GPUs & CPUs & Enthusiast hardware: Questions, Discussion and fanboy slap-fights - Nvidia & AMD & Intel - Separe but Equal. Intel rides in the back of the bus.

[Archeops]

Along with still stuck at 8gb of vram, the Nvidia laptop options continue to fall behind desktop.

Yeah, that's why I'm using my current 4070-powered ASUS laptop for light gaming and why I built a 4070 Ti Super-powered brick shithouse for a PC.

 

[Fcret]

Some people are calling the RTX 5070 Laptop GPU a scam since its computing power is clocked at 3-.8 TFLOPS compared to the RTX 4070 Laptop GPU's 29.1 TFLOPS.

Blackwell is on a very slightly refined process node over Lovelace which allows them to get a little bit more density while utilizing significantly more power for performance gains. I don't expect the laptop chips will be much better than last gen considering they're power-constrained by the laptop form factor rather than the actual physical limits of the silicon.

The point of the laptop 5070 probably has more to do with the fact that they're going to stop producing the laptop 4070 and so it will eventually be the only thing available in the tier. Outside of the xx90 stuff, prices haven't gone up so much of blackwell looks to be direct SKU replacement with some minor improvements rather than actual generational upgrades.

 

[Pascal's Wagie]

I see a fellow ultrawide enjoyer, I'm finding it a little cramped with 3 or more windows up, thinking of upgrading to a 5120x1440 res what do you think?

Two of my three monitors are 21:9 ultrawides, stacked one atop the other with monitor arms (Dell U4025QW on top / Gigabyte M34WQ on bottom), with a 16:9 set to portrait mode to the side. I've sat at a buddy's desk with his duper-puper-ultrawide (more than 21:9) and it was just 2wide4me. The stacked Dell and Gigabyte spread everything out more evenly in my field of view, and would be enough even without the extra portrait mode monitor. Even if you do go with a 5120x1440, see about keeping your current ultrawide in your setup to fill up the vertical space in your view.

 

[barrow greeter]

All signs are that AMD is going to be as aggressive as possible with pricing this generation, although the peak performance of RDNA 4 can't be confirmed yet, but if that leak showing RT performance on par with 4070 Ti is true this is going to be pretty epic.

I still hate the new naming scheme though and think it's retarded to copy Nvidia like that.

 

[4090 Chad]

All signs are that AMD is going to be as aggressive as possible with pricing this generation, although the peak performance of RDNA 4 can't be confirmed yet, but if that leak showing RT performance on par with 4070 Ti is true this is going to be pretty epic.

I still hate the new naming scheme though and think it's retarded to copy Nvidia like that.

I’ve learned my lesson one to many times falling for AMD hype and expecting them to price their cards truly competitively.

I need to see it to believe it.

 

[barrow greeter]

I’ve learned my lesson one to many times falling for AMD hype and expecting them to price their cards truly competitively.

I need to see it to believe it.

not too long ago RDNA 2 the hype was real though, AMD RTG (marketing and pr especially) rarely misses the opportunity to fuck up a product launch, but sometimes they can still surprise us

 

[geckogoy]

I don't understand why AMD isn't going high end since they ostensibly are doing GPU chiplets, high end should be the one area they obviously have an advantage. Makes me think NVIDIA has pictures of someone or something and AMD isn't actually playing their side of the field.

 

[barrow greeter]

I don't understand why AMD isn't going high end since they ostensibly are doing GPU chiplets, high end should be the one area they obviously have an advantage. Makes me think NVIDIA has pictures of someone or something and AMD isn't actually playing their side of the field.

they've done this before, given up competing in the high end for a generation and focused on mid-range - GCN Polaris, RDNA 1

 

[The Mass Shooter Ron Soye]

I don't understand why AMD isn't going high end since they ostensibly are doing GPU chiplets, high end should be the one area they obviously have an advantage. Makes me think NVIDIA has pictures of someone or something and AMD isn't actually playing their side of the field.

1. GPU chiplets are hard. Top RDNA3 dies underperformed. Navi 48 and 44 switched back to monolithic.

2. Competing at the high-end won't work for AMD. AMD's mind/market share in GPUs is not good at all, maybe around 5-15% depending on your source. Nvidia is rightly perceived to have a software advantage with things like DLSS. If the playing field isn't level, they can put out a great GPU but it will be ignored by the market. So they decided to do a repeat of the RDNA1 generation (5700 XT maximum). They are launching AI-based FSR4 so maybe the DLSS advantage will lessen and they will go for the high-end again next-gen.

3. They did prepare to compete at the high-end with chiplets but cancelled it:

https://www.techspot.com/news/102807-amd-cancelled-enthusiast-class-rdna-4-gpu-could.html

 

[barrow greeter]

2. Competing at the high-end won't work for AMD. AMD's mind/market share in GPUs is not good at all, maybe around 5-15% depending on your source. Nvidia is rightly perceived to have a software advantage with things like DLSS. If the playing field isn't level, they can put out a great GPU but it will be ignored by the market. So they decided to do a repeat of the RDNA1 generation (5700 XT maximum). They are launching AI-based FSR4 so maybe the DLSS advantage will lessen and they will go for the high-end again next-gen.

It will be many more years and a lot of money and some severely altered circumstances before AMD can compete with Nvidia in the halo product category again - but nothing is stopping them from accepting a lower profit margin than Nvidia and competing in the mid range (where most consumers are).

That's not to diminish the importance of owning the halo product category, however. - because the way it works is that because Nvidia has the fastest consumer GPU, even if it costs $2000, and even though most people can't buy it, it still influences them towards the Nvidia product they eventually buy, aka a 5070 or 5060. AMD can't help but be seen as a pale imitator while Nvidia is choosing to market enterprise level products as halo tier consumer gaming products. - that and keeping ahead and setting the tone with software and hardware integration ala DLSS, frame gen, etc.

 

[geckogoy]

Top RDNA3 dies underperformed. Navi 48 and 44 switched back to monolithic.

Thats a disaster

 

[barrow greeter]

Thats a disaster

it was! it was a whole thing - MLID did a handful of related analysis vids on what happened

 

[The Mass Shooter Ron Soye]

Thats a disaster

Chiplets can be good but they have to overcome the inherent advantages of monolithic chips. Arrow Lake's problems probably stem from the switch to chiplets. RDNA3 didn't even use more than one GCD (graphics complex die) even though spamming 4/6/8 of those is how you would presumably leave the 4090/5090 in the dust by busting past the reticle limit. Smaller chiplets could be cheaper per mm^2 from having very high yields, but putting several of them together could make the product more expensive overall. And it still has to work correctly.

If the leaks are correct, the 9070 XT will be a good alternative to at least the RTX 5070. 9070 non-XT will be cheaper but retain 16 GB. Top Navi 44 (9060 XT) might be like a cheaper 7700 XT (probably with 16 GB), and the 9050/9040 (8 GB?) could be good low-end replacements for 7600/6600 XT/6500 XT depending on the perf/specs.

---

Strix Halo doesn't connect the Zen 5 desktop chiplets in the usual way. It uses a "fan out" interconnect, like Infinity Links in RDNA3:

Chips and Cheese: AMD's Strix Halo - Under the Hood (archive)

That's a sea of wires. We use fan out, we're for level fan out in order to connect the two dies. So you get the lower latency, the lower power, it's stateless. So we're able to just connect the data fabric through that connect interface into the CCD. So the first big change between a Granite or a 9950X3D and this the Strix Halo always the die to die interconnect. Low power, same high bandwidth, 32 bytes per cycle in both directions, lower latency. So everything that and almost instant on and off stateless because it's just a sea of wires going across. So it's a little [bit of a tradeoff] of course, the fabrication technology is more expensive than the one over there [points to a 9950X3D], but it meets the needs of the customer and the fact that it has to be a low power that can actually connect.

You get full AVX-512 like on desktop (good for some emulators). CPU cores can't use the 32 MiB MALL/Infinity cache as expected, but the interviewee suggests that firmware updates could allow the NPU or VCN (video decode/encode) to use it in the future.

 

[Fcret]

I don't understand why AMD isn't going high end since they ostensibly are doing GPU chiplets, high end should be the one area they obviously have an advantage. Makes me think NVIDIA has pictures of someone or something and AMD isn't actually playing their side of the field.

Their high-end consistently underperforms. 7900 XT and XTX were some of the worst selling cards of last gen because everyone with 7900 XT(X) money just bought nvidia instead. This is one of the speculated reasons as to why they pushed off RDNA4 until 2025 instead of the original Q4 2024 target - massive unsold stock of higher-end GPUs.

The two AMD cards that did okay last gen were the 7600 XT and 7800 XT - cards that exist at a price point where many of the advantages of nvidia aren't as pronounced (the 4070 might be better at raytracing than the 7800 XT but it's still weak enough at it that it's not going to be a major selling point).

AMD desperately needs marketshare because right now there's not even a good incentive for devs to implement FSR at all (to say nothing of FSR4) when nvidia has something like 90% marketshare. They do this by focusing all their volume at lower price points in the hopes of undercutting nvidia enough to displace them.

Also the market for dedicated consumer GPUs is already not that large. Nvidia can sell shit like the 4080 and 4090 because there's a dedicated prosumer class that'll slurp it up if gamers don't bite, but AMD's gaming GPUs don't really work well here due to the RDNA/CDNA split.

 

[The Ugly One]

AMD desperately needs marketshare because right now there's not even a good incentive for devs to implement FSR at all (to say nothing of FSR4) when nvidia has something like 90% marketshare.

Microsoft has abstracted upscaling away in the latest version of DirectX. You no longer have to implement FSR, DLSS, and XeSS paths separately. Just call the DirectX API and it handles that for you.

What AMD needs to do is leapfrog NVIDIA by inventing the next big thing. But that's far easier said than done.

 

[#lurker]

there's not even a good incentive for devs to implement FSR

Yes there is. It literally works on every card, is easy to implement, and gives users more flexibility.

 

[Betonhaus]

I don't understand why AMD isn't going high end since they ostensibly are doing GPU chiplets, high end should be the one area they obviously have an advantage. Makes me think NVIDIA has pictures of someone or something and AMD isn't actually playing their side of the field.

The short version is that there's a very large overlap between people who convince themselves that they need $2,000 GPUs that can play their games at 32k resolution and 7,000fps, and people who blindly buy Nvidia and refuse to even consider AMD because of their preconceived notion that AMD is bad. Right now the only games that aren't comfortably playable on modern midrange cards are AAA slop that's horribly unoptimized and just not very fun.

Is there a single game that cannot run on a Nvidia 3060 that has decent community reviews?

 

[barrow greeter]

Just to provide some context and up to date CES observations, FSR 4 might actually be good :

Not DLSS beating all told, but near enough parity in many mainstream scenarios, and not outright awful anymore

 

[WULULULULU]

For the Corsair 4000D, you should use the front and bottom as intakes, and the back and top as exhausts. Make sure you mount your CPU cooler such that it's aligned with the airflow. The intake fan should point at the front and the exhaust fan at the rear, it's a surprisingly common mistake to mount the cooler upside down such that it's blowing against the airflow of the case, which will harm cooling. Also don't fill the entire top with fans. A fan at the front top of the case will just exhaust cool air, which is completely pointless. Air should pass a hot component before it gets exhausted. With four fans, I would put three at the front and one at the rear. More intake than exhaust means you'll get a positive pressure inside the case, so there will still be air exiting at the top, but you'll make sure more of the airflow is directed toward the CPU and GPU. Also remove the dust filters from any position that doesn't have an intake.

Back. I made the right call and halted my order. By the way, when you say Bottom intake, do you mean the space above the PSU? I got a picture for reference:

 

[Susanna]

Back. I made the right call and halted my order. By the way, when you say Bottom intake, do you mean the space above the PSU? I got a picture for reference:
View attachment 6857164
View attachment 6857166View attachment 6857225

I actually meant inside the bottom, next to the PSU. But I see now that with the HDD cage you probably don’t have the room for that.
You cancelled the order for the Corsair case? Look into the Fractal Torrent or Fractal North. Both are well-designed cases with excellent ventilation preinstalled, you don’t need to buy additional fans for either of them.