NVidia GeForce RTX 40x0 cards - x2 the power consumption, x2 the performance

Kirun

Buzzfeed Editor
<Gold Donor>
18,671
34,770

One of many threads discussing the issue here, seems to be training with the first several boots. Seems like latest BIOS along with these settings had positive effects:

Fast Boot ENABLED

Memory Context Restore AUTO

Boot time: 12 seconds

I haven't noticed my boot times being long after the first few weeks but I spent a long time on my old machine before I switched over to this one full time.
With the 7xxx series of video cards being pretty dogshit and Intel pretty much catching up/surpassing AMD in performance again, I think I'm done with them for a while. Until 2021 when I built my 5900x system, I had last used AMD around 2002ish?(whenever Athlon was relevant). I went fully AMD and while the 5xxx series of CPUs smoked Intel at the time, the little "quirks" that AMD has really just annoy the fuck out of me. I'm too old to be an enthusiast with this shit anymore - I just want it to work and perform well out of the box.
 

Big Phoenix

Pronouns: zie/zhem/zer
<Gold Donor>
44,590
93,109
For better or worse, absolutely not.


I Dont Arrested Development GIF
Something more appropriate long term;

1685238120069.png

I went fully AMD and while the 5xxx series of CPUs smoked Intel at the time, the little "quirks" that AMD has really just annoy the fuck out of me.
I had absolutely zero issues with my 3700x and 5800x systems. Rock solid even with upgrading the bios to go from a 3700x>5800x on the same mobo.

Ive been very satisfied.
 
Last edited:
  • 2Solidarity
Reactions: 1 users

Kirun

Buzzfeed Editor
<Gold Donor>
18,671
34,770
I had absolutely zero issues with my 3700x and 5800x systems. Rock solid even with upgrading the bios to go from a 3700x>5800x on the same mobo.
To be fair, my 5900x has been solid too, sans a DOA Gigabyte motherboard on my first attempt(swapped to ASUS and it was fine). I just get frustrated by shit like Ryzen Master, their chipset drivers, etc. And don't even get me started on what an utter pile of shit Adrenaline is compared to GeForce Experience.
 

Blitz

<Bronze Donator>
5,660
6,178
Something more appropriate long term;

View attachment 475376

I had absolutely zero issues with my 3700x and 5800x systems. Rock solid even with upgrading the bios to go from a 3700x>5800x on the same mobo.

Ive been very satisfied.
I put Hitler and Nazi ones on mine, and watch the puzzled looks I get when people visit, and I show them my office setup. It's just small enough, there's a level of "am I really seeing this" when people notice the heat sink.
 
  • 1Hodjing
  • 1Worf
Reactions: 1 users

Big Phoenix

Pronouns: zie/zhem/zer
<Gold Donor>
44,590
93,109
So I went back and checked the power draw on my old system taking pictures because I thought I typed the wrong numbers

Old system = 5800x, x570 atx mobo, 2x8gb memory, 3080ti(not undervolted or ocd) and a 850watt evga gold psu.

New system = 7800x3d, b650 micro atx mobo, 2x16gb memory, 4080 and a 850 watt evga titanium psu.

End result is a 250-300 watt reduction in power consumption for "demanding" games;

1685253423214.png


Im genuinely shocked. I figured a 100-125 watt reduction.
 
  • 2Like
Reactions: 1 users

Daidraco

Golden Baronet of the Realm
9,196
9,305
So I went back and checked the power draw on my old system taking pictures because I thought I typed the wrong numbers

Old system = 5800x, x570 atx mobo, 2x8gb memory, 3080ti(not undervolted or ocd) and a 850watt evga gold psu.

New system = 7800x3d, b650 micro atx mobo, 2x16gb memory, 4080 and a 850 watt evga titanium psu.

End result is a 250-300 watt reduction in power consumption for "demanding" games;

View attachment 475397

Im genuinely shocked. I figured a 100-125 watt reduction.
At 14.2 cents per kwh, I felt like my power bill rose about 19 bucks on average with a 3090 at the time. The 4090/13900 seems like it pulls about the same, or just within a few dollars less. (At least this is how Ive been watching wattage.)

I installed Cyberpunk just to see the frames drop myself, and ya - that game with max everything still kicks this computers butt a bit. But nothing was scary about the temps really. What was scary was letting CinebenchR23 push the system for a few rounds and for fucks sake, this 13900 shot up to around 100C. I was freaking out thinking my deepcool was fucked, or I did a shitty job with thermal paste - but no, there are other people reporting the same issue. That type of temperature on my gear just makes me have an anxiety attack, regardless. End result was just over 42k points, but uhh....
 

Kajiimagi

<Gold Donor>
1,325
1,699
I still like building my PCs but times have certainly changed. My first PC build was a Celeron 300a when I was in HS and when I picked out all the parts, I had no idea overclocking existed. My motherboard was some Intel 440bx with absolutely no overclocking options. Then after stumbling across HardOCP for the first time my eyes nearly popped out of my skull: Simply apply nail polish to 2 pins on the Celeron's slot card and it would tell the motherboard to run the bus at 100mhz instead of 66, giving me an instant 50% overclock for free. I couldn't believe it worked and it was perfectly stable, I had built a budget PC and I got something that was pretty much as fast as a Pentium II 450 or even faster in games because of the faster cache.

Those days are long gone, pretty much every CPU on the market these days has already been redlined so the only reason to keep building them I think is nostalgia and quality assurance.
Same. My 2nd PC was a pentium pro and I read in a magazine ( I think boot) how to overclock it. Could not believe that changing a few settings gave an immediate boost for 'free'. From then on I overclocked everything that was mine. FF to my current rig which is the 1st one since the early 90's that I didn't overclock as I didn't see the point. With 16 cores what I read was it's actually slower to OC all of them.

Also my wifes current PC, M2 drive, on board video (she just websurfs). It was boring to build.

Weird times
 

Kirun

Buzzfeed Editor
<Gold Donor>
18,671
34,770
So I went back and checked the power draw on my old system taking pictures because I thought I typed the wrong numbers

Old system = 5800x, x570 atx mobo, 2x8gb memory, 3080ti(not undervolted or ocd) and a 850watt evga gold psu.

New system = 7800x3d, b650 micro atx mobo, 2x16gb memory, 4080 and a 850 watt evga titanium psu.

End result is a 250-300 watt reduction in power consumption for "demanding" games;

View attachment 475397

Im genuinely shocked. I figured a 100-125 watt reduction.
How is that even possible? Or is it just because the updated components have to "work less" for the same relative performance?
 

Mist

Eeyore Enthusiast
<Gold Donor>
30,380
22,150
How is that even possible? Or is it just because the updated components have to "work less" for the same relative performance?
The new stuff has slightly lower max power draw but a LOT lower median power draw.

That said, I think running a 7900X in Eco Mode would be better in most cases than a 7800x3D.
 

rhinohelix

<Gold Donor>
2,867
4,669
As I understand it, IANAE, as they reduce the process size they use to make the chips, they are able to pack more and more transistors into the same area, which is one of the ways they are able to increase performance; the electrons have shorter distances to travel across less trace and less resistance, Less resistance means less heat waste and thus they use less power individually. Often times manufacturers pack more on a chip (and more chips on a wafer) and clock them higher, so cumulatively power draw can get higher but depending on how you design them you can also make them much more efficient. So one can end up with both ends of the spectrum: Much more efficient cards like the 4080 and less efficient cards that are going for all out performance like the 4090. When the process was new, they struggle with it, as they did with the 3080/3090 or the design was new, as RNDA2 tried to use chiplets for cards for the first time, IIRC.

Second revision cards can be better like the 1080Ti/2080 Super, as the card makers refine the process and get better yields for the chips and improve, as AMD did with the 6950XT
 
  • 3Like
Reactions: 2 users

Mist

Eeyore Enthusiast
<Gold Donor>
30,380
22,150
As I understand it, IANAE, as they reduce the process size they use to make the chips, they are able to pack more and more transistors into the same area, which is one of the ways they are able to increase performance; the electrons have shorter distances to travel across less trace and less resistance, Less resistance means less heat waste and thus they use less power individually. Often times manufacturers pack more on a chip (and more chips on a wafer) and clock them higher, so cumulatively power draw can get higher but depending on how you design them you can also make them much more efficient. So one can end up with both ends of the spectrum: Much more efficient cards like the 4080 and less efficient cards that are going for all out performance like the 4090. When the process was new, they struggle with it, as they did with the 3080/3090 or the design was new, as RNDA2 tried to use chiplets for cards for the first time, IIRC.

Second revision cards can be better like the 1080Ti/2080 Super, as the card makers refine the process and get better yields for the chips and improve, as AMD did with the 6950XT
Right, a lot of power efficiency is also determined just by how the firmware and drivers are programmed, not the silicon itself. In the case of the 30XX series, the last ~30% of wattage only represented the last 5% of performance. But because that wattage made a wider range of chips (quality is an RNG lottery, basically) stable at higher speeds, that's what they were programmed for.

On the 4080, for instance, the behavior is very different. More wattage allows for almost no additional performance, so the card is just set to run at something much closer to the optimal performance-per-watt curve. The 4080 is by far the most power-efficient card, but hilariously, one of the least price-efficient, until you factor in total lifetime cost of ownership.
 
  • 1Solidarity
Reactions: 1 user

rhinohelix

<Gold Donor>
2,867
4,669
Right, a lot of power efficiency is also determined just by how the firmware and drivers are programmed, not the silicon itself. In the case of the 30XX series, the last ~30% of wattage only represented the last 5% of performance. But because that wattage made a wider range of chips (quality is an RNG lottery, basically) stable at higher speeds, that's what they were programmed for.

On the 4080, for instance, the behavior is very different. More wattage allows for almost no additional performance, so the card is just set to run at something much closer to the optimal performance-per-watt curve. The 4080 is by far the most power-efficient card, but hilariously, one of the least price-efficient, until you factor in total lifetime cost of ownership.
Nvidia decided at the end of the scapler profit-taking that why should middlemen make all that extra money, so they built in shortage pricing to the MSRP. Unfortunately for them, other than a few% at the top who will indeed pay anything, the market will not hold those prices. The people that are paying those prices now have always paid those prices, relatively and in some cases absolutely speaking: 4090's are just 2023's Titan, top of the line cards, Modern Triple monitor/SLA rig setups.

The market for 1% will probably always be there but its pretty small. AMD had a huge opportunity, has a huge opportunity if they come back with a 7950XTX card for 1200 bucks that is competitive with the 4090.
 
  • 1Like
Reactions: 1 user

Big Phoenix

Pronouns: zie/zhem/zer
<Gold Donor>
44,590
93,109
How is that even possible? Or is it just because the updated components have to "work less" for the same relative performance?
So it appears part of the huge disparity was performance being left on the table. I finally updated to the latest BIOS which officially added support for the 7800X3d for this mobo and power consumption increased, along with performance.

BF2042 went from averaging 325watts to 385watts. Its pushing more FPS though, timespy score went from 22841 to 24377 and the CP2077 benchmark went from 118 to 140. Still a huge reduction, just a little less huge with the 7800X3d running at full performance.

Memory Context Restore is something you may have, and will probably have to dig for in the BIOS to enable. Probably not turned on by default since it can introduce instability.

I tend to sleep my computer and only shut down by accident or when installing some hardware update, so it's not as big of a deal for me, but I notice the start up time is definitely longer while it checks the memory when set to EXPO.
Sadly the update didnt seem to fix the boot hanging. Still having to wait 1-2 minutes for it to POST even with turning memory context restore on.
 

Palum

what Suineg set it to
23,359
33,427
So it appears part of the huge disparity was performance being left on the table. I finally updated to the latest BIOS which officially added support for the 7800X3d for this mobo and power consumption increased, along with performance.

BF2042 went from averaging 325watts to 385watts. Its pushing more FPS though, timespy score went from 22841 to 24377 and the CP2077 benchmark went from 118 to 140. Still a huge reduction, just a little less huge with the 7800X3d running at full performance.


Sadly the update didnt seem to fix the boot hanging. Still having to wait 1-2 minutes for it to POST even with turning memory context restore on.
That sounds defective to me, at this point any boot times over 3 seconds feel pointlessly slow.
 

rhinohelix

<Gold Donor>
2,867
4,669
That sounds defective to me, at this point any boot times over 3 seconds feel pointlessly slow.
Please. Google "AM5 long boot times" and you can see what is going on. The AM5 platform since it's launch has a history of doing this, the question is how does he break his machine out of the training cycle get it into a more normal time. That's not to say that there isn't something that ultimately isn't wrong with it but just flipping the table over if its more than three seconds with almost no effort and throwing up one's hands is so mid.

This is the guy in War games who doesn't get up from his workstation and shouts across the room "Put X in the Center square. Oh well, we're all going to die." and goes back to playing Galaga.

I know, mixing metaphors.
 

spronk

FPS noob
22,592
25,632

kinda cool it uses basically a second pci slot but has a custom PCI-connector so it only works on one single motherboard right now lol
 
  • 1Like
Reactions: 1 user

Mist

Eeyore Enthusiast
<Gold Donor>
30,380
22,150
DDR5 typically trains very slowly, but it should only happen on cold A/C power boots, not normal power off states. Normal boots are a bit longer than my previous DDR4 motherboards, but only the boots after I first plug it in after moving it or maintenance does it power-cycle a bunch of times to train the memory.
 
  • 1Like
Reactions: 1 user

jayrebb

Naxxramas 1.0 Raider
13,911
13,765
Memory reading at 49000 with 64gigs plugged in. What's the best way to diagnose which stick is the problem?

Or just don't bother and call warranty on the two sticks?
 
  • 1Like
Reactions: 1 user