News Intel's latest lower-powered CPUs give Ryzen rivals a run for their money — Core i9-14900T beats Ryzen 9 7900 in Geekbench 6 benchmark

[TerryLaze]

Again, you're confusing or conflating Zen 2 and Zen 4. Also, more whataboutism.

Sure dude, don't let facts get in your way.

When ZEN 4 is forced to run at low clocks, due to power limits (or cooling restrains) ,it loses efficiency because any clock below 2.3Ghz still needs the same amount of voltage, we all know how undervolting can dramatically decrease power draw/cooling requirements.

Enter zen c , just like the e-cores they are worse at normal clock levels but better at very low power.
https://zhuanlan.zhihu.com/p/653961282

Looking at the VID tables of the two different Zen 4 cores, we can see that in most cases the voltage of the dense core is higher. The Classic core reaches Vmin at around 2.3GHz, while the Dense core does not reach Vmin until below 1.5GHz. The two V/F curves overlap around 1.5 GHz.

Specific to power consumption, it can be seen that the dense core has certain advantages in energy efficiency below 2GHz, and above 2GHz, the classic core is better. However, in the context of high basic power consumption of the package, this advantage is not obvious. Comparing with the voltage VID table, we can see that in the frequency range of 1.5-2GHz, the dense core can achieve better power consumption performance even when the voltage is slightly higher, indicating that the classic core has a larger current under the same voltage and load. , to some extent, it reflects the price Zen4 classic pays to hit the high frequency >5GHz.

As for the ComputeBase data, I've said all I have to say about that. I think this plot speaks volumes:

7950x | 7900x | 7700x | 7600x all are below the 13900k at 45W.
With the 7950x , that I chose to compare with, being closest to the 13900k ...
Whataboutwhatism?!

 

[bit_user]

Enter zen c , just like the e-cores they are worse at normal clock levels but better at very low power.

Off-topic, though.

7950x | 7900x | 7700x | 7600x all are below the 13900k at 45W.

Irrelevant. AMD doesn't sell a version of them designed to run at or below 45 W.

We ended up where we started:

• When it's boosting, the "35 W" i9-14900T will burn 20.5% more power than a AMD R9 7900.
• At its sustained speed, the i9-14900T will perform worse than the R9 7900.
• Only while it's boosting, will the i9-14900T actually perform better than the R9 7900.

 

[TerryLaze]

Off-topic, though.

Not really though, since we are talking about low power CPUs it's very relevant.

Irrelevant. AMD doesn't sell a version of them designed to run at or below 45 W.

LOL, are you really trying to sell yourself on that the 13900K is designed for 45W ?

We ended up where we started:

• When it's boosting, the "35 W" i9-14900T will burn 20.5% more power than a AMD R9 7900.
• At its sustained speed, the i9-14900T will perform worse than the R9 7900.
• Only while it's boosting, will the i9-14900T actually perform better than the R9 7900.

Based on you closing your eyes real hard and believing?!
The 14900t having 20% more boost overhead when available is good thing, it means that it can be faster if the user needs it to be.
At 45W and below every ZEN cpu loses to the 13900k so why do think that the 7900 is going to be the big exception?

 

[bit_user]

The 14900t having 20% more boost overhead when available is good thing, it means that it can be faster if the user needs it to be.

You have to think about the market for 35 W CPUs. A lot are probably going into mini-PCs, where the noise of a fan spinning up to dissipate that 106 W of heat might be rather unwelcome.

If you read reviews of mini-PCs, reviewers often enough complain about fan noise from models dissipating only 25-40 W. So, the thought of dissipating 105 W of heat, from such a form factor, seems pretty unappealing to me.

At 45W and below every ZEN cpu loses to the 13900k so why do think that the 7900 is going to be the big exception?

That's not what I said. I said that i9-14900T at 35 W will be slower than R9 7900 at 65 W. That's what the article compared, so that's the matchup I'm examining.

 

[TerryLaze]

You have to think about the market for 35 W CPUs. A lot are probably going into mini-PCs, where the noise of a fan spinning up to dissipate that 106 W of heat might be rather unwelcome.

If you read reviews of mini-PCs, reviewers often enough complain about fan noise from models dissipating only 25-40 W. So, the thought of dissipating 105 W of heat, from such a form factor, seems pretty unappealing to me.

Yes so at that 25-40W the 14900t is still going to give you the best possible performance, so what was your issue in the first place?

That's not what I said. I said that i9-14900T at 35 W will be slower than R9 7900 at 65 W. That's what the article compared, so that's the matchup I'm examining.

And you had to make a whole post about this in the first place?
That a CPU at 35W is going to be slower than a CPU at 65W (88 actual) ?

 

[bit_user]

Yes so at that 25-40W the 14900t is still going to give you the best possible performance, so what was your issue in the first place?

But:

1. its fans will probably make quite a racket, during the 106 W turbo boost.
2. once it drops back to 35 W, the impact on performance will be substantial.

And you had to make a whole post about this in the first place?

Yes, because people get lured in by that 35 W number, which is under 1/3rd of its boost power. I'll bet most people wouldn't expect it to boost so high. A few years ago, I sure wouldn't have expected a 35 W CPU to boost to 106 W.

That a CPU at 35W is going to be slower than a CPU at 65W (88 actual) ?

Actually, what the data shows is that at 35 W it's going to be slower than a R9 7900, even when the latter is running only at 65 W.

 

[TerryLaze]

But:

1. its fans will probably make quite a racket, during the 106 W turbo boost.
2. once it drops back to 35 W, the impact on performance will be substantial.

Yes, because people get lured in by that 35 W number, which is under 1/3rd of its boost power. I'll bet most people wouldn't expect it to boost so high. A few years ago, I sure wouldn't have expected a 35 W CPU to boost to 106 W.


Actually, what the data shows is that at 35 W it's going to be slower than a R9 7900, even when the latter is running only at 65 W.

106W is almost half of what the most basic cooler can deal with, so it's going to run at about 50% speed which usually is pretty quiet, it's less than what consoles use, I bet there are even passive coolers that can deal with 100W
https://www.tomshardware.com/reviews/amazon-basics-cpu-cooler-review/2
Why would it drop down to 35W if the cooling is there for 106W?

Actually, what the data shows is that at 35 W it's going to be slower than a R9 7900, even when the latter is running only at 65 W.

That's what I said. That's what anybody would expect.

 

[bit_user]

106W is almost half of what the most basic cooler can deal with,

Not in a small form factor case, which is precisely where people are typically using 35 W TDP CPUs. Those also will have smaller fans, which have to run faster. Smaller fans naturally emit higher-pitched sound, which tends to be more noticeable & annoying.

 

[TerryLaze]

Not in a small form factor case, which is precisely where people are typically using 35 W TDP CPUs. Those also will have smaller fans, which have to run faster. Smaller fans naturally emit higher-pitched sound, which tends to be more noticeable & annoying.

Could you please finally decide if you are talking about a 35W or a 106W system?

 

[Order 66]

Furthermore, the default Tau for these CPUs is 28 seconds. So, that's how long it could boost at 105 W.

Couldn't you use XTU to increase that to unlimited, assuming you had a decent cooler, anyway?

 

[bit_user]

Could you please finally decide if you are talking about a 35W or a 106W system?

Unfortunately, if it's a i9-14900T, then you're stuck building for 106 W of peak dissipation (or further limiting its performance, by lowering the boost/PL2).

I think the ones who couldn't decide were Intel, themselves.

 

[bit_user]

Couldn't you use XTU to increase that to unlimited, assuming you had a decent cooler, anyway?

My understanding is that the max Tau is baked into the hardware. According to this datasheet, the MSR (Model Specific Register) has a max value of 448 seconds. See page 102, in the row labeled "S Refresh-Processor Line LGA", "8P+16E Core 35W", "Power Limit 1 Time (PL1 Tau)", under the column "Tau MSR Max Value"

https://www.intel.com/content/www/u...ation-processors-datasheet-volume-1-of-2.html
​

So, assuming nothing in the system prevents you from setting it to the MSR max value (I'm not sure if manufacturers can block this, somehow), you'd get no more than about 7.5 minutes at 106 W, before it cuts you down to 35 W.

If we use the ComputerBase data from the i9-13900K, the estimated performance at 106 W would be a score of 74, while the estimated performance at 35 W would be 42 (using linear interpolation & extrapolation, respectively). So, you'd get about 56.8% of the performance after Tau expires.

For reference, note that the i9-13900K scores 79, when limited to 125 W, and 95, when limited to 253 W.

 

[TerryLaze]

Unfortunately, if it's a i9-14900T, then you're stuck building for 106 W of peak dissipation (or further limiting its performance, by lowering the boost/PL2).

I think the ones who couldn't decide were Intel, themselves.

Then why are you constantly talking about SFF and 35w tdp cpus?
Why all the talk about the CPU losing performance then?

My understanding is that the max Tau is baked into the hardware. According to this datasheet, the MSR (Model Specific Register) has a max value of 448 seconds. See page 102, in the row labeled "S Refresh-Processor Line LGA", "8P+16E Core 35W", "Power Limit 1 Time (PL1 Tau)", under the column "Tau MSR Max Value"

​

https://www.intel.com/content/www/u...ation-processors-datasheet-volume-1-of-2.html​

​

So, assuming nothing in the system prevents you from setting it to the MSR max value (I'm not sure if manufacturers can block this, somehow), you'd get no more than about 7.5 minutes at 106 W, before it cuts you down to 35 W.

If we use the ComputerBase data from the i9-13900K, the estimated performance at 106 W would be a score of 74, while the estimated performance at 35 W would be 42 (using linear interpolation & extrapolation, respectively). So, you'd get about 56.8% of the performance after Tau expires.

For reference, note that the i9-13900K scores 79, when limited to 125 W, and 95, when limited to 253 W.

Look at it, look at it closely, you even typed it yourself, PL1 that's the low power draw, that's the amount of time turbo should stay at low power,lower than actual PL1 to allow the system to cool off, after boosting to refill the turbo allowance, if it ever runs out of turbo allowance that is.
There is no TAU for the 106W PL2 setting, if you have a happy system that can cool 106W all day long then you are set,

 

[bit_user]

Look at it, look at it closely, you even typed it yourself, PL1 that's the low power draw,

You're misinterpreting it, based on the name, instead of actually reading how it's specified. The document states quite clearly what that setting does:

"PL2- SoC opportunistic higher Average Power with limited duration controlled by Tau_PL1 setting.
the larger the Tau, the longer the PL2 duration."
​

Page 87, note 13.

 

[thestryker]

These have always been fairly interesting parts and they used to be better binned though I don't know that would still be the case. Given how the power controls work these days I think I'd still be looking at a K SKU and power limiting that if putting the system together myself.

The only chance you have to disable it is on unlocked, K-series CPUs. On these T-series, the most you can do is to reduce it from the default of 28 seconds (which is counter-productive, from a performance perspective).

I've never played around with the T SKUs does Intel lock Tau on them? You can definitely modify the time/set unlimited on the standard CPUs (ex 13900). I mean removing it would be counterproductive and make buying the part pointless, but that doesn't mean you can't do it!

My understanding is that the max Tau is baked into the hardware. According to this datasheet, the MSR (Model Specific Register) has a max value of 448 seconds. See page 102, in the row labeled "S Refresh-Processor Line LGA", "8P+16E Core 35W", "Power Limit 1 Time (PL1 Tau)", under the column "Tau MSR Max Value"

I believe that's the maximum value you can set it to, but doesn't have anything to do with whether or not you can disable it (the same figure exists in every desktop part).

 

[TJ Hooker]

Regarding extending/disabling turbo duration limit (tau) on locked CPUs:

"For example, if you install the 12700 on any Z690 motherboard with the exception of entry-level models from Asrock, it will run in the PL2 state indefinitely, despite the fact that it's a locked part. This can also happen on some B660, H670 and H610 boards. For example, the MSI B660M Mortar WiFi DDR4 runs without power limits by default."

Intel Core i7-12700 + Intel B660 Review

Today we have a combo review of all-new Intel hardware, our first look at the affordable Intel B660 platform, the non-K Core i7-12700 and Intel's latest box...

www.techspot.com

So, on paper, non-K chips still have the PL1, PL2, and tau limits we're familiar with (in contrast to unlocked chips, where Intel made PL1=PL2 the new standard). But mobo makers are free to do what they want with those values.

 

[cyrusfox]

These have always been fairly interesting parts and they used to be better binned though I don't know that would still be the case. Given how the power controls work these days I think I'd still be looking at a K SKU and power limiting that if putting the system together myself.

I've never played around with the T SKUs does Intel lock Tau on them? You can definitely modify the time/set unlimited on the standard CPUs (ex 13900). I mean removing it would be counterproductive and make buying the part pointless, but that doesn't mean you can't do it!

I agree, I see no reason to obtain a T part as you can set the power limits in the Bios for any chip (do not need an unlocked chip). As there doesn't appear to be any binning halo, I would recommend getting a normal 14900(or 13900 honestly), and then set the power limit for your application.

The main limitation to the T parts is it will throttle more aggressively and thus perform worse compared to a non-t part due to its tuning. This power envelope only makes sense to me in 1L cases/embedded custom applications. Even there, they usually make a custom 65W capable configuration, and people would be better served by using a standard chip rather than the T parts.

Since all parts can be power gated, I no longer see a segment for T parts, OEM's must still be requesting them...

 

[thestryker]

As there doesn't appear to be any binning halo, I would recommend getting a normal 14900(or 13900 honestly), and then set the power limit for your application.

Locked SA voltage is the reason I personally would go with the K SKU as DDR5 6400 is the only guarantee without being able to modify it.

 

[cyrusfox]

Locked SA voltage is the reason I personally would go with the K SKU as DDR5 6400 is the only guarantee without being able to modify it.

I still am using 32Gb DDR4 3200MHz modules(2x or 4x depending on available slots) on all my Raptor boxes...

Isn't memory performance (Going from one speed to the next) the lowest ROI of any upgrade one can do on their system? Perhaps their are workload dependencies that prove my statement false. For my workload, which is largely creator workload (edit, encode) from what I recall there was negligible benefit going up to DDR5. Perhaps DDR5 prices are more comparable to DDR4 and I am missing out in huge performance improvements. Love to understand this better.

 

[thestryker]

I still am using 32Gb DDR4 3200MHz modules(2x or 4x depending on available slots) on all my Raptor boxes...

Isn't memory performance (Going from one speed to the next) the lowest ROI of any upgrade one can do on their system? Perhaps their are workload dependencies that prove my statement false. For my workload, which is largely creator workload (edit, encode) from what I recall there was negligible benefit going up to DDR5. Perhaps DDR5 prices are more comparable to DDR4 and I am missing out in huge performance improvements. Love to understand this better.

It's extremely workload dependent and unfortunately there hasn't been a good general memory scaling investigation since Ian left AnandTech after doing ADL (his tests were limited to 3200 v 4800). If your workload doesn't need memory bandwidth then it's a wash, because the real world latency between the two is going to be about the same (DDR5 6400 CL32 = DDR4 3200 CL16 latency for example).

Steve from HUB did a gaming overview in December which is a pretty good indicator of where things stand: https://www.techspot.com/review/2777-ddr5-vs-ddr4-gaming/

While it doesn't directly compare to any specific workload aside from gaming it does show the same sort of behavior: sometimes it matters and sometimes it doesn't. Building new though there's no advantage to going with DDR4 over DDR5 unless you know for sure that no workload you have that will ever need additional memory bandwidth and even then the savings will be low.

 

[bit_user]

Regarding extending/disabling turbo duration limit (tau) on locked CPUs:

The key question is to know whether that also applies to the T-series CPUs. With such a low TDP, they might be "special", in this regard.

 

[bit_user]

For my workload, which is largely creator workload (edit, encode) from what I recall there was negligible benefit going up to DDR5.

These benchmarks seem directly relevant to your question:

Impact of DDR5 Speed on Content Creation Performance (2023 update)

DDR5 memory has kits rated for up to 8400 Mbps, while desktop CPUs only officially support up to 5600 Mbps. How much does running at official specifications actually impact performance in common content creation applications?

www.pugetsystems.com

Of course, it also depends on your CPU. With fewer cores or on 65 W CPU variants, memory should be less of a bottleneck.

 

[TerryLaze]

You're misinterpreting it, based on the name, instead of actually reading how it's specified. The document states quite clearly what that setting does:

​

"PL2- SoC opportunistic higher Average Power with limited duration controlled by Tau_PL1 setting.​

the larger the Tau, the longer the PL2 duration."​

​

Page 87, note 13.

You're misinterpreting it based on the notes instead of looking at the actual numbers, the PL1 TAU is set to 28W on a CPU with 35W base power.

It's like I said, PL1 tau goes below base power to replenish the turbo allowance, so yes the longer you save up turbo allowance the longer you can boost afterwards, BUT THIS PERQUISITES THERE BEING ANY ISSUE WITH NOT HAVING ENOUGH TURBO ALLOWANCE. If it never needs to go to pl1 tau because it never runs out of turbo allowance then it does diddly squat.

 

[bit_user]

You're misinterpreting it based on the notes instead of looking at the actual numbers,

LOL. It's a hardware reference. You don't just look at numbers and guess what they mean!

the PL1 TAU is set to 28W on a CPU with 35W base power.

Better take another look at that table, because it has a "Units" column which labels the units of PL1 Tau as "S" (seconds).

​

It's like I said, PL1 tau goes below base power to replenish the turbo allowance,

That's not what they said and doesn't align with what you can independently find about the default Tau being 28 seconds, for these CPUs.

https://www.techpowerup.com/cpu-specs/core-i9-13900t.c3001
​

 

[TerryLaze]

LOL. It's a hardware reference. You don't just look at numbers and guess what they mean!


Better take another look at that table, because it has a "Units" column which labels the units of PL1 Tau as "S" (seconds).

​

​

That's not what they said and doesn't align with what you can independently find about the default Tau being 28 seconds, for these CPUs.

​

https://www.techpowerup.com/cpu-specs/core-i9-13900t.c3001​

​

TAU is just the the interval the CPU checks to see what conditions there are, if everything is fine nothing will change.

You are conditioned to think that TAU means that the CPU is forced to reduce clocks/power because that's the only type of benchmarks and reviews we ever get to see.

In case of the PL2 TAU ,if it is ever enabled, this is also often true if the CPU has worked on a heavy task, it will have used up all of the power/cooling and will have to slow down for a while.

Intel Turbo Boost 2.0 Technology - SkatterBencher

Intel Turbo Boost 2.0 Technology allows the processor to run faster than the base frequency when it's below its rated specification limits.

skatterbencher.com

• Tau, in seconds, is the time window for calculating the average power consumption. The CPU will reduce the CPU frequency if the average power consumed is higher than PL1.

Also why are you linking tpu that shows a PL2 TAU number to justify anything about PL1 TAU, you were strictly against that when I showed numbers for different things.
This is power limit 1 you are trying to talk to death.