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

[Post Reply]

Make sure the MB supports resizable BAR. Not sure if it's critical on the A380, but apparently on the A770 it causes significant performance loss.

I'm sad I got an A770. It can game, but that's about it, all the AI stuff is currently more or less unsupported. Luckily I have a sucker, er, friend, I can give it to who will be happy with the upgrade.

Stable Diffusion supports the Intel GPUs doesn't it? Or are you referring to other AI related capabilities?

 

[DavidS877]

Stable Diffusion supports the Intel GPUs doesn't it? Or are you referring to other AI related capabilities?

For the libraries comfy uses, it's been broken almost all the time. I think I got it to work once, and then they promptly broke it again.

https://github.com/comfyanonymous/ComfyUI/discussions/476

Text-Generation-UI I don't remember if that every worked.
And non-mainstream things like WhisperX don't have a chance. Although I guess it should work with plain old Whisper.

But by far the worst part was that the BiFrost version from Acer had no way to disable the LEDs in Linux though.

 

[The Mass Shooter Ron Soye]

10-year-old Nvidia GPU is still receiving love from vendors — Asus debuts GeForce GT 710 EVO with 2GB GDDR5 memory (archive)

What's good *new* at $50, $100, $150 price points? If the RTX 3050 6GB (70W TDP) went down to $100, I'd probably buy one. It's currently $175 on Newegg.

There is the A580 at $160 if you are willing to brave Intel and use more power than the PCIe connector.

 

[Just Some Other Guy]

I'd just use Intel at that price point. AMD and Nvidia put out some real gutted shit for the budget segment.

If I was staying that cheap I'd probably just buy used tbh.

*Edit* my local microcenter actually has 2 new rx580s for $130.

There's also like $160 1660ti and 5700xt, but those are refurbs.

 

[Betonhaus]

I'd just use Intel at that price point. AMD and Nvidia put out some real gutted shit for the budget segment.

If I was staying that cheap I'd probably just buy used tbh.

oh yeah anything under $200 I'd just buy used, and get something equivalent in power to something up to double it's price in stores

 

[The Mass Shooter Ron Soye]

Stuff from the AMD Computex keynote:

Desktop CPUs

AMD claims a +16% IPC increase for Ryzen 9000 (Zen 5), launching in July 2024. That figure may be biased towards AVX-512, which could have gigantic uplifts from possibly having a full 512-bit implementation.

Base clocks regressed for the 6-core 9600X (4.7 -> 3.9 GHz), 8-core 9700X (4.5 -> 3.8 GHz), 12-core 9900X (4.7 -> 4.4 GHz), and 16-core 9950X (4.5 -> 4.3 GHz). I am comparing them to the 7600X, 7700X, 7900X, and 7950X. There's no 3D V-Cache on these particular CPUs, so it will be interesting to hear AMD's explanation for this. It may not affect real world performance at all if most people have the cooling needed to hit the higher clocks. The 9900X drops to 120W TDP / 162W PPT from the 7900X's 170W / 230W, matching the 7900X3D's lower power limits. So it's likely that Zen 5 is going to be very power efficient.

Turbo clocks went up for the 6-core 9600X (5.3 -> 5.4 GHz) and 8-core 9700X (5.4 -> 5.5 GHz).

AMD is now committed to supporting the AM5 socket through 2027+, and is in fact launching two "5000XT" CPUs for the AM4 socket in July. We can infer this means that Zen 6 is definitely coming to AM5.

New X870/X870E motherboards will all supposedly support USB 4, PCIe 5 for graphics/NVMe, and higher memory clocks. However they may be lying about this for X870, so take the whole thing with a grain of salt.

Mobile APUs

Strix Point mobile APUs are officially announced, with up to 12 cores (4x Zen 5 + 8x Zen 5c), 16 CUs RDNA3.5 (Radeon 890M), XDNA2 at 50 TOPS INT8.

AMD adopted a GayI naming scheme for at least two Strix Point SKUs, which were the only ones they announced: the 12-core Ryzen AI 9 HX 370 and 10-core Ryzen AI 9 365. The 10-core disables two of the Zen 5c cores and has 12 CUs (Radeon 880M).

L3 cache is 24 MB, but it's going to be split between the Zen 5 and Zen 5c clusters. The Zen 5 cores get 16 MB, the Zen 5c cores get 8 MB. So that shouldn't make much difference from Phoenix/Hawk which had 16 MB for all cores, but maybe there's an argument to be made here about giving the big cores more breathing room. We'll have to see if it makes any difference in testing.

They have a 15-54W TDP range, indicating that the U and H(S) chip lineups may be merged now. Maybe a good idea at first glance, but it could confuse consumers when they get a 15W thin-and-light laptop model and expect it to perform like the unchained 54W models.

Lisa Su joked on stage with the Microsoft guy that the XDNA2 NPU is taking up a lot of die space. If this image is drawn to scale, then by my quick estimate the NPU is taking up about 59-60% of the die area of the integrated graphics:

"We could have had 24 CUs!" screamed the Gamer, forgetting that the APU is stuck on the usual 128-bit memory bus. You could argue it makes the chip more expensive. Blame Microsoft, they are the company practically forcing AMD, Intel, and Qualcomm to add 40+ TOPS NPUs to everything.

XDNA2 adds support for "Block FP16", which supposedly has the performance of INT8 but the accuracy of FP16, without quantization. So it also has a full 50 TOPS of this new Block FP16 performance, not just INT8.

Just as we say Hawk Point raise clock speeds to hit 16 TOPS with XDNA1, we can expect different implementations to have different performance. Strix Halo is now rumored to hit 60-70 TOPS with the NPU. Maybe there will be lower chips like Kraken Point that only hit 40-45.

Other

Turin Epyc CPUs using Zen 5/5c will have up to 192 cores (Zen 5c). AMD showed some comparisons of the 128-core version (Zen 5) against Intel. They drop into the same SP5 socket.

AMD saw Nvidia switch to an annual cadence for their AI GPU releases, and said "Aw fuck, follow the leader!" So now their roadmap has the Instinct MI325X accelerator this year, which simply adds 50% more memory (from 192 to 288 GB) and slightly faster memory clocks for ~13% more bandwidth.

The MI350 (coming 2025) will be based on CDNA4, with up to the same 288 GB of HBM3E. AMD made a claim of a 35x performance uplift from the MI300X, which is certainly some bullshittery exploiting FP4 precision and the larger pool of memory. But from what I hear these lower precision data types are becoming perfectly fine for many models, so the speedup could be real for some scenarios.

AMD Slims Down Compute With Radeon Pro W7900 Dual Slot For AI Inference

Stable Diffusion 3 (the Medium-sized model) will be available on June 12 (discuss here).

 

[Overly Serious]

It's not raw-compatibility I'm talking about, but speed. Apple's M-series chips implement certain aspects of x86's memory model specifically for the purposes of faster execution of translated x86 code. This is why you get ~80% performance of a purely native ARM application on Apple silicon when using rosetta2.

Qualcomm has already confirmed that it's not implementing things like TSO and is more or less sticking to a pure ARM implementation, so I imagine a lot of x86 binaries will run quite poorly.

That's interesting thanks. For my case it's probably not going to matter. I shared the game list mainly because I was impressed at the level of compatibility at all. It will be interesting to see any performances differences. For me, I expect most of the software I use to be available in ARM compiled versions. They interviewed a guy from Affinity. I use their graphic design software. He said it had taken them very little time to port their code over to ARM and they were ready to release it. The less big name software I use - weird little niche things like WizTree, might suffer performance loss but they're so light that what will matter to me is that they run. So I think these devices could be a good fit for me. In the video I linked they put quite a lot of focus on developing and compiling on ARM and it looked pretty slick, so I'm hopeful going forward we see companies targeting both as standard.

 

[Post Reply]

There's no 3D V-Cache on these particular CPUs, so it will be interesting to hear AMD's explanation for this. It may not affect real world performance at all if most people have the cooling needed to hit the higher clocks.

I guess we will have to wait the next gen x3D chips. That will probably be my next upgrade point, although my 5700x is holding up just fine.

and is in fact launching two "5000XT" CPUs for the AM4 socket in July.

AM4 owners stay winning. The 5900XT is an interesting one, +4 cores over the 5900x.

 

[The Mass Shooter Ron Soye]

I guess we will have to wait the next gen x3D chips. That will probably be my next upgrade point, although my 5700x is holding up just fine.

Launching X3D months after the regular chips has become tiresome, and many will skip these first four SKUs. Zen 4 sales were poor out of the gate, with the new socket not helping, until prices dropped and everyone was getting free DDR5 RAM.

AM4 owners stay winning. The 5900XT is an interesting one, +4 cores over the 5900x.

It's shit naming, but it doesn't matter because they can throw any slop into the channel (due to lower quality chiplets from ongoing Epyc production) and it will end up somewhere. These AM4 releases are probably helping countries that don't have good access to tech (high prices, low availability, lower incomes).

 

[indomitable snowman]

I guess we will have to wait the next gen x3D chips. That will probably be my next upgrade point, although my 5700x is holding up just fine.

AM4 owners stay winning. The 5900XT is an interesting one, +4 cores over the 5900x.

Having longed lived sockets is Based. I was able to upgrade my early 1800x that had issues to a 3900xt a couple years ago and it was glorious. I may just get a 5900xt in a couple years if my x370 board gets an update to support lol.

 

[The Ugly One]

(due to lower quality chiplets from ongoing Epyc production)

They're planning to keep making zen 3 EPYC until 2027.

 

[Overly Serious]

Wendell of Level1Techs, the Hank Hill of Tech and all round knowledgeable tech person, posted this video earlier of Intel's Lunar Lake reveal.

It's under 20mins but some interesting stuff in there. He's still waiting to get his hands on it for testing but looking to be very efficient, the NPU takes up a crazy amount of die space and the whole thing is looking like a pretty neat SoC. As well as big efficiency gains it's up around 14% IPC but the bigger deal is the new NPM and graphics and efficiency. Looks like a really strong entry for mobile.

 

[Blue Miaplacidus]

Interesting.

 

[Just Some Other Guy]

Interesting.

Will noctua finally beat $30-$40 coolers?

 

[Fcret]

Will noctua finally beat $30-$40 coolers?

No, because the $30-$40 coolers are already overbuilt for what 80% of the DIY market needs. Their real niche is in long-term support, but how many people are really looking for a lifelong relationship with the manufacturer of their CPU cooler?

 

[Susanna]

No, because the $30-$40 coolers are already overbuilt for what 80% of the DIY market needs. Their real niche is in long-term support, but how many people are really looking for a lifelong relationship with the manufacturer of their CPU cooler?

I use exclusively Noctua fans for my own builds, and have only once had to RMA one due to failure, but that also happened to be an entirely painfree experience. "Absolutely, we'll mail you a new one. Could you please cut the cable on the broken one and send a picture?"
Noctua aren't good value for money, but they do make solid products, and if you actually care about IP67 or the manufacturer warranty, they're pretty much the only ones I'd expect to honour that (short of actual enterprise equipment like Delta, but those are as far from low-noise as you can get). The one time I bought Corsair fans, the aRGB splitter cable advertised on the box wasn't included and I had to mail them about it. They sent it without hesitation, which tells me that not including it was probably done deliberately. Skimping on a cable to save a few cents per box was something I found entirely disagreeable, since I'd already spent a ton extra getting the aRGB version of the fans (don't judge me, I was building a computer for my nephew).

 

[Just Some Other Guy]

*Shrug*

Shit happens and sometimes you get unlucky. I doubt they left a cable out on purpose.

Otoh can't say I've ever bought Corsair new. Much like noctua, they're overpriced with a "brand name" tax. Though one time there was a good deal on some basic black ml120s I did grab.

 

[The Ugly One]

(short of actual enterprise equipment like Delta, but those are as far from low-noise as you can get)

I work with a guy who's deaf from 20 years of not wearing hearing protection in the server room.

 

[WULULULULU]

Interesting.

They can keep making coolers but I'm still sticking to the Peerless Assassin.

I'm close to my setup but so far the GPU is what I'm fearing because the budget is fucking insane. Already have plans on the CPU (7800X3D) and the PSU (750 or 850 Corsair) but the prices for GPUs are still fucking absurd. Any ideas on what GPU's work best?

 

[JoseRaulChupacabra]

I very much enjoy the aesthetics of dual towers, but every benchmark I've seen shows they don't perform that much better than a single 140 class tower.

Is there even any point to them?

The thermosyphon looks to be some kind of loop heatpipe based AIO. I do find that interesting as I have a very irrational fear of water cooling.