OcUK RX7900 series review thread

[Hotwired]

The the pro 7900XTX overclockers will get a 4GB and 6GB vram XOC BIOS, us plebs won't see that but, ultimately that's one of the solutions.

The Tier 1 AIB 7900XTX cards will get binned vram chips and binned GPU's so, there's a good chance performance will do another leap, even without a big power-limit increase as the base components will be chugging less juice.

The XOC Bioses will be the one though, I imagine disabling 20GB of vRAM would have a big effect on power-draw and OC stability.

I've heard of having minimum ram for cpu oc but hiding gpu memory through bios also?

 

[fs123]

@Grim5 The Mesh Shader performance in 3DMark is a known issue under investigation, debug is ongoing you’ll be pleased to know. Doesn’t affect any games.

There aren't any games that use mesh shaders. Even if it is a bug that is fixed, it's not going to improve performance for any current games. I'd rather AMD fix the real issues like why some cards are not even reaching their boost clocks in many games.
The 7900XTX is only reaching 2.25GHz in this video..

 

[LtMatt]

There aren't any games that use mesh shaders. Even if it is a bug that is fixed, it's not going to improve performance for any current games. I'd rather AMD fix the real issues like why some cards are not even reaching their boost clocks in many games.
The 7900XTX is only reaching 2.25GHz in this video..

I know, I had a little chuckle about that in another thread.

Regarding your second question, the MBA card is bouncing off the power limit most of the time at 4K, so that's why it doesn't reach the maximum boost clocks. If you undervolt, which reduces power draw and gives more headroom before you hit the power limit, you'll see higher core clocks. Note my front end clock and shader clock in this video.

That's with the card undervolted from 1.150mv stock to 1.030v. Core clocks are running stock which is 3.1Ghz.

No link to post?

I can't link you to an internal engineering ticket Grim, you'll just have to take my word for it.

 

[Grim5]

You haven't yet posted a link to where AMD acknowledges the mesh shader issue and says they will fix it

I do agree it's not an urgent issue, more important to fix within the next few months is:

* Inconsistent performance when playing the same game for some users

* Black screen crashing with some configurations

* Long system boot times with some configurations

* Inability to translate raw compute performance into significant improvements in rasterisation game framerates in many games vs rdna2

* Inability to produce higher framerates in VR games than rdna2

* Issues with hotspot temps for some users; reports of 110c temps - more of a QC build issue not a software issue for AMD

* High power draw in low or no load conditions

 

[LtMatt]

You haven't yet posted a link to where AMD acknowledges the mesh shader issue and says they will fix it

See my edited post above.

 

[Blackjack Davy]

I agree. There are one or two posting in AMD threads of late with a clear enjoyment of posting anything that trashes AMD and adding nothing of value to the discussions. It's becoming a pain reading the threads and visiting the graphics forum. I won't name people but I think we all know who they. Some post weighted arguments or a balanced understanding which is great. Don't understand why they can't stick to their own threads if they dislike one brand so much. They're as unwanted here as Ian Watkins in a creche.

Put them on ignore and flter out the garbage they mostly skulk here and rarely break out into the light of day onto other subforums so its no great loss.

 

[Hotwired]

jays video was interesting: https://www.youtube.com/watch?v=-KvkVumbgvw

If you try to push the gpu it will do it but once it hits the power limit it takes power away from the memory especially when the gpu has to work hard. So you can end up with a clock speed which is just a paper value since the performance is crippled by the memory losing power to the gpu priority.

Can't really say if that's a bad thing or not. Overclocking is an adventure beyond what the manufacturer guarantees and I'm sure between amd fiddling it and people fiddling it some kind of optimum way of fiddling an overclock will be decided on.

At any rate the higher power aib cards have a lot more power to play with so it is mostly a reference card thing.

 

[Grim5]

jays video was interesting: https://www.youtube.com/watch?v=-KvkVumbgvw

If you try to push the gpu it will do it but once it hits the power limit it takes power away from the memory especially when the gpu has to work hard. So you can end up with a clock speed which is just a paper value since the performance is crippled by the memory losing power to the gpu priority.

Can't really say if that's a bad thing or not. Overclocking is an adventure beyond what the manufacturer guarantees and I'm sure between amd fiddling it and people fiddling it some kind of optimum way of fiddling an overclock will be decided on.

At any rate the higher power aib cards have a lot more power to play with so it is mostly a reference card thing.

So it's acting like a stability mechanism; that's a good thing though - it's either that or you get system/driver crashes

 

[humbug]

So it's acting like a stability mechanism; that's a good thing though - it's either that or you get system/driver crashes

Its taking the delibratly limited board power the reference cards have, taking it from memory and applying it to the core as you clock it up, that's actually quite clever even if the result isn't what you wanted. < that is negated by AIB models with 3X 8 pin PCIe plugs that have a higher board power.

 

[Nexus18]

 

[Mr Evil]

I hear that the vapour chamber in the reference cooler may not perform well in certain orientations. Does anyone know if any of the custom coolers are better in that regard (edit: the Sapphire Nitro+ in particular, since those are the most easily available at the moment)? I have my GPU mounted vertically (long-axis vertical, with the PCI bracket at the top), which is uncommon.

EDIT: To late now my original order has been shipped. I suppose I'll find out soon if the reference cooler has a problem standing upright.

 

[turbot1984]

has anyone benched a 7900 on neon noir?

CRYENGINE | Neon Noir Ray Tracing Benchmark

Note: This product comes with a custom EULA. You will find it in your download folder. We are delighted to share ‘Neon Noir,’ a free ray tracing benchmark based on the ray tracing demo of the same name, which was made available as a video during GDC 2019. Neon Noir was developed on a bespoke...

www.cryengine.com

 

[Grim5]

Let the rumours begin; red gaming tech is waiting no time and starts blasting click bait specs for multiple cards

Prepare yourself for

AMD rdna4: 144 CUs split across 3 GCDs with 3.5ghz clock speed, targeting 60-80% higher performance over rdna3.

Nvidia Blackwell: Multiple chiplets including Cuda core chiplets, memory chiplets and fixed function RT and denoising accelerator chiplets. Targeting "largest ever performance increase for Nvidia"

Intel Battlemage: Multiple Chiplets, targeting 7900xt/rtx4070 performance at 225w

 

[Grim5]

Fantastic investigative work from the DF crew; they've been the first ones to figure out AMD's slides. Turns out that the "up to 1.7x" number is very literal; it means when you look up at the sky in your game the performance on the 7900xtx skyrockets relative to the Nvidia cards. It sounds like a joke, but that's actually what happens in the video lol

 

[Harlequin]

Digital Shilleries at their finest. PR department for Nv

 

[Nexus18]

Fantastic investigative work from the DF crew; they've been the first ones to figure out AMD's slides. Turns out that the "up to 1.7x" number is very literal; it means when you look up at the sky in your game the performance on the 7900xtx skyrockets relative to the Nvidia cards. It sounds like a joke, but that's actually what happens in the video lol

Exactly and they have footage to back that up too but nope "shilling for nvidia", even though they have stated, RDNA 3 is good and they're very pleased with the RT perf. increase but nope, still "shilling for nvidia"

 

[Harlequin]

Digital Shilleries have always been pro nvidia, especially with there `exclusive before everyone else` Ampere videos and subsequent non objective coverage of anything AMD or Intel. Keep going lad. Whats so funny is the focus on AMD`s poor Ray Tracing, even though it is near identical to Ampere, a shill copied by many on here.

 

[Hotwired]

I think that's an odd thing to say.

This is what they said in that video when they spent a few moments mentioning AMD raytracing performance.

- I was actually pleasantly surprised at the leap in raytracing performance which does actually seem to be consistently good whereas the gains in rasterisation can range from extremely impressive to really rather poor

- I think that's one of the best things Rich, finally AMD has a card where you can enjoy raytracing. It's not the fastest out there of course, but it's good enough now where I think most games with raytracing features can be enjoyed on these new cards which is awesome.

- It's 3080 to 3090Ti performance

- It's quite good.

I've written it down so it's not vaguely claimed to be somewhere in a video that people aren't actually watching. But if you do want to confirm, this is said at 32:10.

 

[GoogalyMoogaly]

Honestly I don't actually remember the numbers but the impression I got from reviews was that no card (even the 4090) are outputting grat numbers with RT turned on. Without using DLSS/FSR I don't remember great framerates from any of the cards.
I get the impression that at the minute the technology isn't quite there but we're just starting to get to the point where it's getting interesting.

 

[JGF111]

The Digital Foundry hate from circles is pretty amusing. I find their content overall actually very good, and their takes generally quite reasonable.

But then some circles accuse the Hardware Unboxed guys of being nvidia shills too, sooooo in summary the internet is going to internet i guess...