will both struggle with, in future. In AMD's case Zen 5 is on a lower-density node than Arrow Lake, which might somewhat reverse the situation between Ryzen 9000 and Arrow Lake.

I even wondered if AMD's decision to use TSMC N4P for Zen 5, while they're using N3 for Zen 5C, was based partly on thermal density.

It's not just about the density, aren't pcores much bigger in transistor count anyways? So even if they had same density theyd be easier or cool, no? Also even though the 13900k can pull more power for example, it's spread out over a bigger number of cores than the 7950x.

It's the whole reason why eg 12900k hits 100c at 260w while the 13900k is sitting at 80, both with the same cooler.

 

[bit_user]

It's not just about the density, aren't pcores much bigger in transistor count anyways? So even if they had same density theyd be easier or cool, no?

I'm getting a figure of 100.7 MTr/mm^2 for the Intel 7 node and 138.2 MTr/mm^2 for TSMC N5. That's 37.2% higher density, which seems hard to dismiss.

Yes, it doesn't fully account for the discrepancy, but it'd sure make something like the KS-series CPUs a lot harder to cool, with the thermal density @ 350W going up to 1.87 W/mm^2, which is even higher than the R9 7950X's figure @ 230 W.

Of course, Arrow Lake is not using TSMC N5, but rather will use Intel 20A. I haven't found figures on the supposed density improvements of that node, but it's a good bet it'll be denser than TSMC N4P.

 

[-Fran-]

You keep saying it but it's not true. The 13700 is more efficient than the 7700x in every mt workload at all the ranges that matter for example. Why are people like you claiming otherwise I don't get...

As I said, no matter how much of a colourful interpretation you want to give to any graph and piece of data you come across, it's already stablished AMD is more efficient.

Can you say a 13700K makes more sense than a 7700X for specific workloads? Yes, you can. Is the 13700K more efficient than the 7700K in all cases? No, it is not.

Look at the 7800X3D and gaming against the 13700K in games if you want. Or the 14900KS. Does the 7800X3D make sense for every situation and game? No. Does the 13700K make sense for every sitation and game? No. Different needs different CPUs you should get. Across the board, you can find a more efficient AMD CPU than the Intel counterpart for any given task, and that is a problem for Intel. Would they be cheaper? Not necessarily. Would they run cooler? Not necessarily. They are* separate things.

Remember that power efficiency is a measure of unit of work per unit of energy consumed. Temperature has nothing to do with it and it's a consequence.

Regards.

 

[TheHerald]

Can you say a 13700K makes more sense than a 7700X for specific workloads? Yes, you can. Is the 13700K more efficient than the 7700K in all cases? No, it is not.

Yes it is, but let's ignore that for a moment. Let's just go with your argument, that sometimes the 13700k is more efficient and sometimes it isn't. So how the hell do you then conclude that amd is more efficient???

Ok find me a more efficient amd chip for mt workloads at or below 265$. I'll be waiting.

 

[-Fran-]

Yes it is, but let's ignore that for a moment. Let's just go with your argument, that sometimes the 13700k is more efficient and sometimes it isn't. So how the hell do you then conclude that amd is more efficient???

Ok find me a more efficient amd chip for mt workloads at or below 265$. I'll be waiting.

Going by this: https://gamersnexus.net/cpus/intel-problem-cpu-efficiency-power-consumption

And their power efficiency chart using Blender, I can say the 5600X is your CPU for efficiency and, funnily enough, the 7700X is as well (since it's right at the price point you mentioned). Also note where the 13700K is.

"AH! BUT IT DOESN'T PERFORM THE SAME!"

Well, yeah; Power Efficiency is not a raw Performance metric. Hence my statements.

"AH! BUT IT'S JUST BLENDER! WHAT ABOUT OTHERS?!"

Well, I did say it doesn't make sense in all scenarios, didn't I?

Regards.

 

[TheHerald]

Going by this: https://gamersnexus.net/cpus/intel-problem-cpu-efficiency-power-consumption

And their power efficiency chart using Blender, I can say the 5600X is your CPU for efficiency and, funnily enough, the 7700X is as well (since it's right at the price point you mentioned). Also note where the 13700K is.

"AH! BUT IT DOESN'T PERFORM THE SAME!"

Well, yeah; Power Efficiency is not a raw Performance metric. Hence my statements.

"AH! BUT IT'S JUST BLENDER! WHAT ABOUT OTHERS?!"

Well, I did say it doesn't make sense in all scenarios, didn't I?

Regards.

You were talking about iso power before. Now you are changing your argument to out of the box, and then you are only using the power unlimited intel chips to compare.

Cant you see that your method is incredibly biased. ComputerBase has run the tests for us, for most scenarios the 13700k is faster than the 7700x while consuming the same power, meaning it's both faster and more efficient in their entire mt suite.

People just don't want to admit it but in most segments intel is leading in efficiency either out of the box (non k and t) or at iso efficiency.

 

[-Fran-]

You were talking about iso power before. Now you are changing your argument to out of the box, and then you are only using the power unlimited intel chips to compare.

Cant you see that your method is incredibly biased. ComputerBase has run the tests for us, for most scenarios the 13700k is faster than the 7700x while consuming the same power, meaning it's both faster and more efficient in their entire mt suite.

People just don't want to admit it but in most segments intel is leading in efficiency either out of the box (non k and t) or at iso efficiency.

I've never talked about "ISO" power.

You can see a T CPU in the chart well above, but still under AMD. So...

Regards.

 

[TheHerald]

I've never talked about "ISO" power.

You can see a T CPU in the chart well above, but still under AMD. So...

Regards.

I can't see any T cpu in the chart. I only see a power limited 14700 which is on top of the efficiency chart, only losing to the 4999 euro threadripper part. Every desktop amd part is below it.

If you never talked about iso power then what exactly is your point? That amds power limited chips are more efficient when compared to Intel's power unlimited chips? Sure, I agree with that. But why would you compare the two? Either restrict power on both or compare vs the non k and t chips. If you do either, intel wins.

 

[-Fran-]

I can't see any T cpu in the chart. I only see a power limited 14700 which is on top of the efficiency chart, only losing to the 4999 euro threadripper part. Every desktop amd part is below it.

If you never talked about iso power then what exactly is your point? That amds power limited chips are more efficient when compared to Intel's power unlimited chips? Sure, I agree with that. But why would you compare the two? Either restrict power on both or compare vs the non k and t chips. If you do either, intel wins.

Ah, I misread the name of the CPU; apologies.

And no, I don't see any Intel CPU at the top beating a desktop AMD part:
https://gamersnexus.net/u/styles/la...ages/vlcsnap-2023-12-27-13h49m14s208.jpg.webp

The 13600K is at the middle and it is the highest in the chart. I'm sure T variants should be higher in the list, or at least should be.

And the subsequent power efficiency charts reflect that one closely. (edited)

Regards.