Any PC Builders Out There?

The 5090 is, over the benchmarks I've seen, 27% faster than a 4090. It sounds fantastic, until you realize it has 25% more cores and 30% more TDP than the 4090. There's near zero uplift in terms of per core or per watt performance. All for at least $2K USD, with early leaks showing AIB cards starting in the $2300 range.

nVidia is going to sell piles of them, but much disappoint in my opinion.

 

Does whoever put out that chart say if any frame gen is on for those results? I'm assuming no, yeah?

 

It's raster performance across 17 games.

You can't empirically compare the 4000 and 5000 series with frame gen on. The 4000 series can only use up to DLSS 3 and its variants, while the 5000 series can use DLSS 4.

Also the story is even worse, that 27% is at 4K. At 1080P and 1440P the card bottlenecks behind both the 9800X3D and 13900K with under 10% improvements, if any.

Something weird is also happening in some results, where AMD or Intel CPUs, but not both, win for no apparent reason. The early thoughts are that the nVidia drivers are not optimized game by game yet, which is not a good sign going forward. It could mean current 5000 series only DLSS 4 performance is being prioritized by studios accepting nVidia sponsorships and help. It's possible DLSS 4 is the priority nVidia is pushing over standard raster performance or past DLSS versions.

 

Right, that's why I asked, just to make sure DLSS 4 wasn't skewing it and I was reading it correctly.

 

LTT and a few others put out good frame generation graphs. Basically, with the new way frames are generated, it's completely possible to quadruple your base frame rate by turning DLSS on and frame generation to maximum. However, ghosting is a problem with all games showing artifacts. What's funny is DLSS 4 looks like it has FSR problems. With the new model being a simpler algorithm, moving objects with edges like fans gain extra blades at random, or flash or sparkle as they appear and disappear. While the new models do allow for better frame generation gains, they seem to sacrifice accuracy at lower frame rates. Given the only time DLSS and FSR make sense is when the frame rate drops below 60 fps, it's disheartening to see nVidias bright new solution has significant issues that get worse as the native frame rate drop, much like FSR 1 and 2.

 

Funny story, nVidia removed the hot spot temperature monitoring for the 5090. Most suspect it's because the card can easily hit 95c when pushed for hours, der8auer had his card hit 40 dB on fan nose and this temperature without overclocking. If the hotspot was shown, it would probably be in the 110-115c range and people would snap lol.

This might be the first FE card to skip. It doesn't seem like the flow through design is enough to compensate for being a 2 slot card trying to cool a 600 watt GPU.

 

Jesus.

Igor's labs noticed the transient spikes on the 50 series, which the 30 series was known for and the 40 series mostly did not exhibit, and decided to dive deep into the actual surge draw.

The 5090, with no overclock, will surge draw over 900 watts. That's right, if you want to be completely safe, you shouldn't put this card behind anything with less than 1200 watts of power, assuming your CPU and other components draw under 300 watts. If they don't, 1600 watt monsters may be the new requirement for top end cards.

Also how did nVidia mess this up? The 3090 would occasionally throw a ms of 600 watt draw which was crazy at the time, but the 4090 didn't. It rarely broke 500 watts surge on a 450 watt TDP. Now we have a 600 watt TDP card that wants a little over 900 watts frequently enough to trip 1000 watt PSUs? Crazy.

 

.....

 

I'd be happy with a 4080 super, 4090 or a 5080. Ideally under $1k lol. I'm not dropping $2k on a graphics card. Just no.

 

I'm really not into the whole RGB color scheme on every component either. Why does the entire build need to look like a carnival wheel at cheap circus?

 

This tested at the PSU, running cyberpunk 2077 at 4k.

 

As a first time builder, I'm going to answer this question honestly. Never having built a PC before, I wanted two things from my machine: a little more performance than an xbox series S, and a badass desk ornament that shined for all to see. For my build, I wanted it to be entirely obvious that it was hand-built by me alone, looked nothing like a store-bought tower, and I wanted the satisfaction of knowing it was "unique" and "customizable." I wanted it to be loud. I wanted it to shine brighter than the monitor. I wanted it to be something more than just a tool; maybe more like a RGB covered Snap-on tool

After making the switch from console to PC entirely and spending a couple years with this machine, I can understand where you're coming from now. Before, I wanted the aesthetic to be self-fulfilling, it was a big factor, and performance was just the not-so-fun part I had to manage when researching parts. Being on this side of ownership, I can now understand how LED's everywhere can be a real hassle, an unneeded bump in cost, and just something else to mess up. While I absolutely love my LED's still, I've never been a fan of anything other than static colors, so I don't know that I'll ever understand the unicorn vomit themes. If I were to build again, I'd keep in mind just how little the LED part of the tool means to me now. I'd keep in mind that a giant LED lit case with 100 fans may not be the move for a tower I'd want to mount on a sim rig shelf or anywhere other than on top of a desk, out of sight.

Til then, I'm just gonna be lost in the pool of orange LED sauce

 

Geeking over the mounting method you have for the wheel and shifter. Thats sick. Was always my gripe when I was using a wheel a lot, and now I’ve been limited by storage space for last few years but I’m sure I’ll be back to it once I move.

that and the riser for the pedals… never thought of that to get a more low down feel

 

Thanks! I'd love to have a dedicated "rig"/extrusion chassis to mount it all to, but don't really have the space. This was the best solution I could come up with. Its just a piece of extrusion bolted to the frame of my desk, with some smaller pieces of extrusion to house the wheelbase/shifter/ebrake. I essentially created a track system using some quick release type handle fasteners to be able to move the wheelbase or shifter/ebrake down to the far end of the desk when not in use. The pedal platform was also just a scrap wood project to get the pedals to a more realistic orientation for driving. I can recline my chair back one notch and it pretty much feels like a Toyota Corolla lol. The wheel being quick detach, I mount it on the opposite end of the track when not in use.

 

Tested using the nVidia official power draw kit. It measures draw at the PCI-E slot and PCI-E 12V connection. nVidia started selling the kits to reviewers a few years ago and they're very accurate. A few reviewers have done third party testing on them. Without them, seeing transient spikes like this is tough, until the PSU is tripped.

 

No, I'm saying that the graph you shared is peak power, that's a full system load analysis (the test bench equipment is included)

The 901-watt spike, came from the power consumption analysis during native testing for cyberpunk 2077 but it's still full system load, not component level, the blue indicating gpu consumption. (I edited this just to reference the difference)



It should also be noted that transient spikes are completely normal due to transistors constantly switching states. When you have 92billion transistors switch their states, a power spike should be expected this was true in the 30 series as you referenced, the 40 series was also no stranger to this. The difference was atx3.0 came out two years after the 30 series launched, so we were left with just having to have a overbuild psu's to absorb the spikes. ATX 3.0/3.1 psu have up to 200% excursion protection at 0.1ms during a 10% duty cycle, 180% for a 1ms at 20% (any psu more than 450w using the stupid 12VHPWR cable (which i know blackwell is on 12v2x6 so 650w I believe.) Even while gaming you most likely won't be running 100% duty cycles (maybe, I dunno) so many things switching on and off. That's when you que AI tech though, which would probably allow lower duty cycles since it's not constantly processing data (rasterizing)

In the end, Wallossek even mentions that a 1000w psu is sufficient to run the 5090, now that's based on his opinion running a 9800x3d...I think the intel chips are getting more power hungry so it's possible to actually want a 1200w psu for that but damn if you still usin an intel chip, just damn.

I'm gonna generalize here (I love doing that) overclocking for the most part (for most gamers) has been pretty dead in the water, in my experience the 30 series performed far better by undervolting which would also reduce the transient spikes, I can't imagine that the 5090 is gonna be much different in the regard.

Again, this is native. Even the consumption analysis for mfg and dlss, you don't see these spikes (or I guess that spike)

EDIT: This isn't really apples to apples here, as I'm not certain what a gaming loop is but this is the consumption for the 4090, it's quite similar to the 5090 in regards to power spikes. Where it's just a millisecond give or take some microseconds. So, I don't really think nvidia messed that up. Both are leaps and bounds improved upon the 30 series.

 

Ah. Makes sense. The social media I got the image from had it broken down completely wrong then and so did I. That's on me. I saw an image and did a double take at the 900 watts without digging farther than the tweet.

I still don't trust the new power connector. While I agree consumer error was the main cause of issues, the fact that so many issues have happened tells me the connector should never of been released as it was. Time will tell if the new changes will actually fix the installation issues. Personally, I wish nVidia would let AIBs use 8 pin classic connectors. Currently, nVidia dictates that the 12 pin must be used, which at up to 580 watts of stock draw is a lot of power in a small connection.

Undervolting as the main means of tuning has been the trend for a while. nVidia has been okay for undervolting, and you can get some crazy low numbers on AMD cards. Everything is being dialed up to 11 now from factory. Turning the power down by 20% of CPUs while sacrificing under 5% performance is the new normal.

I'd argue the 50 series appears to really be an evolution of the 40 series, for now. Maybe we will see the new architecture of the 50 series shine in a few years as new features are better utilized. My opinion is that nVidia is asking a lot for an upgrade that isn't that good. If I had a 3090, and I didn't need to upgrade, I'd consider waiting another generation.

 

I caved. Installed a 9800X3D today when my buddy, who works sales at a new local computer place, called me and said he had one with my name on it if I wanted it lol.

 

Microcenter supposedly has a 25 in stock for in store pick up only. Unfortunately they have no stores in AZ. Still waiting on restock...

 

If the rumors are true, and who really knows, they should be coming in stock every week from this point. I can't say how many, but the store near me got a sizable amount this time and they were told to expect similar amounts every 7-14 days.

I will also say my buddy told me they've sold one, just one, 200 series processor this year. Everything going out the door is AMD. I don't think we've ever seen Intel have it this rough in the PC builder market. Obviously they still dominate the OEM's, but it's not a good sign that everyone is going AMD this generation.