Dr. DroI wouldn't be so quick to judge, but there is little rush to adopt this. I will happily wait for the 52C version, the Z990 Apex, and hopefully some improvements on RAM pricing. By then their quirks and what they can do will be well documented, and life will be much easier.

unless there is a pop, there is likely to be little improvement through 2030, and even then... might only happen with regulatory investigation into price fixing.

Hecate91I care because electricity isn't getting any cheaper, anyone who pays their own power bill will usually care about a CPU chugging down 300W, a more power hungry CPU also requires additional cooling, a high end motherboard, and a larger case.

14700KF, Peerless Assassin 120, 250W TDP, B760 board, Corsair 3000D Mid-Tower case, no issue with thermals. If I can do it with Intel, you can do it with AMD, right? Right??

Everyone's talking like like an Intel CPU consumes 300W 24/7 when it's on... Unless you work with your PC, you'll never use PL2 consistently.
Funnily enough, it seems like AMD's Ryzen 9 don't consume 225W 24/7 for whatever reason when this argument is brought up.

There is a point after which performance stops being important.

I mean, such CPU should never be a desktop part, just like RTX5090 should not be desktop part also.

We quickly normalized having 1 kW PSU in desktop PC...

It is crazy.

Sol_BadguySo this overhyped POS is going to have a power draw similar to the 14900K instead of the 270K/285K???
:nutkick:

lol alittle harsh

Wait for it to launch and 3rd Party reviews.

Hecate91I care because electricity isn't getting any cheaper, anyone who pays their own power bill will usually care about a CPU chugging down 300W, a more power hungry CPU also requires additional cooling, a high end motherboard, and a larger case.

Do you run your system at maximum load all the time then? I definitely don't, but if you do and you need what high end processors can bring and don't like the power consumption that's what power limits are for. Not often running my system at full load is why ARL was much more attractive to me than RPL ever was: at the heaviest workload end everything uses a lot of power, but for the in between ARL is much better and is similar to AMD's behavior.

It also doesn't require additional cooling, high end motherboard or a larger case. I used a $44 dual tower air cooler with my 265K set to 295W and never had a problem. On the motherboard side pretty much any Z series ATX board above bottom of the barrel can easily handle that power consumption. I'm not sure where the "larger case" misconception comes from as larger doesn't mean better. One just needs a case with good airflow and something as cheap and relatively compact as the Lian-Li LANCOOL 207 fits that bill.

Hecate91Also it seems like people forgot that even high end hardware used to not be extremely power hungry, there was OC room if you wanted your PC to double as a room heater.

Said hardware had a low performance ceiling which isn't the case anymore. My overclocked 6900K didn't use more than 160W while my 270K Plus can use over 300W yet my power bill didn't go up nor did my room get hotter. Turns out it depends on how efficient your chip is more than the peak power consumption.

Mmmm, very toasty. It makes sense though, I was not really expecting these chips to be low power draw. Considering the supposed boost in clock, bigger cores, and a 50% in cores, Zen 6 desktop will likely be around 300-320W on their flagship. Will be interesting to see the power draw comparisons for next-gen. I suspect only the 44-core Nova Lake would start to match or beat a Zen 6 24-core however. Could be wrong, the IPC gain for the E-cores for Nova Lake is supposedly even better than Raptor Lake to Arrow Lake.

Hecate91I care because electricity isn't getting any cheaper, anyone who pays their own power bill will usually care about a CPU chugging down 300W, a more power hungry CPU also requires additional cooling, a high end motherboard, and a larger case.
Also it seems like people forgot that even high end hardware used to not be extremely power hungry, there was OC room if you wanted your PC to double as a room heater.

What modern desktop cpu takes 300w all the time? Funny how all the TPU reviews show gaming power consumption well below that. Can you let us all know which cpu spends more time at 300w than any other power limit? My 285k spends most time around 80w while gaming with spikes to 120w here and there. But I'm sure you will dismiss that because someone who runs a 5090 doesn't care about power consumption right? While playing around with my 5060ti it is around 150ish w. So worst case with that setup im just under 300w total system power......thats terrible power consumption right, but somehow you say cpus are taking 300w all by themselves?

Well genius yes sometimes when I cpu is under 100% load it may take that amount of power. But I'll ask you in gaming how many cpus are pegged at 100% usage?

bLLC will help keep cores fed with data, so there's less time for it to ramp down to save power, just like X3Ds. As usual though, the firmware will manage heat and current progressively in scenarios where PL2 is reached. The 300W limit seems to be Intel pre-empting productivity compute demand where parallel tasks may push consumption high, yet will finish faster than before. How much faster determines its overall efficiency gain vs prior generations.

Let's use a hypothetical compute scenario:

300W to process a heavy workload taking 8.02 minutes vs 200-250W (230W avg) taking 11.7 minutes. That's a sizable 46% improvement in performance at a cost of higher initial power. That's also 144,360J over 481.2s vs 161,460J over 702s. So, the 300W power usage also improved overall power efficiency by nearly 12% even with higher power draw because the task completed faster. You have to treat computation as a process that consumes power over time, not simply focus on a headline power figure.

Gaming hasn't really pushed CPUs hard in quite some time now because multi-threading is difficult when there are innumerable systems running within the game, many of which have data dependencies and/or require a series of actions to occur (serial processes). It's difficult to get full parallel compute from games on anything more than the GPU.

With all due respect... who cares how much power a desktop PC uses? Why is it relevant?

More power = more heat.

More heat = more problems.

And more expensive components.

And less reliability.

So, it is not about power per-se (although with PCs going to 1 kW it starts to be about that, too), it is about everything else such increase in power consumption causes...

In idle, the wall socket power of amd is about 20w higher than intel, which is a platform power difference that holds true for all the time the PC is running.
When gaming, the cpu package power of 270k plus is 35w higher than 9800x3d. So subtracting the higher amd platform power from the wall socket power, 270k plus is 15w higher than 9800x3d in gaming.
And the wall socket power of 250k plus/265k during gaming should be even lower than 9800x3d.
For most people, idle or low-to-medium load time is longer than gaming time, so intel cpu roughly has an average power saving advantage of 15w compared to amd cpu. According to AI, based on 3 years, 300 days per year, and 12 hours per day, it comes out to $28 or €47.

tsurugiWhen gaming, the cpu package power of 270k plus is 35w higher than 9800x3d. ...
For most people, idle or low-to-medium load time is longer than gaming time, so intel cpu roughly has an average power saving advantage of 15w compared to amd cpu. According to AI, based on 3 years, 300 days per year, and 12 hours per day, it comes out to $28 or €47.

Keep in mind that you are comparing a 24 core CPU - one of the most powerful CPUs for the desktop today, with a wimpy 8 core CPU, this does not seem to be a fair comparison at all, an 8 core CPU in terms of power consumption has a huge advantage.

kondaminHere running a computer like that 4hours a day 320day in a year would cost about 230Euro's if all your power comes from the grid.

If you can keep the CPU running at 300w for 4 hours every day, you have likely already made money with it, or put its computing power into an interest that is meaningful to you personally; either way, it is cheap.

BoggledBeagleKeep in mind that you are comparing a 24 core CPU - one of the most powerful CPUs for the desktop today, with a wimpy 8 core CPU, this does not seem to be a fair comparison at all, an 8 core CPU in terms of power consumption has a huge advantage.

Yes, being able to deliver sufficiently powerful performance when needed is also a very important capability, the key is that 24-core can do it, while 8-core cannot.

Cliff Bungalowhe Nova Lake CPU tile is larger, and it will run hotter, draw more power, and cost a lot more.

IMO its gonna be the opposite. Larger surface area for better heat transfer. Since intel moved the power delivery to the backside it should be easier to cool as the heat would be on the top of the die.

LOL, just another Intel Toaster. I don't believe it will be better than Zen6.
- Because of Zen 10-15% IPC
- Clocks will be ridiculoud High
- I am not even counting x3D

kazuvikingSince intel moved the power delivery to the backside

NVL-S is using N2P which means no BSPDN.

thestryker6NVL-S is using N2P which means no BSPDN.

18A for the bLLC from the leaked CPU-Z screenshot.

rodrigorrasLOL, just another Intel Toaster. I don't believe it will be better than Zen6.
- Because of Zen 10-15% IPC
- Clocks will be ridiculoud High
- I am not even counting x3D

I find it funny you think intel is a "Toaster" when my 9800X3D runs warmer than my Arrow Lake cpus......low quality opinion with zero facts

Okay since this news is about the "single tile flagship" so to speak I thought about an earlier estimation of mine.
Again compared to the 270K/285K this "single tile flagship" will be like the 9850X3D over the 7700X and with something extra. Something which I didn't explore earlier but I will now.
Frequency.

There are some frequency differences between the 270K and 285K which give each some pros and cons depending on the scenario, but for the sake of simplicity I'll just take the 285K as a reference.
tpucdn.com/review/intel-core-ultra-9-285k/images/boost-clock-analysis.png
As I analyzed earlier the 9850X3D is pretty much on par frequency wise with the 7700X across the thread range, so the perf gains come from the newer arch and the 3D-V Cache.
Max power is only slightly increased from 140 W to 150 W.
The same will be for the NL "single tile flagship" over the 285K, but with a higher frequency as well.

The past has the answer.
tpucdn.com/review/intel-core-i9-14900k/images/boost-clock-analysis.png
tpucdn.com/review/intel-core-i9-14900ks/images/boost-clock-analysis.png
200 extra MHz for P-cores and 100 extra MHz for E-cores translate into:
www.techpowerup.com/review/intel-core-i9-14900ks/26.html
3% more app perf and 1.3% more gaming perf at 720p.

Okay but will the NL chip have only that much over the 285K? I don't know but if I were to guess I'd say this is the minimum frequency increase, it's likely more. Let's say it's double or even a bit more (somewhat higher than 6 but under 6.5 GHz), to the point where the previous percentages are doubled, so 6% higher in apps and 2.6% in gaming.

tpucdn.com/review/amd-ryzen-7-5800x3d-10th-anniversary/images/relative-performance-cpu.png
tpucdn.com/review/amd-ryzen-7-5800x3d-10th-anniversary/images/relative-performance-games-1280-720.png
Okay the 9850X3D over 7700X example (aka NL chip over 285K) would give 18% more app perf and 29.5% more gaming perf.
And with the doubled frequency gain percentages above we would have 125.1% higher app perf and 32.9% higher gaming perf.
At this point we see that with some hot-rodding (which is emphasized by the news subject) the NL chip can realistically get to 25% more app perf and 33% more gaming perf than the 285K.

How does this compare with Zen5 X3D and Zen6 X3D?
In gaming at 720p the 9850X3D is 22.4% better than the 285K which is almost the same to the 270K.

Thus the NL chip can realistically at the very least be on par with the 9850X3D if not clearly surpass it by 8.6% let's say.

Previously, some people including me have expressed some doubt over how effective the bLLC will be for Nova Lake, something along the lines of "can Intel get it right from the first try?".
Looking at how the 5800X3D currently holds up against newer non-X3D CPUs in gaming the answer is a resounding yes.
There is nothing stopping Intel from implementing this properly. Actually if AMD got it right (performance wise but not position wise) from the first try, Intel could potentially do even better as they had the time to learn from AMD's experiences.

What about the "10800X3D"?
Well if we look at the 9850X3D vs 7800X3D we have 12.8% more gaming perf by means of newer arch (and smaller node), higher power (150 vs 80 W) and higher frequency (600 MHz).
Now with the Zen5 to Zen6 some have stated that the bigger node jump than from Zen4 to Zen5 will give Zen6 a bigger gain over Zen5 than previously.
Really?
Well some of the extra frequency from Zen4 to Zen4 was possible (for X3D chips) due to positioning the 3D-V Cache underneath the CCDs as opposed to above like before.
So now AMD has already used up this trick meaning the extra frequency will come solely from the node jump.
Given that the node jump is bigger it offsets the lack of cache repositioning thus the frequency increase should be about the same as with Zen4 to Zen5.
Meaning the frequency increase gain won't be significantly higher either.

When it comes to non-X3D we could use the 9700X vs 7700 as those have the same power ceiling, so any gain comes from newer arch and frequency increase.
tpucdn.com/review/amd-ryzen-7-7700-non-x/images/boost-clock-analysis.png
tpucdn.com/review/amd-ryzen-7-9700x/images/boost-clock-analysis.png
At the start of the thread range there is a 170 MHz advantage for the 9700X but at the end of the range the frequency is almost identical (avg between the 3 tests).
So on average sort of 100 MHz more for the 9700X let's say.
That translates to 7.67% better in apps and 5.4% better in 720p gaming:
tpucdn.com/review/amd-ryzen-7-9800x3d/images/relative-performance-cpu.png
tpucdn.com/review/amd-ryzen-7-9800x3d/images/relative-performance-games-1280-720.png
For the old data, just to check.
tpucdn.com/review/amd-ryzen-7-5800x3d-10th-anniversary/images/relative-performance-cpu.png
tpucdn.com/review/amd-ryzen-7-5800x3d-10th-anniversary/images/relative-performance-games-1280-720.png
And 8.77% better in apps and 7.24% better in 720p gaming for the new data.
Let's assume we increase the frequency further until we double the percentages.
After all Zen6 is rumored to have very high frequency.

Another aspect is core count increase. This is one area for which Zen4 to Zen5 provided 0 gains.
But now there will be extra cores on the CCD which for some games might make a difference.

The power aspect can be ignored, I don't think that the 12 core "10800X3D" will go higher than the 6+6 core 9900X(3D) maybe a bit but inconsequential to gaming. The 9850X3D is already close to the 9900X(3D) and is heavily hot-rodded (~150 W max vs ~170 W max).

Going back to the 12.8% gain of the 9850X3D over the 7800X3D, if the "10800X3D" is another 12.8% over the 9850X3D then it would be 38.1% higher than the 285K which would make it 3.9% better than the NL "single tile flagship" in gaming at 720p.

When it comes to app perfomance I've touched on that before.
The 9900X3D is the stand-in for a "9900X 12 core single CCD".
It would be at 93% in the relative performance chart if the 270K would be at 100% and the 285K at 98.7%.
But we also need a "9900X3D 12 core single CCD" so then the "10800X3D" has newer arch and more frequency over it, for the same power ceiling.
Using the gain from 9950X to 9950X3D this would be at 96% in the performance chart.

The "10800X3D" over the "9900X 12 core single CCD" would be pretty much like the 9850X3D over the 7700X -> newer arch (smaller node), same frequency and power but with 3D-V Cache.
So the "10800X3D" would sit at 109.7% (18% better than the 93% 9900X3D) in the app perf chart compared to the NL chip at 123.5% (25.1% better than the 98.7% 285K) -> 12.6% better than the "10800X3D".

The "10800X3D" over the "9900X3D 12 core single CCD" would be pretty much like the 9700X (with doubled percentages) over the 7700 -> newer arch (smaller node), more frequency (at least 200 MHz) and same power.
So the "10800X3D" would sit at 112.8% (17.5% better than the 96% "9900X3D2") in the app perf chart compared to the NL chip at 123.5% (25.1% better than the 98.7% 285K) -> 9.5% better than the "10800X3D".

The "10800X3D" over the "9900X3D 12 core single CCD" would be pretty much like the 9800X3D over the 7800X3D -> newer arch (smaller node), more frequency (300 MHz), more power (150 vs 80 W).
So the "10800X3D" would sit at 115.6% (20.4% better than the 96% "9900X3D2") in the app perf chart compared to the NL chip at 123.5% (25.1% better than the 98.7% 285K) -> 6.8% better than the "10800X3D".

In conclusion the "10800X3D" could be 3.9% better in gaming at 720p compared to the NL "single tile flagship" while the NL chip could range from 6.8-9.5 to 12.6% better in apps (let's average those and say ~9.6%).
However max power draw could be under 190 W for the "10800X3D" while apparently it will be over 260 W for the NL chip.

Could this be realistic or is it too optimistic for one and too pessimistic for the other?

kazuviking18A for the bLLC from the leaked CPU-Z screenshot.

There's no way to poll what node is being used so it's ripe for human error. Early ARL showed 20A which we know was never used because the node was canceled.

Sol_BadguyOkay since this news is about the "single tile flagship" so to speak I thought about an earlier estimation of mine.
Again compared to the 270K/285K this "single tile flagship" will be like the 9850X3D over the 7700X and with something extra. Something which I didn't explore earlier but I will now.
Frequency.

Why are you comparing "single tile flagships", when Intel does not have any double tile CPU, while AMD does, and Intel tile has 24 cores and AMD tile has 8? Does your comparison make any sense at all?

BoggledBeagleWhy are you comparing "single tile flagships", when Intel does not have any double tile CPU, while AMD does, and Intel tile has 24 cores and AMD tile has 8? Does your comparison make any sense at all?

Are you OK bro?
Yes my comparison makes sense for those who are capable of understanding my words.
Comparatively your post doesn't make any sense.

You ask me why do I compare single tile flagships when Intel doesn't (yet) have dual tile ones?
That's why because Intel only has single tile ones.
Who would've guessed... :rolleyes:

If you want me to ELI5 it to you fine.

thestryker6There's no way to poll what node is being used so it's ripe for human error. Early ARL showed 20A which we know was never used because the node was canceled.

Intel 100% tried their nodes on NVL. My prediction is TSMC ran out of production for the N2 node and intel just decided fuck it we ball and use 18A for the other half. Or the bLLC is 18A while the normal tiles are 2N. Or IO made with 18A, cores made with 2N and bLLC is made with 18A.

kazuvikingSince intel moved the power delivery to the backside it should be easier to cool as the heat would be on the top of the die.

The HUB tile in Novalake uses 18a-p, which uses BSPD, while the compute tile uses n2p. For the BSPD, traditional chips are powered from the top side where the heatsink is located, with power rails traveling from the substrate through the entire chip to reach the transistors at the top of the cpu.
In contrast, BSPD delivers power directly to the transistors from the bottom of the substrate. Therefore, in terms of chip thermal dissipation efficiency, the traditional method undoubtedly holds the advantage. The BSPD method leaves the transistors in the middle of the chip, which is disadvantageous for heat dissipation. BSPD reduce IR Drop that offset this negative impact to some extent.

kazuvikingIntel 100% tried their nodes on NVL. My prediction is TSMC ran out of production for the N2 node and intel just decided fuck it we ball and use 18A for the other half. Or the bLLC is 18A while the normal tiles are 2N. Or IO made with 18A, cores made with 2N and bLLC is made with 18A.

There might be a Razorlake 8+0+4 bllc(Novalake uarch) built on 18A-P.

zusBut Intel won't have the node advantage this time, so I don't know how things will play out for Nova Lake vs Zen 6.

This. It is node parity this time that likely forces them to increase power budget.

Even if we pessimistically assume that 24 core Zen6 SKU has the same maximum power of ~300W, this would be a power budget for 24 big cores with SMT, whereas in Intel's case, there will be only 8 big cores and 20 small cores screaming 'moar powaaa'! Big cores have been bloated for years. Switching off SMT/HT and AVX512 were some of ways to curtail the power budget. It is not surprising that they are moving in a direction of a unified core in a few years, as e core team seems to be bringing more promising results.

Let's take an example from AMD's playbook of new Zen6 EPYC. The 9686F SKU with 96 cores (8 CCDs with 12 big cores each) features a power budget of 500W. It reaches 5 GHz and it features two large IODs. If we assume that each IOD consumes ~50W when fully loaded, there is ~400W left for eight CCDs, roughly 50W per CCD that can run at 5 GHz.
www.amd.com/en/products/processors/server/epyc/9006-series/amd-epyc-9686f.html

On desktop, they will push the same CCD well beyond 6 GHz. Do they need to bring extra 100W to do that? Most likely not. I'd assume that '10800X3D', or whatever the name, would still be ~120W CPU and dual CCD flagship ~200W. We will see in a few months.

zusBut Intel won't have the node advantage this time, so I don't know how things will play out for Nova Lake vs Zen 6.

I expect Nova lake to be leading (massively actually) in Multithreaded efficiency at every power level as they do today. In fact, I expect Arrow lake to be leading Zen 6. There is no way a 10800x (or whatever it's named) 12 core CPU to be faster at similar power to the 270k. And that's even assuming it's at least 300$

I assume AMD is facing a severe power draw issue the last few generations which is why they are stuck with 6 and 8 core chips in thier mainstream CPUs, and even their high end are at 16. Let's hope Zen 6 fixes a bit of the power inefficiencies.