• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

Intel Core Ultra 9 285K

Bro really went on a technology forum to tell us we don't need new technology.
No I am saying most people do not care much about this either, same way they don't care about PCIe 5.0, or having a gazillion PCIe lanes and high speed ports in general, I'd be amazed if even 0.1% PC users are doing something that necessities 40Gbps over USB.

And what you're saying sounds really stupid if you stop and think, you were saying nobody gives a shit about PCIe 5.0 but that's new technology as well go figure.
 
Last edited:
So you know the 9950X3D will be the best CPU for the next couple of years now! :toast:
Not for what I do. I read these reviews and went straight to the Science and Physics tests, because that is what matters to me most. Gaming is a secondary concern as ANY CPU that performs well with these kinds of tasks will do games VERY well.
In these tests, the 285K kicks the crap out of most of the competition and handily beats the 9950X. The 9950X3D likely won't do much better if at all. We'll see though.
 
Not for what I do. I read these reviews and went straight to the Science and Physics tests, because that is what matters to me most. Gaming is a secondary concern as ANY CPU that performs well with these kinds of tasks will do games VERY well.
In these tests, the 285K kicks the crap out of most of the competition and handily beats the 9950X. The 9950X3D likely won't do much better if at all. We'll see though.
Those NAMD results look weird, compared to the ones from Phoronix:
Screenshot 2024-10-25 at 15.35.06.png


The core ultra does win in some tests, but gets heavily beaten in many others, so it's more of a matter of which specific workloads are important to you.
 
Did you look at all of the rest of those scores? "Heavily beaten" is not what I'm seeing.
Well, I did give you a clear example of it getting heavily beaten in NAMD.
If all you want to do is cherry pick results to ones were the Ultra 9 wins, and they're in fact the ones most relevant to you, then great!
This doesn't change the fact that it still gets beaten in most tests, other wise the geomean wouldn't look like so:
Screenshot 2024-10-25 at 16.21.06.png
 
Well, I did give you a clear example of it getting heavily beaten in NAMD.
If all you want to do is cherry pick results to ones were the Ultra 9 wins, and they're in fact the ones most relevant to you, then great!
This doesn't change the fact that it still gets beaten in most tests, other wise the geomean wouldn't look like so:
View attachment 368912
It doesn't get beaten in the things that I care about. So everything else doesn't matter. Still, It's only topped by 2 other CPU models(depending on RAM speed), and one of them only just. That does not count as "heavily beaten". I'm not going to spend more money on a 9950X that doesn't do as well at what I am looking for.

I was waiting for these CPU models to release before upgrading. The reviews show clearly that this lineup is a mixed bag of results. For what I need, the 285K hits the mark.
 
Last edited:
It doesn't get beaten in the things that I care about. So everything else doesn't matter. Still, It's only topped by 2 other CPU models(depending on RAM speed), and one of them only just. That does not count as "heavily beaten". I'm not going to spend more money on a 9950X that doesn't do as well at what I am looking for.

I was waiting for these CPU model to release before upgrading. The reviews show clearly that this lineup is a mixed bag of results. For what I need, the 285K hits the mark.
How's the pricing where you live? So far I've seen the 9950x going for cheaper.
The CUDIMMs to make the ultra 9 better are also not that cheap.
 
But 2x x16 at full throttle is not possible on Z890 either.
The CPU only has 20 PCIe 5.0 lanes and 4 PCIe 4.0 lanes, total of 24. You could either run an AIC at x16 and two other at x4.
The chipset does have an impressive 24 PCIe 4.0 lanes, but that's coming from an x8 4.0 uplink, so you'll be bottlenecked by that anyway.

You could in theory do x16 out of the CPU + x8 saturating the chipset, but no manufacturer so far has done that. Max you'll see is x8/x8 5.0 on the CPU + x4 4.0 through the chipset, with a ton of NVMe slots available. AMD has a similar offering, but with way fewer NVMe slots.
Don't need full bandwidth, realistically the worst-case is I'm copying from one M.2 drive to another which is only going to consume 4 lanes. Even assuming I do that for four drives (2 read 2 write) the DMI link will handle it. And if all four of my M.2s are sitting in an x16 4.0 slot off the chipset, ideally copying between them will be entirely chipset-local i.e. the data never has to hit the DMI link nor the CPU.
 
Don't need full bandwidth, realistically the worst-case is I'm copying from one M.2 drive to another which is only going to consume 4 lanes. Even assuming I do that for four drives (2 read 2 write) the DMI link will handle it. And if all four of my M.2s are sitting in an x16 4.0 slot off the chipset, ideally copying between them will be entirely chipset-local i.e. the data never has to hit the DMI link nor the CPU.
Then you just don't need an x8 slot whatsoever? From what you said now, all you need are x4 links for NVMes and nothing else.
AMD's x670(e) and x870e can also do tons of downstream links, with x12 4.0 and x8 3.0, and coping data among devices in this link would mean full bandwidth as well since it's not going through the CPU either.

Do you often move vast amounts of data across 4+ drives at once? I'm curious about that use case.
 
Well, I did give you a clear example of it getting heavily beaten in NAMD.
If all you want to do is cherry pick results to ones were the Ultra 9 wins, and they're in fact the ones most relevant to you, then great!
This doesn't change the fact that it still gets beaten in most tests, other wise the geomean wouldn't look like so:
The real problem is that it's doing this with 8 more cores and similar power consumption, what's the point of all of those E cores if it's not more efficient and it's not really faster either than having 16 P cores.
 
The real problem is that it's doing this with 8 more cores and similar power consumption, what's the point of all of those E cores if it's not more efficient and it's not really faster either than having 16 P cores.
I don't think I get your idea.
It has 8 more cores, yes. Performance for stuff where SMT is not relevant ends up matching the 9950x, so we could say in a rought manner that the 8x P-cores match AMD's 8 regular cores, whilst the other 16 E-cores also manage to match another 8 cores from the red side.
We will understand this relationship better once we see tests with only the P-cores, Phoronix should be releasing those in the following weeks, then we can estimate how good are the P-cores alone and how much the E-cores help.

Now, when it comes to power, 24 hybrid cores with the same performance of 16 big cores with similar power consumption does seem like a good thing IMO. Either side would need to cram more cores for higher performance, which would mean higher power as well. I don't see the problem with this.

Also important to remind that E-cores are not focused on being power-efficient, but rather space-efficient. So Intel could have gone with either the current design, or 12P cores instead (since 1 P-core roughly equals a 4x E-core cluster in area), which would likely be worse at multi threading, and maybe use more power.

Gamers would be likely happier, but I don't think Intel is trying to catter solely to gamers.
 
In sum: Arrow Lake costs more and delivers less. Quite literally, including energy consumption.

Intels architecture isn't making TSMCs 3nm process any justice.

And to think that the uninformed will still buy it, just because it says Intel and has "ULTRA" written on the name.
 
Not for what I do. I read these reviews and went straight to the Science and Physics tests, because that is what matters to me most. Gaming is a secondary concern as ANY CPU that performs well with these kinds of tasks will do games VERY well.
In these tests, the 285K kicks the crap out of most of the competition and handily beats the 9950X. The 9950X3D likely won't do much better if at all. We'll see though.
Course you do buddy, tell me some more things that didn't happen :laugh:
 
Also important to remind that E-cores are not focused on being power-efficient, but rather space-efficient. So Intel could have gone with either the current design, or 12P cores instead (since 1 P-core roughly equals a 4x E-core cluster in area), which would likely be worse at multi threading, and maybe use more power.
I don't know if even that's true, you have to keep in mind a lot of those area savings come from much smaller caches, every E core cluster has 4MB (so just 8MB for all E cores) and every P core has 3MB of L2 cache (24MB total). In reality they could probably cram 8 P cores in there instead of 16 E cores with reduced caches and the overall size would be about the same. As it stands I simply cannot see even one advantage this approach has, it's not faster, it's hardly more power efficient and it barely saves die space.

1729894167502.png
 
In sum: Arrow Lake costs more and delivers less. Quite literally, including energy consumption.

Intels architecture isn't making TSMCs 3nm process any justice.

And to think that the uninformed will still buy it, just because it says Intel and has "ULTRA" written on the name.
I always thought there was more to it than node size for how good a product is in terms of power and heat.
 
I always thought there was more to it than node size for how good a product is in terms of power and heat.

I can't even imagine how bad this would have been if intel stuck to their own process nodes.... At least it trades blows on application performance if not it would be a disaster.

The design likely matters a ton but the process node dictates how many transistors can be crammed into it, more usually means better.

People only tend to care about what makes their products of choice look good and what makes the product they didn't choose look bad at the end of the day all that matters is the performance level at a given wattage for an individuals use case.

This cpu doesn't move the needle forward in anything I use a cpu for and actually goes backwards overall that's really all that matters to me. Process node, efficency, design don't really make any difference to me otherwise.

The only cool thing is those new ddr5 modules that don't seem to help much but are still neat.
 
How's the pricing where you live? So far I've seen the 9950x going for cheaper.
Here the 9950X is $699. 285K is $630. Newegg prices. Amazon doesn't have either yet.
So yeah, the 285X is the better value IMPO.
The CUDIMMs to make the ultra 9 better are also not that cheap.
While true, I'm going to spend the extra 20%(ish). I'm also going to get 128GB, maybe more.

Course you do buddy, tell me some more things that didn't happen :laugh:
Got a problem bucko?
 
Last edited:
Why would a someone go from 5800x3d to 9800x3d either, considering the costs? Too many people conflate wants & needs here & yes the earth is going to sh!t as a result of that!

Not the only reason but one of the main ones :ohwell:
If you can afford it it would be worth it. The 5800X3D is slower than the 7000X3D chips in Gaming.
 
I don't know if even that's true, you have to keep in mind a lot of those area savings come from much smaller caches, every E core cluster has 4MB (so just 8MB for all E cores) and every P core has 3MB of L2 cache (24MB total). In reality they could probably cram 8 P cores in there instead of 16 E cores with reduced caches and the overall size would be about the same. As it stands I simply cannot see even one advantage this approach has, it's not faster, it's hardly more power efficient and it barely saves die space.

It's rougly 8 P-core vs 24 E-core

thats 8000 vs 18000 CPUz score and 8x3MB vs 6x4MB The E-core demolishes anything P-core.

1729895832317.png
 
Here the 9950X is $699. 285K is $630. Newegg prices. Amazon doesn't have either yet.
So yeah, the 285X is the better value IMPO.
Ouch, that's way higher than the current MSRP for the 9950x (which should be ~$550), and also higher for the Ultra 9.
While true, I'm going to spend the extra 20%(ish). I'm also going to get 128GB, maybe more.
I do wonder how well Ultra 9 will handle higher density DIMMs, and if CUDIMMs will help with that.
Both AMD and Intel currently have issues with higher density setups, getting 128/192GB going at reasonable frequencies is really annoying.
 
I do wonder how well Ultra 9 will handle higher density DIMMs, and if CUDIMMs will help with that.
The specs say that the max supported RAM is 192GB, which directly implies 48GB packages. I'm betting the actual functional limit is likely 384GB but Intel stated what they can certify in-house.
TPU's own specs page doesn't state RAM limits, so it's like unknown ATM.
In the past the stated max from Intel was usually half of what the actual max was, so who knows?
 
It's rougly 8 P-core vs 24 E-core

thats 8000 vs 18000 CPUz score
It would be more like 12000 and this categorially would not scale nearly as well in most other things. No matter how you spin it the E cores are duds in these CPUs, nothing you wouldn't be able to do with 16P cores and smaller caches vs 8P/16P at similar die sizes and you wouldn't need to deal with the asymmetric architecture in software which will forever remain a problem.

And if you're trying to save space this is the wrong approach anyway, AMD figured it out years ago with chiplets, you simply break up the chip in smaller dies with higher yields and you lower the cost that way.
 
Last edited:
The specs say that the max supported RAM is 192GB, which directly implies 48GB packages. I'm betting the actual functional limit is likely 384GB but Intel stated what they can certify in-house.
The biggest available UDIMMs at the moment are 48GB sticks, so you can use 4 of those to achieve 192GB. Be aware that you'll not be able to reach high frequencies with this config.
64GB UDIMMs are supposed to come out next year, I'm just waiting for those to be avaiable to do the jump to DDR5 and double my current RAM amount.

TPU's own specs page doesn't state RAM limits, so it's like unknown ATM.
It's 192GB, there's no physical way to get more than that ATM.
 
Back
Top