Been away from PC gaming, a bit lost...

[Why_Me]

I this board. It's like Roman gladiators but with techies.

 

[dgianstefani]

What a twist, the non X parts and the X3D parts (both undervolted/underclocked compared to X parts, top efficiency. And there's the Alder Lake 13400F (a rebranded 12600 from... 2021).

Lets look to see where the 7950X is (the first X series Zen 4 chip on this list), oh, behind the two year older 2021 i5 12600. Gosh.

The "but power draw" "muh nuclear reactor" posts get boring. Guess what. Your super efficient Zen 4 chip idles at around 50 watts, with total system power draw edging close to 100 W due to dual chipset/dual chiplet design and the infinity fabric. Guess what the 13600K system I have idles at? CPU only is less than 10 W, and total system is around 45 W.

Browsing the web? Again, Intel chips more efficient.

Loading up Cinebench to do a synthetic all core load (a normal use case...)? Wow, you can point to the big number and laugh, ignoring the fact that the 14900K has 24 cores which clock at up to 6.2 GHz, and has approximately 20% faster ST performance and more than double the MT performance of the "efficient" 7800X3D.

It's really not some impressive thing to see a 5nm undervolted/underclocked Zen 4 part (X3D chips) be more efficient in synthetic all core loads than a 10 nm derivative part with more cores that clocks higher.

It's almost like desktop architectures don't need to purely optimize for peak efficiency like laptops or phones do. And people who actually work on their computers don't care about a few dollars more on their power bill, if it means they can get that render done faster, or compile that code faster. But guess what, if you care so much about efficiency, simply buy a non K part.

/rant

 

[kapone32]

Literally because I already had an Optimus AMD waterblock so going to AM5 was a little cheaper, rather than buying new watercooling parts moving from AM4 to LGA 1700. AM5 was an experiment I learned from, Intel it is for the next build. Same goes for other professionals I know who I won't name.

But honestly, you need to stop using emotional arguments "love" "fanboy" etc. it's tedious to read and doesn't make you come off as informed.



Here you go. From the 7800X3D review, since you love bringing that CPU up as if it's magic.

View attachment 348471
Something you might notice is that the last gen (Alder Lake) non K parts top efficiency charts. It just so happens that TPU hasn't reviewed Raptor Lake parts that aren't K series 600K/700K/900K since they weren't sampled this time around, but has reviewed the non X AMD parts, which are equivalent, lower clocked tuned for efficiency. Hence why those SKUs along with X3D parts (similar story, undervolted underclocked, but in this case due to cache voltage sensitivity) cause AMD to look so good in "efficiency" charts (100% all core synthetic load is not a typical use case BTW).

I guess in your world there has not been media about the new BIOS to turn down the power on 14th Gen chips. You can call that emotional if you want but the truth cannot be refuted. That is why you don't post raw power as it would blow your argument into shreds. Just because you didn't sample it does not mean the entire industry is wrong. It is just like how you tell me my 7900X3D is a bad CPU. My other question would be how can you recommend to the OP if you don't know what you are talking about?

 

[dgianstefani]

I guess in your world there has not been media about the new BIOS to turn down the power on 14th Gen chips. You can call that emotional if you want but the truth cannot be refuted. That is why you don't post raw power as it would blow your argument into shreds. Just because you didn't sample it does not mean the entire industry is wrong. It is just like how you tell me my 7900X3D is a bad CPU. My other question would be how can you recommend to the OP if you don't know what you are talking about?

"truth" "to shreds" "don't know what you are talking about"

More emotion. Post some evidence and back it up with a coherent statement or please cease wasting our time.

Power draw is completely irrelevant. Work done/power used is what matters. Hence "efficiency". Please look up the definition.

Third party motherboard vendors doing their own thing out of Intel specs to do better in out of the box benchmarking "reviews" for techtubers catering to "enthusiasts" is completely irrelevant to this discussion. In fact this entire red herring you keep bringing up is completely irrelevant to the OP, and you keep inserting topics noone has brought up "my 7900X3D is not a bad CPU".

OP messaged me asking about a motherboard with WiFi and different memory, so here's a slightly revised part list. Comes with a white CPU cooler too as a perk for the theme.

PCPartPicker Part List: https://pt.pcpartpicker.com/list/CZjb9c

CPU: Intel Core i7-14700K 3.4 GHz 20-Core Processor (€454.90 @ Globaldata)
CPU Cooler: Deepcool AK620 68.99 CFM CPU Cooler (€66.80 @ Switch Technology)
Motherboard: MSI PRO Z790-A MAX WIFI ATX LGA1700 Motherboard (€247.01 @ PC Componentes)
Memory: Corsair Dominator Titanium 32 GB (2 x 16 GB) DDR5-7200 CL34 Memory (€230.75 @ PC Componentes)
Storage: Western Digital Black SN770 2 TB M.2-2280 PCIe 4.0 X4 NVME Solid State Drive (€146.90 @ Império Multimédia)
Video Card: Zotac GAMING Trinity GeForce RTX 4070 Ti SUPER 16 GB Video Card (€863.04 @ PC Componentes)
Case: Fractal Design Torrent Compact ATX Mid Tower Case (€119.90 @ PCDIGA)
Power Supply: MSI MAG A750GL PCIE5 750 W 80+ Gold Certified Fully Modular ATX Power Supply (€116.50 @ Switch Technology)
Total: €2245.80
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2024-05-22 21:15 WEST+0100

White CPU cooler, RAM, case, mobo with some light elements on it.

White GPUs are very expensive.

WiFi 7 on that board too. Although for a desktop I personally would always use ethernet.

Could get a white GPU too but it's an extra 60 euros just for a color matching part.

Man, sometimes I'm glad I live in the UK, parts of Europe have expensive PC parts relatively.

 

[Bobaganoosh]

Here, I fixed your build.

[snip]

I took dgianstefani's list (which I don't have complaints about), but just tried to see what an AMD (and more white) build looks like:
https://pt.pcpartpicker.com/list/hrnjVW
7800X3D
Deepcool AK620
ASRock X670E Steel Legend
Teamgroup 2x16GB DDR5 6000 CAS30
WD SN770 2TB
Inno3D Twin X2 OC 4070 Ti Super
Fractal Torrent Compact White
SeaSonic Focus GX 850W

Total: €2219.12

So pretty much the same price and it's more white. This was not done to dispute any of the other points made in the thread, just having some fun with PCPartPicker and selecting an AMD rig as I hadn't seen anybody do that for the full price comparison yet. The only thing not white is the power supply and the cables for it. You can get the MSI MAG A850L PCIE5 that's white, but I don't think I could bring myself to buy an MSI (or Gigabyte) power supply. Check reviews, maybe that one is ok, but I haven't looked into those in a bit. I'd always check reviews for everything before buying it, but power supplies can take down everything else in the system when they go, so I don't like to skimp on those. I have had great luck with SeaSonic, Super Flower, EVGA (older ones of specific models, but can't recommend now), Be Quiet!, and Corsair (these aren't in any particular order and there are other ones that are fine, these are just the ones I personally know have a long history of high quality in their better supplies).

 

[dgianstefani]

I took dgianstefani's list (which I don't have complaints about), but just tried to see what an AMD (and more white) build looks like:
https://pt.pcpartpicker.com/list/hrnjVW
7800X3D
Deepcool AK620
ASRock X670E Steel Legend
Teamgroup 2x16GB DDR5 6000 CAS30
WD SN770 2TB
Inno3D Twin X2 OC 4070 Ti Super
Fractal Torrent Compact White
SeaSonic Focus GX 850W

Total: €2219.12

So pretty much the same price and it's more white. This was not done to dispute any of the other points made in the thread, just having some fun with PCPartPicker and selecting an AMD rig as I hadn't seen anybody do that for the full price comparison yet. The only thing not white is the power supply and the cables for it. You can get the MSI MAG A850L PCIE5 that's white, but I don't think I could bring myself to buy an MSI (or Gigabyte) power supply. Check reviews, maybe that one is ok, but I haven't looked into those in a bit. I'd always check reviews for everything before buying it, but power supplies can take down everything else in the system when they go, so I don't like to skimp on those. I have had great luck with SeaSonic, Super Flower, EVGA (older ones of specific models, but can't recommend now), Be Quiet!, and Corsair (these aren't in any particular order and there are other ones that are fine, these are just the ones I personally know have a long history of high quality in their better supplies).

It's a nice list but that PSU doesn't have a native 12+4 pin cable. You'd have to use an adapter with the 4070 Ti Super, it also isn't ATX 3.0/3.1 spec. Not a huge issue since the 4070 Ti Super is very efficient, being only a 290 W card with spikes to ~350 W, but it's still a concern, and will somewhat limit future GPU upgrades.

An 12VHPWR or even better a 12V-2x6 connection like the unit in the list I made has is borderline essential for new builds going forward IMO.

Seasonic make great PSUs but I just wouldn't put a 2019 non ATX 3/3.1 unit in a new high end build.

 

[Bobaganoosh]

It's a nice list but that PSU doesn't have a native 12+4 pin cable. You'd have to use an adapter with the 4070 Ti Super, it also isn't ATX 3.0/3.1 spec. Not a huge issue since the 4070 Ti Super is very efficient, being only a 290 W card with spikes to ~350 W, but it's still a concern, and will somewhat limit future GPU upgrades.

An ATX12VO or even better a 12V-2x6 connection like in the list I made is borderline essential for new builds going forward IMO.

Seasonic make great PSUs but I just wouldn't put a 2019 non ATX 3/3.1 unit in a new high end build.

Fair enough (Edit: I was definitely trying to save a few bucks there and just stuck with a brand I'm more comfortable with). That'd be one more reason to look at reviews for the white MSI one I linked, as it's got that connector on it. hwbusters gave it a pretty good rating for a budget supply. Could be better, but doesn't seem dangerous as far as they could tell, so maybe it's fine.

 

[evernessince]

Literally because I already had an Optimus AMD waterblock so going to AM5 was a little cheaper, rather than buying new watercooling parts moving from AM4 to LGA 1700. AM5 was an experiment I learned from, Intel it is for the next build. Same goes for other professionals I know who I won't name.

But honestly, you need to stop using emotional arguments "love" "fanboy" etc. it's tedious to read and doesn't make you come off as informed.



Here you go. From the 7800X3D review, since you love bringing that CPU up as if it's magic.

View attachment 348471
Something you might notice is that the last gen (Alder Lake) non K parts top efficiency charts. It just so happens that TPU hasn't reviewed Raptor Lake parts that aren't K series 600K/700K/900K since they weren't sampled this time around, but has reviewed the non X AMD parts, which are equivalent, lower clocked tuned for efficiency. Hence why those SKUs along with X3D parts (similar story, undervolted underclocked, but in this case due to cache voltage sensitivity) cause AMD to look so good in "efficiency" charts (100% all core synthetic load is not a typical use case BTW).

Great, low-end last gen and mid-range 2gen old Intel CPUs are more efficient when you aren't using your PC.

That's about the only time you are going to be using one thread. How does your chart apply to the topic? It doesn't, you are off topic and derailing the thread as usual. OP isn't looking one or two gen old Intel CPUs and they would bottleneck his GPU choice anyways. Here is that same chart but with CPUs relevant to the discussion:

Now let's actually look at charts that demonstrate power consumption in scenarios the OP actually intends to use the PC in:

These charts echo the broader concensous that AMD is more power efficient.

(100% all core synthetic load is not a typical use case BTW).

It demonstrates CPU power consumption when fully loaded. You can't tout the core advantage of Intel without coming to the realization that in order to utlize that core advantage you will be at an large disadvantage when it comes to power consumption.

Mind you it doesn't have to be full load, the 7800X3D is a whopping 229% more efficient in gaming workloads.

Third party motherboard vendors doing their own thing out of Intel specs

Intel is on record in a statement to GN that motherboard vendors were in spec, mostly because Intel didn't really define much of a spec in the first place. Intel only introduced the baseline profile very recently, a total of 7-8 months after launch and only after instability was reported. It'd be one thing if this was at product launch but 7-8 months later? Even if we assume that Intel had some sort of spec it has not made public, that would without a doubt be negligence on their part to allow nearly all their vendors violate the specs for such a long period of time. It is beyond a stretch to call it the partner's fault given the facts.

It's a nice list but that PSU doesn't have a native 12+4 pin cable. You'd have to use an adapter with the 4070 Ti Super, it also isn't ATX 3.0/3.1 spec. Not a huge issue since the 4070 Ti Super is very efficient, being only a 290 W card with spikes to ~350 W,

ATX 3.1 is more of a downgrade than an upgrade:

ATX 3.0 is just the 12VHPWR / 12V2X6 cable, which you don't need because adapters exist. It's a terrible design with low safety margins and should be replaced

but it's still a concern, and will somewhat limit future GPU upgrades.

Having multiple regular 8-pins from the PSU side going into the adapter is less of a concern. All having an 12v2X6 on the PSU side does is introduce another point of failure for a connector with low safety tolerances.

In what way does it limit future GPU upgrades?

 

[dgianstefani]

Great, low-end last gen and mid-range 2gen old Intel CPUs are more efficient when you aren't using your PC.

That's about the only time you are going to be using one thread. How does your chart apply to the topic? It doesn't, you are off topic and derailing the thread as usual. OP isn't looking one or two gen old Intel CPUs and they would bottleneck his GPU choice anyways. Here is that same chart but with CPUs relevant to the discussion:

View attachment 348475



Now let's actually look at charts that demonstrate power consumption in scenarios the OP actually intends to use the PC in:

View attachment 348476

View attachment 348477


These charts echo the broader concensous that AMD is more power efficient.



It demonstrates CPU power consumption when fully loaded. You can't tout the core advantage of Intel without coming to the realization that in order to utlize that core advantage you will be at an large disadvantage when it comes to power consumption.

Mind you it doesn't have to be full load, the 7800X3D is a whopping 229% more efficient in gaming workloads.



Intel is on record in a statement to GN that motherboard vendors were in spec, mostly because Intel didn't really define much of a spec in the first place. Intel only introduced the baseline profile very recently, a total of 7-8 months after launch and only after instability was reported. It'd be one thing if this was at product launch but 7-8 months later? Even if we assume that Intel had some sort of spec it has not made public, that would without a doubt be negligence on their part to allow nearly all their vendors violate the specs for such a long period of time. It is beyond a stretch to call it the partner's fault given the facts.



ATX 3.1 is more of a downgrade than an upgrade:

ATX 3.0 is just the 12VHPWR / 12V2X6 cable, which you don't need because adapters exist. It's a terrible design with low safety margins and should be replaced




Having multiple regular 8-pins from the PSU side going into the adapter is less of a concern. All having an 12v2X6 on the PSU side does is introduce another point of failure for a connector with low safety tolerances.

In what way does it limit future GPU upgrades?

Still conveniently ignoring that high end non K Intel Raptor Cove parts aren't on that list I see. Despite X3D/non X AMD chips being there (and leading in efficiency to the surprise of noone).

"when you aren't using your PC"

Talk about cherry picking hahahaha.

Reminder that the vast majority of system power on time is light workloads or intermittent loads then back to idle. Hence why the ATX12VO spec was designed to massively improve idle and low load efficiency, for peak load it didn't do much. For gaming I don't think a stock K CPU using ~130 W (14700K, assuming a 4090 with uncapped framerate, not a 60 FPS target) is going to kill the user from heatstroke and send their power bill to the moon, despite what you may want to imply with percentages.

ATX 3.1 is more of a downgrade than an upgrade:

Irrelevant, since both ATX 3.0 and 3.1 are better than ATX 2.x.

ATX 3.0 is just the 12VHPWR / 12V2X6 cable, which you don't need because adapters exist. It's a terrible design with low safety margins and should be replaced

Yes, the 80% consumer marketshare NVIDIA has or the 90%+ datacentre marketshare, where products for the past three generations have used this or a similar connector, yet where are the burned down houses?

Your comment is factually wrong too BTW, ATX 3.0 is not just the PCIE5/12VHPWR cable, it's a thorough spec that requires units to tolerate spikes up to 200% of rated power etc, amongst many other things.

https://www.cybenetics.com/attachs/52.pdf

A whole 1.4% of gaming performance lost by halving the power limit to 125 W, or 0.2% by limiting to 200 W (even at the stock 14900K power limit, actual gaming power draw is 140 W, so these charts aren't implying that the chip being limited to 200 W always means that it draws 200 W, just that gaming performance changes when you change the V/F curve by changing power limits without manually setting clocks). For those reading, that's how a non K Raptor Cove core CPU would perform.

Still 9.9% faster than a 7950X, and 2.2% faster than a 7950X3D.

One setting in the bios.

 

[Dr. Dro]

I this board. It's like Roman gladiators but with techies.

Lump me in with whatever gladiators hate half-baked technology, stuttering, driver crashes and substandard feature sets

 

[alwayssts]

I have not read this entire thread, so please excuse my ignorance.

Now, truly, is a bad time to build a PC. Computex is literally in a couple weeks, in which the whole landscape/thought-process may change depending upon what happens there (and pricing of new/old parts).

I'm not going to create a sprawling post, but even something like AMD's next GPU is likely worth holding out to build. IMHO, that gpu will be '4k upscaling and/or goodies" to the 1440p60 ps5pro.

When you look at the performance charts, imo, it's pretty clear it has to compete with the 4070 Super in price, the 4070 Ti Super in performance (stock), and with any luck (and really to make sense without a higher-end sku) it should OC to around/above the potential of a stock 4080. As a guy that has recommended the 7900xt for eons, for example, as it will OC to the level of a 4080 and is much cheaper, I really do think that's an important thing to consider.

While gaming has (largely) been scaled off around the level of a 5700xt for around four years, I think that will be bumped up relatively shortly; not just in raster but (eventually, perhaps more slowly) in RT.

I'm not saying people can't be happy with something less; of course they can be. But, imho, the 4080 has been and will be the benchmark for a happy high-end PC gamer for this whole console generation.

And so, it becomes a race to when that level of performance fits within your budget (if not just gaming on a console).

With nVIDIA all-but-stating the cards they will release this year will be faster/more-expensive (a 4080 would be similar to something like a 5070/5070ti) to keep margins high, it's something to at least consider.

As for CPUs, I'll let everyone else fight it out, but I feel a 7700x will suffice until the next console generation (at 1440p, which let's be real...that's what most people use/want to use even if up-scaling to 4k).

Again, I'm not saying it has to be THOSE parts; you do you. I'm just saying those are the performance metrics I would recommend to someone that wants something greater than a console experience until PS6.

That level of performance will likely be much cheaper shortly, and yeah...that's because of AMD. The next (zen 5) iteration of the 7500F (cheap six-core) might (probably will) still be enough for a nice gaming rig.

All JMO.

Some people are very particular about certain aspects, and I respect that. I'm just trying to give you a baseline and what I feel people can expect if they want to save some cash and keep up with the Jones'.

 

[dgianstefani]

Now, truly, is a bad time to build a PC. Computex is literally in a couple weeks, in which the whole landscape/thought-process may change depending upon what happens there (and pricing of new/old parts).

I agree, however sometimes people want to see what the options are right now, and the thread has done a good job of showing those.

My attitude is new parts don't magically make the old ones slower, so a 4080 like you said is still a benchmark card for high end gaming, even if a 5080 releases in six months and is ~35% faster for the same price or whatever.

Hopefully the AI powered FSR is actually decent, and RDNA 4 doesn't flop like RDNA 3, some competent competition like when RDNA 2 was around would be nice.

Raptor Lake and the 7800X3D can already sustain 240 FPS and you're GPU limited most of the time, so I doubt new CPUs will really bring much to the table besides efficiency improvements, some AI stuff, improved platform etc.

 

[evernessince]

Still conveniently ignoring that high end non K Intel Raptor Cove parts aren't on that list I see. Despite X3D/non X AMD chips being there (and leading in efficiency to the surprise of noone).

Well first TPU didn't review the non-k parts, hence they aren't in the charts: https://www.techpowerup.com/review/?category=Processors&manufacturer=&pp=25&order=date

Implying that such an omission was intentional reeks of an argument out of bad faith given that finding comprehensive reviews of non-k 14th gen at all is a difficult task.

Of course it is your point to prove so surely you can foster such charts. An argument made without evidence can be dismissed without evidence.

I very much doubt they'll change the fact that X3D parts are more efficient.

Reminder that the vast majority of system power on time is light workloads or intermittent loads then back to idle.

Is it? My system is almost always running handbrake with other background programs? Your statement would also hold false for anyone gaming or streaming as well.

It's pretty much only valid for people who leave their system idling for long periods of time or only do light work. The problem with that is those workloads are going to have a small power consumption regardless of if you go Intel or AMD. The Intel might be 3 watts more efficient at idle but 60-100w worse in gaming. It doesn't take long for that 3w idle advantage to be completely dwarfed by the energy expended when you are actually using your PC's power.

I'd also like to point out that heat output during max and mixed workloads is a lot more important that during idle / ST. Idling over hours provides more than enough time for entropy to make the extremely minor added heat output irrelevant. Dumping the heat all at once though does not and in the summer that means higher room temps or necessitated AC.

What exactly do you even define as "light workloads" because intel is loosing in the provided charts in workloads using some thread and max threads.

For gaming I don't think a CPU using ~30-60 W more (assuming a 4090 with uncapped framerate, not a 60 FPS target) is going to kill the user from heatstroke and send their power bill to the moon, despite what you may want to imply with percentages.

No, that's what YOU are saying I said, trying to put words in other people's mouths to make your own argument look better.

This is a classic scarecrow argument because you argue in bad faith. I never once mentioned anything about a power bill, you assumed that as well.

Yes, the 80% consumer marketshare NVIDIA has or the 90%+ datacentre marketshare, where products for the past three generations have used this or a similar connector, yet where are the burned down houses?

2000 series uses 2x8pin, not 12VHPWR. "past three generations" is factually incorrect.

You are nuts if you think there need to be houses burning down for there to be an issue. That is absolutely an extreme requirement to even admit there is an issue.

The issue at hand is that there is significantly less safety tolerance. Hence why a single repair shop still receives 200 4090s with baked connectors / month.

Mind you, there's shouldn't need to be an obvious catastrophe when you can look at the numbers and see the connector is objectively less safe. Your policy towards safety is on the same level as Boeing apparently, it's not an issue until people die.

Your comment is factually wrong too BTW, ATX 3.0 is not just the PCIE5/12VHPWR cable, it's a thorough spec that requires units to tolerate spikes up to 200% of rated power etc, amongst many other things.

And so is your comment as well, it depends on this table:

Ironic isn't it.

A whole 1.4% of gaming performance lost by halving the power limit to 125 W, or 0.2% by limiting to 200 W. For those reading, that's how a non K Raptor Cove core CPU would perform.

I don't see a 14900, 14700, or other non-K 14th gen part in that chart Mr. "Um Technically". Power limited k parts are not equal to non-k SKUs due to the better binning of the K-parts.

One setting in the bios.

Yes and there's always that "one more setting" argument for any product. You could also set eco mode on an AMD GPU or all motherboard vendors have an option to both apply an undervolt and overclock in one setting.

It's moving the goalposts plain and simple.

Also important to note that the non K chips have different V/F curves, so they perform slightly better than simply power limiting a K part, since it's a different V/F curve, so these charts are understating the impact having non K chips on efficiency charts to contrast with their counterpart non X/X3D parts (tuned for efficiency out of the box).

Or they are worse because the K-parts are better binned. Hence why all your charts with K parts with limited wattage are not 100% representative of actual non-K parts. AMD had some 3000 series low end CPUs that exhibited worse power consumption due to the lower bin. You made the argument that CPUs like the 14900 and 14700 are more power efficient but even you couldn't find a review showing that data and you had the gall to call my omission intentional.

 

[dgianstefani]

I very much doubt they'll change the fact that X3D parts are more efficient.

See post #59. Voltage limited parts are more efficient. Whether that's due to X3D cache voltage intolerance, non X/non K variants or simply power limiting a K series chip, efficiency improves as voltage is limited. If the deciding factor for you is efficiency, power limit a 14900K to 35 W and you have the most efficient desktop x86 CPU on the planet, it takes about 30 seconds. But the reality is performance is what matters, and +/- 50 W on the CPU side of things when gaming isn't really what people care about, as marketshare would indicate.

Is it? My system is almost always running handbrake with other background programs? Your statement would also hold false for anyone gaming or streaming as well.

Ah, and you are representative of the average computer using person? Gaming is the minority of computer usage, PC gamers are the minority of gamers, people who game while running handbrake 24/7 (if they indeed exist) are the minority of PC gamers.

It's pretty much only valid for people who leave their system idling for long periods of time or only do light work. The problem with that is those workloads are going to have a small power consumption regardless of if you go Intel or AMD. The Intel might be 3 watts more efficient at idle but 60-100w worse in gaming. It doesn't take long for that 3w idle advantage to be completely dwarfed by the energy expended when you are actually using your PC's power.

3 watts huh? The blue bars are total idle system power draw. Seems more like 40 W if we don't make up numbers. Zen 4 drastically increased the minimum system power draw over Zen 3.

To emphasise my point, this chart shows a 13900K series Raptor Lake CPU system drawing 22-37 W less power than the non X Zen 4 parts, such as the 7600, 7700 and 7900, and 39 W less power at idle than the 7950X.

This idle state is where a typical PC will be for most of it's lifetime.

Single Thread load shows another 20-40 W difference (it's never just single thread, all cores are loaded if Windows is booted and software is loaded, they're just not at 100% synthetic load).

So, these Zen 4 CPUs (specifically the 7800X3D, excluding all others, especially the dual CCD chips, which are significantly less "efficient") turn out to be pretty decent at gaming and synthetic all core load efficiency, but pretty bad at idle/low load. So, very efficient at the states the average computer spends the least amount of time in, and very inefficient at the states the average computer spends the most amount of time in.

Furthermore, the average CPU is not a K series, nor an X3D chip, they're standard OEM parts, likely a laptop, but if in a desktop either a non K or a T series part.

It's because of reasons like this that the simplistic mindset "AMD CPUs are more efficient" irritates me.

 

[alwayssts]

I agree, however sometimes people want to see what the options are right now, and the thread has done a good job of showing those.

My attitude is new parts don't magically make the old ones slower, so a 4080 like you said is still a benchmark card for high end gaming, even if a 5080 releases in six months and is ~35% faster for the same price or whatever.

Hopefully the AI powered FSR is actually decent, and RDNA 4 doesn't flop like RDNA 3, some competent competition like when RDNA 2 was around would be nice.

Raptor Lake and the 7800X3D can already sustain 240 FPS and you're GPU limited most of the time, so I doubt new CPUs will really bring much to the table besides efficiency improvements, some AI stuff, improved platform etc.

Totally; I think the 4080 will be relevant for some time (until not only [RDNA5] next-gen consoles release, but they become the leading platforms). The problem has *always* been the price.

This is something nVIDIA has known since before it was released, and likely why it will continue to prop it up as long as they possibly can.

Eventually I'm sure we'll get a cheaper cut-down (16GB) GB203 (or next-gen) to compete...like the 4070 Ti Super but actually good-enough...but not until they've run their margins dry/AMD gains momentum.

I can't speak to how FSR will improve, although it surely will, but I *can* speak to how AMD *may* fix the RDNA3 arch with RDNA4.

You likely know. First, you likely have (4?/)8-bit precision units. Then it's bandwidth (likely greater cache if not memory speed), in which RDNA3 units can have up to a 20% performance improvement over RDNA2 using dual-issue when properly fed...and then fixing whatever the problem is with their pipeline (hotspots?) to sustain high-performance boost clocks (rather than 'efficient' clocks) on the process. It's completely achievable, but IMHO those cards are (likely) going to run hot and will probably use beefy coolers for a (relatively) small chip.

To me this makes sense, though. I've always thought they should create the smallest, most-efficient (spec-wise) chip that will cap out the max clock frequency of the high-performance libraries at 375w (2x8-pin).

Hopefully similar will be true with the half-N48 and ~200w (or whatever the threshold is for cheaper less-layer boards/components/coolers/lengths; it's somewhere around there) to compete with consoles.

Is this the most power-efficient when running at full load (which nobody actually does 24/7)? No. Does it make sense for most gamers (because cost/performance)? Yes.

As I've mentioned before, that is literally the lineage of ATi and why I loved them. Hopefully it shall return.

As for CPUs...yeah. I'm not particularly-enthused by Zen 5. The reason is, while there is an argument for a chip that is slightly faster than a 7800x3D, that's truly entering native 4k territory (for gaming)...

The rest of the ecosystem just ain't there yet (unless you're the guy buying a 5090, which ain't many). Hence I look forward to more-efficient chips that will fit that criteria when that time arrives...zen 6 or 7.

Hence why I said something like a 7700(x) being good-enough for most. If you want to spend more, or your particular niche situation requires more, I respect that...but that's not the general use-case.

Similarly, I think we need at least 3nm for a *great* 1080p hand-held device. I also wouldn't be sad if AMD's memory controller were much better and we had faster memory (be it cache, DDR6 or GDDR7) for APUs.

While Zen 5 may be a great architecture, it might be more than what most people need now but not reach the requirement (and certainly not efficiency) of what may be needed in the future...when it's applicable.

(edit: Changed the last sentence to be more clear/make more sense. )

 

[dgianstefani]

Totally; I think the 4080 will be relevant for some time (until not only [RDNA5] next-gen consoles release, but they become the leading platforms). The problem has *always* been the price.

This is something nVIDIA has known since before it was released, and likely why it will continue to prop it up as long as they possibly can.

Eventually I'm sure we'll get a cheaper cut-down (16GB) GB203 (or next-gen) to compete...like the 4070 Ti Super but actually good-enough...but not until they've run their margins dry/AMD gains momentum.

I can't speak to how FSR will improve, although it surely will, but I *can* speak to how AMD *may* fix the RDNA3 arch with RDNA4.

You likely know. First, you likely have (4?/)8-bit precision units. Then it's bandwidth (likely greater cache if not memory speed), in which RDNA3 units can have up to a 20% performance improvement over RDNA2 using dual-issue when properly fed...and then fixing whatever the problem is with their pipeline (hotspots?) to sustain high-performance boost clocks (rather than 'efficient' clocks) on the process. It's completely achievable, but IMHO those cards are (likely) going to run hot and will probably use beefy coolers for a (relatively) small chip.

To me this makes sense, though. I've always thought they should create the smallest, most-efficient (spec-wise) chip that will cap out the max clock frequency of the high-performance libraries at 375w (2x8-pin).

Hopefully similar will be true with the half-N48 and ~200w (or whatever the threshold is for cheaper less-layer boards/components/coolers/lengths; it's somewhere around there) to compete with consoles.

Is this the most power-efficient when running at full load (which nobody actually does 24/7)? No. Does it make sense for most gamers (because cost/performance)? Yes.

As I've mentioned before, that is literally the lineage of ATi and why I loved them. Hopefully it shall return.

As for CPUs...yeah. I'm not particularly-enthused by Zen 5. The reason is, while there is an argument to be for a chip that is slightly faster than a 7800x3D, that's truly entering native 4k territory (for gaming)...

The rest of the ecosystem just ain't there yet (unless you're the guy buying a 5090, which ain't many). Hence I look forward to more-efficient chips that will fit that criteria when that time arrives...zen 6 or 7.

Hence why I said something like a 7700(x) being good-enough for most. If you want to spend more, or your particular niche situation requires more, I respect that...but that's not the general use-case.

Like-wise, I think we need at least 3nm for a *great* 1080p hand-held device. I also wouldn't be sad if AMD's memory controller were much better and we had faster memory (be it DDR6 or GDDR7).

While Zen 5 may be a great architecture, it might just be be "better" than what most people need now, but not reach the potential (and certainly not efficiency) of what can be done in the future.

I mean, according to the latest Steam Hardware survey, 1% of Steam gamers have an RTX 4090. Small percentage I know, but that's still 1.32 million cards out there (132 million monthly active Steam users). 8/12 GB cards dominate, and 16/24 GB cards are less than 5% combined.

The mid range xx60/xx70 class cards dominate, but are still significantly faster than console GPUs in terms of framerates/comparable quality settings.

Optimum Tech did a video on this recently, an RTX 4060 PC, while in up front costs is slightly more expensive, is much, much faster than an Xbox Series X, which is arguably the console with the most GPU horsepower.

RTX cards are the vast majority within PC gaming it seems, so if AMD want to disrupt that they need to offer something appealing, with software that is competitive, not just "worse but 10% cheaper".

The first dedicated AMD card on the list, the RX 580, is about as popular as an RTX 4080, which is one of the least popular, and most expensive RTX Ada cards.

 

[64K]

I mean, according to the latest Steam Hardware survey, 1% of Steam gamers have an RTX 4090. Small percentage I know, but that's still 1.32 million cards out there (132 million monthly active Steam users). 8/12 GB cards dominate, and 16/24 GB cards are less than 5% combined.

The mid range xx60/xx70 class cards dominate, but are still significantly faster than console GPUs in terms of framerates/comparable quality settings.

Optimum Tech did a video on this recently, an RTX 4060 PC, while in up front costs is slightly more expensive, is much, much faster than an Xbox Series X, which is arguably the console with the most GPU horsepower.

RTX cards are the vast majority within PC gaming it seems, so if AMD want to disrupt that they need to offer something appealing, with software that is competitive, not just "worse but 10% cheaper".

The first dedicated AMD card on the list, the RX 580, is about as popular as an RTX 4080, which is one of the least popular, and most expensive RTX Ada cards.

View attachment 348501

You have to be careful how much info that you take from the Steam Survey. As a percentage of members it's alright to a degree but bear in mind that Steam doesn't survey everyone. It's a random sample and It's voluntary. If you are chosen then you will be asked when you logon if you would like to participate in the survey. Choose yes and Steam will search your PC automatically for hardware. Choose no and it won't. There is a way to force Steam to let you participate but hardly anyone knows about it or does that. So, a percentage doesn't equate to any actual amount of GPUs being used on Steam.

In all the many, many years that I have had a Steam account I have only been asked 3 times to participate and only 1 time I let it and 1 time I forced it just to see if it would happen and it worked.

 

[dgianstefani]

You have to be careful how much info that you take from the Steam Survey. As a percentage of members it's alright to a degree but bear in mind that Steam doesn't survey everyone. It's a random sample and It's voluntary. If you are chosen then you will be asked when you logon if you would like to participate in the survey. Choose yes and Steam will search your PC automatically for hardware. Choose no and it won't. There is a way to force Steam to let you participate but hardly anyone knows about it or does that. So, a percentage doesn't equate to any actual amount of GPUs being used on Steam.

In all the many, many years that I have had a Steam account I have only been asked 3 times to participate and only 1 time I let it and 1 time I forced it just to see if it would happen and it worked.

That's true. But sometimes you have to work with the data you have and I'm not aware of any other data that breaks down GPU market share as clearly or with Sean's scale. I don't consider mindfactory sales to be representative for example, but just about every pc gamer uses Steam.

*Steam's scale, lol

 

[kapone32]

That's true. But sometimes you have to work with the data you have and I'm not aware of any other data that breaks down GPU market share as clearly or with Sean's scale. I don't consider mindfactory sales to be representative for example, but just about every pc gamer uses Steam.

*Steam's scale, lol

That is funny considering that the only PC Steam registers for my hardware is my 3060 laptop. In what world have Steam charts become reliable? Maybe when you are trying to pigeon hole a point. I mean it includes Intel IGPUs in the middle of the chart you posted.

 

[dgianstefani]

That is funny considering that the only PC Steam registers for my hardware is my 3060 laptop. In what world have Steam charts become reliable? Maybe when you are trying to pigeon hole a point. I mean it includes Intel IGPUs in the middle of the chart you posted.

Do you understand what a representative sample is?

If you poll, for example, 10% of a very large population, at random, you get pretty accurate data that's representative of the whole population.

Steam can't use data from 100% of users since data protection laws exist, hence you have to approve the survey for it to gather your system information.

The survey that includes Intel and AMD iGPUs is accurate. Despite what enthusiasts may think, the vast majority of PCs do not have dedicated graphics cards. Or would you prefer the data wasn't shown accurately?

 

[Dr. Dro]

That is funny considering that the only PC Steam registers for my hardware is my 3060 laptop. In what world have Steam charts become reliable? Maybe when you are trying to pigeon hole a point. I mean it includes Intel IGPUs in the middle of the chart you posted.

Protip if you want to boost the AMD numbers on Steam a bit. Won't change the fact that the Radeon is effectively worthless, though... at least it serves the purpose of offloading general processing and saving precious video memory for my 4 GB 3050...

 

[kapone32]

Protip if you want to boost the AMD numbers on Steam a bit. Won't change the fact that the Radeon is effectively worthless, though... at least it serves the purpose of offloading general processing and saving precious video memory for my 4 GB 3050...

View attachment 348559

Thanks for your opinion

 

[Dr. Dro]

Thanks for your opinion

I'm surprised you have an Nvidia GPU on your laptop, considering how adamant you were about the AMD Advantage™ brought by Ryzen | Radeon

Time to play some games the way they were meant to be played™

 

[Bobaganoosh]

I have not read this entire thread, so please excuse my ignorance.

Now, truly, is a bad time to build a PC. Computex is literally in a couple weeks, in which the whole landscape/thought-process may change depending upon what happens there (and pricing of new/old parts).
...

I mostly agree with this post, but with a couple caveats or side points.

As far as rumors go it's a complete crap-shoot for what's going to launch in the next 6-8 months.
Nvidia:
Only the 5090 is launching this year (and maybe Decemberish), with 5080 likely ~January. No idea when/if other skus will launch and they may just say "well 4060-4080 are still good so what's the rush?"
AMD:
I have no idea what AMD is launching, but it's a good bet they will want to be price competitive with it as they're supposedly not going after the 5090 (it's unclear if it'll even be 4090 level performance...again based on rumors only). I haven't heard a real timeline for launch of the next CPUs. As is typical with AMD launches, the projection rumors are aaalllll over the place. My gut feeling is that there's really not much there for 1440p+ gaming. I'm hoping they have a better IHS design and that lets them boost higher with better thermals...and if they can improve stability and memory performance, there may be a reason to hold out for them, but that's a few if's.
Intel:
It sounds like Battlemage is a few months out (maybe?), but it's quite unclear where performance or price will be. Seems people are guessing ~4070 range. Their CPUs are a total mystery at this point. same core-count as 14900k for top-sku, but no hyperthreading and lower clocks. Will it be substantially better than 14th gen for gaming (especially at 1440p+) and worth paying top dollar for? Probably not. I'm betting similar to 11900k where if you OC'd it, it would win most gaming benchmarks over 10900k, but not by much...and only if you OC it. 14900k can't really OC much if at all, so I imagine the new sku might have a slight edge. Worth waiting for? Not in my opinion.

So there are some times where it really does behoove you to wait and in some previous years, it was a real benefit because they replaced skus and the old ones got price drops so if there were still previous-gen cards on the shelf, you could get some great deals. The last few years they've been completely killing inventory before launching new cards OR leaving the old prices and bringing in new cards at higher prices. So the price/performance hasn't moved and top skus are just more expensive. This all reinforces what I've believed 90% of the time: The right time to build is when you want to use it. There will always be new skus "just around the corner" if you adjust the corner definition lol. I'd say we're in a pretty great place where most of next gen looks pretty skippable as of today and it's not even coming for several months. The only thing definitely coming soon are announcements.

 

[dgianstefani]

The right time to build is when you want to use it. There will always be new skus "just around the corner"

Agreed.

However, I'd still wait for Arrow Lake personally, there's several very interesting process technological leaps for it, even if FPS is the same, which I doubt, the platform itself and the way the CPU works is changing quite a bit. For one thing security and simplicity of threads seem to be a focus due to the lack of HT.

Even if it's a mild upgrade, 13-14th gen will become cheaper, so it's really win-win to just wait. Zen 5 won't be revolutionary IMO, just typical gen on gen improvements, Zen 6 will be interesting if rumours hold true. Not like Intel moving in one generation from monolithic to disaggregated, introducing backside power delivery (huge), no HT, tiles on foveros in it's third iteration finally on desktop, DDR5 only memory controller, new process node, 3MB per P core compared to 2 MB cache on Raptor, RibbonFET (huge). Leaks say 6-21% faster, so without HT, that's pretty significant per core improvements considering the frequency is also dropping by ~5%.

TBH though the Lunar Lake SoC, though simpler than Lakefield, which had DRAM stacked vertically on top of the CPU/IO etc dies with foveros, six months before M1 was released, the LPDDR5X 8500 or whatever 16/32 GB being on package will be significant for efficiency and latency. Apparently it's competitive/faster/more efficient than current Zen/ARM designs in power efficiency, despite the RAM power budget being included with the SoC, rather than a separate budget for those comparison chips.

So Arrow Lake is somewhat conservative compared to the laptop chips, but that's not unusual, Intel usually does some demo chip low power on laptop first, then a full range of SKUs for low/mid power laptops, then desktops, maybe 16th gen will introduce some of the more cutting edge stuff seen with Lunar Lake for desktop. Interesting times ahead. 16th gen vs Zen 6 will be a fascinating duel IMO, and I expect Arrow Lake will perform better than Zen 5, with X3D chips perhaps taking the lead in gaming, but being behind in productivity, as with 14th gen vs Zen 4.