Review AMD Ryzen 7 7800X3D Review: New Gaming Champ Beats Pricier CPUs

[colindog]

and a $500 console is not justifiable when there are no games on the console that a person wants to play, or if they are on consoles AND a comp, and the person has them on their comp already, whats your point ?

and a $500 console is not justifiable when there are no games on the console that a person wants to play, or if they are on consoles AND a comp, and the person has them on their comp already, whats your point ?

I think he's just sad he's poor.

 

[nerdvous]

This motherboard has a BIOS flashback feature which allows you to update the BIOS with a USB flashdrive with no parts installed, just PSU plugged into the motherboard for power. I have done this many 10's of times with no issues, though if power is cut to the motherboard in the middle of the flash it can cause some serious issues.

Thanks for your post. Am I right in thinking that regardless of what AM5 motherboard you choose to pair with a 7000x3D processor, for best results you're going to want to update it in this way to the latest available BIOS?

 

[helper800]

Thanks for your post. Am I right in thinking that regardless of what AM5 motherboard you choose to pair with a 7000x3D processor, for best results you're going to want to update it in this way to the latest available BIOS?

Usually motherboards have a sticker or other such way of informing you what BIOS version it has out of the box. If you get a motherboard with a BIOS that does not support the CPU you have purchased with it, then yes you will need to do a BIOS update for support with those parts. If your motherboard does not have a BIOS flashback type feature then you will need to use another method of update. The only other way to update the motherboard would be to get a hold of a supported CPU and then update the BIOS through whatever method is available to that motherboard (there are several methods).

 

[bit_user]

So it's 16.4% faster than the 13600K in gaming and about 14% slower in productivity (multithread) while costing 40% more.

Would you pay 40% more to get 16% faster gaming performance? I sure wouldn't.

You can apply this same logic to everything at the high-end! Most people buying premium products know they're not getting the best value - that's not the point!

The whole idea behind premium products is to deliver the best performance, for those with bigger budgets. Some people just want to have the best framerates possible, no matter the cost (or, more often, within reason). In absolute terms, the price difference pales in comparison to what we see between mid-tier vs. top-end GPUs.

 

[bit_user]

Intel has T-series chips specifically for low power builds.

That's 35W, which is frankly a bit low for a decent mini-ITX case. You give up a lot of performance, when you scale down that much. The 65W models (which lack either a T or K suffix) are generally a better move, especially if you limit PL1 to keep turbo in check.

If you want even less power consumption, you should look at embedded ARM systems.

I think Phoenix-based Mini-PCs are going to be a NUC killer. AMD's chiplet-based CPUs don't scale down too well, below 65 W. However, their APUs tend to deliver very compelling perf/W, and 35 W tends to be right in their sweet spot.

 

[bit_user]

I'm starting to wonder if AMD limited the clocks a bit too much on the VCache CCDs in order to avoid a "3D be too good; Zen5 could not shine as much", kind like it happened with the 5800X3D?

IMO, that doesn't make a ton of sense. The better they perform, the more AMD can charge for them and the more they sell. AMD needs to sell CPUs now, a lot more than they need to worry about "protecting" their Zen 5 launch.

Also, if you're right, then their X3D's should overclock a lot better than what I think we've seen so far.

In either case, given the power envelope of this thing, I'm expecting AMD to bring this to a laptop. At least, Zen5's version should be in a laptop. One can dream, right? 😀

Look for it on their laptop APUs. That would be something!

 

[bit_user]

Thanks for the reply!

This is all to say that the 7800X3D will cost less than $449 in the very near future, so it will not be limited to upper-tier builds forever. These reviews live a long time (hopefully forever), so we have to also account for the long term.

You touch on this, but it's worth emphasizing that, by the time 7800X3D experiences significant price drops, the gap between DDR4 and DDR5 will be even less. And because people do tend to refer back to the launch reviews for quite a long time, I think that argues even more in favor of de-emphasizing the lack of DDR4 support.

Thanks for quoting the line from Micron on both the expected crossover point and the pricing fundamentals. Good info.

This quibble aside, many thanks for your thorough reviews and articles.

 

[bit_user]

I also think that talking about Raptor Lake like it drops into existing boards is a totally invalid comparison, since data shows that the Intel chips pretty much aren't competitive to 7000 Ryzen chips with DDR4. The 5800X3D drops into existing boards, while we're comparing worse/old platforms.

Eh, not to switch sides on you or anything, but here's the only data on that I've seen:

Core i9 13900K DDR5 7200 MHz (+memory scaling) review

Recently look at the performance differential between DDR5 6000 and 7200 MHz on Raptor-Lake, Intels Gen 13th series processors. Today we review a TeamGroup T-Force Delta 7200 CL34 (!) DDR5 kit to see... Performance - Gaming RTX 3090 - Watch Dogs: Legion

www.guru3d.com

It does not, in fact, invalidate the notion of using a high-end CPU like the i9-13900K with DDR4, if you're primarily interested in gaming. For certain other tasks, the delta is bigger. However, you really have to get into some of the heavily-threaded SPECbench type workloads to see a massive difference.

That said, people buying a top-spec CPU and GPU will typically not want to leave that last few % on the table, simply to save a little on memory + motherboard. Those looking to max-out their framerates will be well-advised to opt for DDR5.

I should acknowledge that my complaints about Intel heat in above posts isn't valid for 13th generation.

Maybe not for gaming, but the Gen 13 i9-K did up the PL1 from 241 W to 253 W, and all the way up to 320 W in the KS variant.

However, when referenced against the competition that does offer dual support for DDR4/5, then the lack of DDR4-support indeed becomes a "relative" con. It's a matter of having more options.

It's not a "con" if anyone likely to buy such a CPU is unlikely to choose DDR4, even if given the opportunity. That's why I draw a distinction between premium and mid-market CPUs.

 

[bit_user]

Intel boards with the W680 chipset and DDR4 support remained paper tigers or simply not available for purchase, nor would they accept Raptor Lake CPUs...

I'm pretty sure I've run across a few. AFAIK, all W680 boards support Raptor Lake, even if it took some manufacturers a while to update their list of supported CPUs. There is no W780 chipset, so W680 is all we get for entry-level workstations on both Gen 12 and Gen 13 CPUs.

even ECC DDR5 no longer has prohibitive prices, but you don't get it at the kind of high-speed variants, that makes perform significantly better.

My solution (since I don't need a lot of RAM) is to go with 2x 16 GB DIMMs of DDR5-4800. After DDR5-5600 ECC UDIMMs hit the market and prices level off, I'll upgrade to 2x 32 GB of that. Still a bit wasteful, but not too much.

without throwing away 128GB of perfectly capable ECC RAM, which I can't really sell to anyone,

I'd consider buying used ECC memory on ebay. ECC provides that added safety margin and RAM typically has a good amount of life in it.

I have no idea what the cost of having the IP block on the IOD would have been in terms of die space or perhaps IP license cost, but I can't believe it to be that high.

The bigger issue might be the degree to which complicates the cache/memory subsystem, since DDR5 is comprised of channel-pairs that are half the width of DDR4 channels. Keep in mind that if AMD added DDR4 support on AM5, they'd have to support that on all future AM5 CPUs! That makes any design or cost tradeoffs a lot more weighty.

I'm not claiming it makes business sense, but I think it's fun to imagine AMD making a different I/O Die that supports DDR4 and AM4. Then, sell the hybrid as Ryzen 6000-series, for the AM4 socket. That avoids market confusion, keeping AM5 as a DDR5-only platform, but gives AM4 customers one more step in their upgrade path and lets more budget-constrained users adopt Zen 4 on a cheaper platform and with cheaper RAM.

I think, considering when it launched, that it was fair for AMD to take a stand and assert DDR5 as the standard for AM5. And it would've worked out fine, except for bad timing relative to the whole market downturn.

Yes, I agree with Paul that the lack of DDR4 backward compatibility for non-APU chips is still a detriment today, perhaps not so much towards the end of the AM5 life cycle in 2025.

If that's your standard, then you should hold him to account when he fails to list lack of DDR4-compatibility as a CON of Intel Gen 14 CPUs, in 2024.

 

[bit_user]

If you see a review that has intel thermal throttling then delete them from your trusted reviewer list...
They use the "out-of-the-box" excuse to use the mobo with the absolute most terrible settings just to make a clickbaity article.

With everything else being equal, using the same cooling, the 13900k reaches 330W ~30% above their advertised power draw at 86 degrees which is 15-20 degrees below the throttling point while the 7950x reaches the thermal throttle point of 95 degrees without even reaching the advertised power draw.

Just like AMD the better the cooling the higher the possible boost...oh wait, that's not right, AMD throttles before it can go above their performance target.

https://www.anandtech.com/show/17641/lighter-touch-cpu-power-scaling-13900k-7950x/3

The 13900k is a 125W base TDP CPU (143W )and here you see why, going to the max turbo TDP of 253W (330W ) only gives you like 20% more performance for more than 100% more power.
On the other hand the 7950x is already at max at 125W (166W ) ,you gain like 5% going to 230W (215W ) because it's already overclocked to the max and thermal throttling at 95 degrees.

(Fromt he same link on the next page)

It must be noted that review utilized the following cooler:

"EKWB EK-AIO Elite 360 D-RGB 360mm"​

AFAIK, that's a $200 cooler. Any claims regarding what that shows about the i9-13900K's thermal throttling behavior are limited in scope to those using a similar or better cooler.

Same argument, with everything being the same the intel one is still much cooler...

The only thing that's cooler is the die, itself. The amount of power consumed is what determines the amount of heat expelled into the room. Thermodynamics 101.

@dalauder

 

[dalauder]

Eh, not to switch sides on you or anything, but here's the only data on that I've seen:

Core i9 13900K DDR5 7200 MHz (+memory scaling) review

Recently look at the performance differential between DDR5 6000 and 7200 MHz on Raptor-Lake, Intels Gen 13th series processors. Today we review a TeamGroup T-Force Delta 7200 CL34 (!) DDR5 kit to see... Performance - Gaming RTX 3090 - Watch Dogs: Legion

www.guru3d.com

It does not, in fact, invalidate the notion of using a high-end CPU like the i9-13900K with DDR4, if you're primarily interested in gaming. For certain other tasks, the delta is bigger.

I really thought I saw a larger gaming difference, but it was probably other tasks. Thanks for the info.

 

[dalauder]

It must be noted that review utilized the following cooler:

"EKWB EK-AIO Elite 360 D-RGB 360mm"​

AFAIK, that's a $200 cooler. Any claims regarding what that shows about the i9-13900K's thermal throttling behavior are limited in scope to those using a similar or better cooler.


The only thing that's cooler is the die, itself. The amount of power consumed is what determines the amount of heat expelled into the room. Thermodynamics 101.

@dalauder

Thermodynamics--why AMD's chips are so great for servers lately.

That $200 cooler is exactly what I'm complaining about. OEMs are putting in $40 AIO 140mm water coolers.

 

[TerryLaze]

Stop with the 7950X throttling nonsense, 7950X doesn't throttle at peak temp; AMD designed the chip to go as high as 95 Celsius and operate at that temperature. This is not a throttling point but a thermal target. You simply don't understand how Zen4 works or choose to ignore it because your employers at Intel wouldn't look kindly at that.
Techpowerup put the 7950X through rigorous thermal testing and found the CPU to perform great even with handicapped air cooling measures.

Did you even read the thing you posted?!
"Your CPU won't get damaged, or the PC won't turn off the moment it hits this temperature, but rather the processor will aim to run at this temperature, which is the highest safe operational temperature, and adjust its boost frequency and voltages accordingly, to keep the processor at this temperature (not under this temperature). "
So it slows down when reaching 95 degrees but it's not thermal throttling?
What do you think that thermal throttle means?
Spoiler: It means that the CPU throttles its clocks( adjust its clocks/boost) when reaching a certain temp.

nice try twisting words around terrylaze. go back an re read my post.
as that is not what i said. but you cant help find ways to bash amd in some way when intel looks bad. no wonder most of your posts always come across as " i hate amd, lets see if i can post something just to make them look bad

sorry terry, but the times i have done cpu swaps, on intel OR amd, the bios is usually reset when i 1st turn the comp on after switching cpus. i cant remember the wording on the screen, but it then says to enter the bios and load defaults, or adjust things, something to that effect.

If the bios resets after you switch the CPU then what I said is correct.

It must be noted that review utilized the following cooler:

"EKWB EK-AIO Elite 360 D-RGB 360mm"​

AFAIK, that's a $200 cooler. Any claims regarding what that shows about the i9-13900K's thermal throttling behavior are limited in scope to those using a similar or better cooler.


The only thing that's cooler is the die, itself. The amount of power consumed is what determines the amount of heat expelled into the room. Thermodynamics 101.

@dalauder

No matter what cooler you use the intel chip is cooler than the ryzen chip.
They have tables with lower TDP settings where you don't need a $200 cooler to get decent temps.
Yes, the starting comment was about the cost of cooling the CPU/die and not about how much heat you get into your room.
If you care about heat in your room thn lock both to 140W max, and the intel die will still be cooler with the same amount of cooling and ryzen will be ~10% more efficient but only at full load.

 

[TerryLaze]

It must be noted that review utilized the following cooler:

"EKWB EK-AIO Elite 360 D-RGB 360mm"​

AFAIK, that's a $200 cooler. Any claims regarding what that shows about the i9-13900K's thermal throttling behavior are limited in scope to those using a similar or better cooler.

Yeah, think about that for a moment.
The 7950x can't hit its advertised power target even with a $200 cooler...

 

[dalauder]

Yeah, think about that for a moment.
The 7950x can't hit its advertised power target even with a $200 cooler...

Wait...I thought we all agreed that Intel's 13th gen has nice thermals?

I think the argument stemmed from criticism of DDR4 and platform cost for the 7800X3D when Intel got away with NOT being criticized for 10th, 11th, and 12th gen thermals? That is, reviewers tend to nitpick AMD when they don't care about legitimate performance limiters in the platform's price range with Intel--just idealized simulations. Even when the 7800X3D should get a clean win across the board, there's two totally invalid dings in the CONS list.

 

[bit_user]

No matter what cooler you use the intel chip is cooler than the ryzen chip.

Not necessarily, but that's also somewhat beside the point.

The reason the i9 is cooler in that article is because the AIO liquid-cooling system they used is capable of dispelling heat almost as fast as it reaches the heatspreader. Were a significantly less capable cooler used, it would be the bottleneck and the i9 would probably reach its throttling temp. It's not hard to find data on this.

Source: https://www.tomshardware.com/features/intel-core-13900k-cooling-tested/2​

The data above uses delta above ambient. Unfortunately, they don't state the ambient temp, but it's a good bet it's somewhere in the neighborhood of 20 C, which would put the CPU with weakest cooler at its throttling temp. The article text indeed confirms that the CPU throttled with this cooler.

What's interesting is that even coolers which don't reach the throttling temp still have trouble unleashing the CPU to its full potential. From the same article:

The moral of the story is that buyers should pay close attention to the cooler used in CPU reviews, and can't necessarily assume the same performance from a lesser cooler.

Yeah, think about that for a moment.
The 7950x can't hit its advertised power target even with a $200 cooler...

It's a limit, not a target! There's no rule that says you have to reach the power limit before you hit the thermal limit!

We've been through this before, and I think we all agree that Ryzen 7000/AM5 has an inferior package for thermal conductivity. This data suggests it's mainly an issue just for compute workloads, so long as you use something better than a Wraith Spire for cooling:

Source: https://www.techpowerup.com/review/amd-ryzen-9-7950x-cooling-requirements-thermal-throttling/​

​

I see @Elusive Ruse already quoted the performance data from the same article, which shows that you merely need a decent cooler to unlock > 99% of the performance the CPU can achieve with a 420 mm AIO.

 

[bit_user]

Wait...I thought we all agreed that Intel's 13th gen has nice thermals?

Depends what you mean by "thermals" and entirely on what you're doing. For gaming, it's generally not too bad.

Even while 114 W isn't that much, it's still well above both the i9-11900K and i9-12900K - both of which sit above any AMD CPU.

It's the power consumption on non-gaming workloads that really make Intel blush:

​

Source: https://www.techpowerup.com/review/intel-core-i9-13900k/22.html​

I think the argument stemmed from criticism of DDR4 and platform cost for the 7800X3D when Intel got away with NOT being criticized for 10th, 11th, and 12th gen thermals?

Most Intel vs. AMD arguments seem to end up on power consumption, because it's become such a weak spot for Intel. Terry tries to confuse the issue by treating CPU temperature as a proxy for power consumption, but the relationship is fairly indirect and involves several other variables.

 

[Elusive Ruse]

Did you even read the thing you posted?!
"Your CPU won't get damaged, or the PC won't turn off the moment it hits this temperature, but rather the processor will aim to run at this temperature, which is the highest safe operational temperature, and adjust its boost frequency and voltages accordingly, to keep the processor at this temperature (not under this temperature). "
So it slows down when reaching 95 degrees but it's not thermal throttling?
What do you think that thermal throttle means?
Spoiler: It means that the CPU throttles its clocks( adjust its clocks/boost) when reaching a certain temp.

As I guessed, you have no clue what you are talking about, yet insist on repeating the same talking point as if it would make it true. Not only that, you misrepresent facts so you can make yourself look good. As the aritcle says, the 7950X stays at 95C which means it doesn't throttle at that temperature, so the 7950X does not throttle at 95C, if it did, then it would lose performance at that temperature and "throttle" itself in order to cool down and protect itself against damage. That's what thermal "throttling" means.
I'm gonna post these charts again, not for your benefit since you simply don't want to accept the reality but so that your misinformation campaign won't run rampart here without response. How can a CPU throttle if it gives you the same performance with an AIO or an air cooler at the same temperature under different loads? You get over 97% performance even with the NH-U14 running at 20% FFS!

 

[Elusive Ruse]

Not necessarily, but that's also somewhat beside the point.

The reason the i9 is cooler in that article is because the cooler is capable of dispelling heat almost as fast as it reaches the heatspreader. Were a significantly less capable cooler used, it would be the bottleneck and the i9 would probably reach its throttling temp. It's not hard to find data on this.

​

Source: https://www.tomshardware.com/features/intel-core-13900k-cooling-tested/2​

The data above uses delta above ambient. Unfortunately, they don't state the ambient temp, but it's a good bet it's somewhere in the neighborhood of 20 C, which would put the CPU with weakest cooler at its throttling temp. The article text indeed confirms that the CPU throttled with this cooler.

What's interesting is that even coolers which don't reach the throttling temp still have trouble unleashing the CPU to its full potential. From the same article:

​

The moral of the story is that buyers should pay close attention to the cooler used in CPU reviews, and can't necessarily assume the same performance from a lesser cooler.


It's a limit, not a target! There's no rule that says you have to reach the power limit before you hit the thermal limit!

We've been through this before, and I think we all agree that Ryzen 7000/AM5 has an inferior package for thermal conductivity. This data suggests it's mainly an issue just for compute workloads, so long as you use something better than a Wraith Spire for cooling:

Source: https://www.techpowerup.com/review/amd-ryzen-9-7950x-cooling-requirements-thermal-throttling/​

​

I see @Elusive Ruse already quoted the performance data from the same article, which shows that you merely need a decent cooler to unlock > 99% of the performance the CPU can achieve with a 420 mm AIO.

Finding data about 13900K throttling is easy:

View: https://www.youtube.com/watch?v=7o74u1u642I

 

[abufrejoval]

I'm pretty sure I've run across a few. AFAIK, all W680 boards support Raptor Lake, even if it took some manufacturers a while to update their list of supported CPUs. There is no W780 chipset, so W680 is all we get for entry-level workstations on both Gen 12 and Gen 13 CPUs.

I live in Germany and use geizhals.de to check availability, which covers most of Europe: we don't get quite everything Newegg or Amazon US offer and I'm quite sure there are things available in Asia we never even get to see.

Supermicro was their usual best, had product and stock early last year, but they offer only DDR5 and DDR5-ECC could not be bought from EU retail for months. When it became available, prices stopped all interest.

Several ASRock DDR4 variants were announced on Anandtech in March 2022, but never made available here. I went ahead and asked ASRock themselves, but they gave me the complete run-around and no source to buy it from. If I remember correctly, they had done one production run, which was sold out and they didn't know if they'd do another.

The only DDR4 variant that ever did pop up late last year is the GIGABYTE MW34-SP0 which evidently does not support Raptor Lake with revision 1.0 of the board and has plenty of other issues.

Raptor Lake support should be no problem, because socket 1700 chipsets are designed by Intel to be compatible, but if for some BIOS reason the mainboard doesn't actually work with Raptors, you're out there in the rain alone.

People tried with that Gigabyte and failed. And revision 1.1 of the board cites 13th gen support as defining feature... Unfortunately that revision isn't yet available in retail...

Entry level servers/workstations have been at 95% attainability for years and I'm ever more convinced, market segmentation plays a large role.

My solution (since I don't need a lot of RAM) is to go with 2x 16 GB DIMMs of DDR5-4800. After DDR5-5600 ECC UDIMMs hit the market and prices level off, I'll upgrade to 2x 32 GB of that. Still a bit wasteful, but not too much.


I'd consider buying used ECC memory on ebay. ECC provides that added safety margin and RAM typically has a good amount of life in it.

Most ECC RAM on eBay is registered server DIMMs, which won't work on these boards.
I haven't checked DDR5 ECC on eBay (it tends to be picked up by geizhals.de), but it not likely to be cheap yet.

The bigger issue might be the degree to which complicates the cache/memory subsystem, since DDR5 is comprised of channel-pairs that are half the width of DDR4 channels. Keep in mind that if AMD added DDR4 support on AM5, they'd have to support that on all future AM5 CPUs! That makes any design or cost tradeoffs a lot more weighty.

I'm not claiming it makes business sense, but I think it's fun to imagine AMD making a different I/O Die that supports DDR4 and AM4. Then, sell the hybrid as Ryzen 6000-series, for the AM4 socket. That avoids market confusion, keeping AM5 as a DDR5-only platform, but gives AM4 customers one more step in their upgrade path and lets more budget-constrained users adopt Zen 4 on a cheaper platform and with cheaper RAM.

I think, considering when it launched, that it was fair for AMD to take a stand and assert DDR5 as the standard for AM5. And it would've worked out fine, except for bad timing relative to the whole market downturn.
If that's your standard, then you should hold him to account when he fails to list lack of DDR4-compatibility as a CON of Intel Gen 14 CPUs, in 2024.

These are all valid points and I wonder what the end-user price uptick for DDR4 would have been in the end.

Pretty sure it might have been as low as €10 per system, which I'd have paid easily, but perhaps 1 Billion others not.

 

[bit_user]

Supermicro was their usual best, had product and stock early last year, but they offer only DDR5 and DDR5-ECC could not be bought from EU retail for months. When it became available, prices stopped all interest.

A good price on the X13SAE is about $430, in the US. The last Supermicro workstation board I bought cost about $350, approx 10 years ago. Adjusting for inflation, I'd say it's about right. You really can't get workstation boards for much less, these days.

Raptor Lake support should be no problem, because socket 1700 chipsets are designed by Intel to be compatible, but if for some BIOS reason the mainboard doesn't actually work with Raptors, you're out there in the rain alone.

People tried with that Gigabyte and failed. And revision 1.1 of the board cites 13th gen support as defining feature... Unfortunately that revision isn't yet available in retail...

It's just a waiting game, until they release updated BIOS.

Sometimes, the easiest thing to do is buy the cheapest LGA 1700 CPU you can find, used. Use that to update the BIOS, and then pass it on. Probably shouldn't cost more than about $40, and you should be able to recoup most of that.

Entry level servers/workstations have been at 95% attainability for years and I'm ever more convinced, market segmentation plays a large role.

Unobtainable, you mean? For DDR5, it kinda doesn't matter if you can't even source the ECC UDIMMs, which was the situation until about December 2022 (as you mentioned). I was only paying attention to availability of the DDR5 boards.

Maybe that had a lot to do with W680 availability, in general. Otherwise, I assume there were mainly supply-chain issues behind that. The chipset itself should be the same silicon Intel uses in its consumer version, and I'm guessing they sell it for a higher margin. Given Sapphire Rapids Xeon W wasn't yet on the market, there really was no good reason for them to drag their feet.

Most ECC RAM on eBay is registered server DIMMs, which won't work on these boards.

That's why you adjust your search query to specify "unbuffered" or "UDIMM". I literally tried exactly that, as I wrote that post.

https://www.ebay.com/sch/i.html?_nk...w=ecc+udimm+32GB&_osacat=0&LH_ItemCondition=4​

 

[abufrejoval]

It's just a waiting game, until they release updated BIOS.

Sometimes, the easiest thing to do is buy the cheapest LGA 1700 CPU you can find, used. Use that to update the BIOS, and then pass it on. Probably shouldn't cost more than about $40, and you should be able to recoup most of that.

I don't think it was just a BIOS issue, because typically you can update a BIOS these days even without a CPU: a new revision seems to hint at a hardware change as well.

And in those days when an older CPU was still required for BIOS updates (e.g. Kaby Lake not booting on Skylake boards), I typically just took advantage of the 14-day free return window e-tailers have to give customers by law in the EU.

Unobtainable, you mean? For DDR5, it kinda doesn't matter if you can't even source the ECC UDIMMs, which was the situation until about December 2022 (as you mentioned). I was only paying attention to availability of the DDR5 boards.

Maybe that had a lot to do with W680 availability, in general. Otherwise, I assume there were mainly supply-chain issues behind that. The chipset itself should be the same silicon Intel uses in its consumer version, and I'm guessing they sell it for a higher margin. Given Sapphire Rapids Xeon W wasn't yet on the market, there really was no good reason for them to drag their feet.

Yes, whenever I tried to put all my requirements into a shopping basket, some key element was always missing. And there is just no way of knowing if and when that might eventually get fixed.

That's why eventually I switched from an Intel Xeon to a Ryzen 5000 to get affordable ECC and still quite a bit of a hardware boost over the previous Haswell Xeon E3.

And an important part of that switch was to retain "unnoticeable performance", meaning noise levels that would never distract even at continued peak loads.

And there the 5800X3D was perhaps the better bet than a Raptor, because it will never even try to convert 250 Watts of power into heat. Sure, I like peak scalar speed like anybody else, but not when it distracts or annoys.

That's why you adjust your search query to specify "unbuffered" or "UDIMM". I literally tried exactly that, as I wrote that post.

https://www.ebay.com/sch/i.html?_nk...w=ecc+udimm+32GB&_osacat=0&LH_ItemCondition=4​

I just got 64GB (2x32GB) of brand new Kingston DDR4-3200 ECC for €160 so eBay sellers don't seem to account for the current free fall in DRAM prices. DDR5-4800 ECC would still be around €280.

BTW: It was SK Hynix instead of the Microns I had already, and I might have 4 Samsung DIMMs on my 5950X, but Kingston evidently ensures 100% compatibility between all variants of KSM32ED8/32**

DRAM sticks get moved around quite a bit in my home-lab and one of my gripes is that I got sizeable DIMM and SO-DIMM populations I can't mix, even if they are otherwise the same specs...

 

[dalauder]

Not necessarily, but that's also somewhat beside the point.

The reason the i9 is cooler in that article is because the AIO liquid-cooling system they used is capable of dispelling heat almost as fast as it reaches the heatspreader. Were a significantly less capable cooler used, it would be the bottleneck and the i9 would probably reach its throttling temp. It's not hard to find data on this...

What's interesting is that even coolers which don't reach the throttling temp still have trouble unleashing the CPU to its full potential. From the same article:

The moral of the story is that buyers should pay close attention to the cooler used in CPU reviews, and can't necessarily assume the same performance from a lesser cooler....

One thing that people often overlook is that many AIO liquid coolers simply aren't that good. Many aren't as good as the DeepCool AG620 air cooler shown above, and that 3% is already enough to mess up the i9's performance advantage.

Honestly, all of these CPUs are overkill for 95% of users and even if you use 16+ threads, getting 97% of the performance is good enough. So any $500+ processor SHOULD actually make anyone happy.

I'm happy with a Ryzen 5 3600X though and an RTX 2080 Ti (refurb for $247). I don't know who most of this new hardware is made for. $1000 graphics cards, sheesh!

 

[dalauder]

Depends what you mean by "thermals" and entirely on what you're doing. For gaming, it's generally not too bad.

Even while 114 W isn't that much, it's still well above both the i9-11900K and i9-12900K - both of which sit above any AMD CPU.

It's the power consumption on non-gaming workloads that really make Intel blush:

​

Source: https://www.techpowerup.com/review/intel-core-i9-13900k/22.html​

Most Intel vs. AMD arguments seem to end up on power consumption, because it's become such a weak spot for Intel. Terry tries to confuse the issue by treating CPU temperature as a proxy for power consumption, but the relationship is fairly indirect and involves several other variables.

Very true. But the 13th generation doesn't have the crippling thermals that make all reviews flat-out lies, like the generations right before it.

And considering that Ryzen 7000 throttles as much as Intel's 13th Gen, I'll call that argument on performance a push. But I might be wrong.

If you want to talk about efficiency...well don't talk about Intel.

 

[dalauder]

As I guessed, you have no clue what you are talking about, yet insist on repeating the same talking point as if it would make it true. Not only that, you misrepresent facts so you can make yourself look good. As the aritcle says, the 7950X stays at 95C which means it doesn't throttle at that temperature, so the 7950X does not throttle at 95C, if it did, then it would lose performance at that temperature and "throttle" itself in order to cool down and protect itself against damage. That's what thermal "throttling" means.
I'm gonna post these charts again, not for your benefit since you simply don't want to accept the reality but so that your misinformation campaign won't run rampart here without response. How can a CPU throttle if it gives you the same performance with an AIO or an air cooler at the same temperature under different loads? You get over 97% performance even with the NH-U14 running at 20% FFS!

I was going to nitpick about the slight thermal throttling demonstrated, but then I saw that any cooler can really come out on top and that, somehow, the lowest air setting somehow is fastest. So it's all really within the error of measurements.