AMD Ryzen

[Keyrock]

On 4/16/2023 at 4:36 PM, Zoraptor said:

The problem with efficiency is... well, if you're after efficiency you should pretty much only consider a chip from a plucky start up company from Cambridge UK, or one of its many derivative designs, rather than anything from Intel/ AMD (though both, of course, have an ARM license and indeed AMD's NSA back door vector  PSP is actually ARM rather than x86).

In terms of pure power efficiency something RISC-V based might be the way to go. Obviously, OS and application compatibility is an issue when you step outside the x86 and ARM ecosystems, but in that regard, as a Linux weirdo, I have a leg up, given that when a piece of hardware comes out, no matter how niche, someone out there with enough interest and skill will port Linux and it's many applications to said hardware and then it's out there in the aether because open source (try to beat that run-on sentence). Indeed, in a few years, when the battery on my current phone starts to die, I will consider a phone running RISC-V with probably a custom version of Linux. I'm not a power user with phones, I use it to make calls, send text messages, and watch an occasional video, that's pretty much it. I'm growing increasingly frustrated with Android and wouldn't mind distancing myself from "Do No Evil" Google, but I have no desire to have anything to do with Apple's walled garden either.

Speaking of efficient, and getting back to AMD, 7040U Phoenix ultrathin laptops and handhelds should be hitting the market any day. I've seen some dubious benchmarks out there that look really promising, but nothing reliable yet. Being able to run pretty much anything at 1080p at ~20W is the dream. 

Edited April 22, 2023 by Keyrock

[majestic]

Nothing like buying 700$+ of hardware only to have it literally blow up in your face. GG.

Edit: Probably best to make something clear for the inevitable TL;DR crowd, the mainboard melting is an ASUS-specific issue, the CPUs dying is not, so if you don't have a mainboard from ASUS, at least "only" your CPU will fry itself and you can still use the mainboard afterwards, and from what one can gather from the affected users, it is not limited to 7800X3D CPUs at all (der8auer has a 7900X that died in this exact same way), they're just more affected than the others. Why would be pure speculation at this point, a follow up video with the analysis from an external lab is coming soon.

Either way, AMD, perhaps it is best to not make fun of melting cables next time. Almost enough to make one believe in karma.

Edit 2: I remember Steve telling me that AMD said that they're not going to bin lower grade 7950X3Ds as 7800X3D CPUs, would perhaps be interesting to see if that is really the case, because otherwise the higher amount of affected 7800X3Ds could be explained by them being the CPUs that already failed the silicon lottery, combined with board manufacturers setting EXPO values in a way that slowly (or sometimes more quickly) cook your CPU. Never mind the other hilarious bugs that came up, like the failsafe shutdown temperature values not being correct for X3D CPUs (funnily enough, ASUS got that properly, the others... not so much).

Some board manufactueres apparently used Palpatine's solution to quickly fix instabilities. Just shoot it with lightning.

Edited April 30, 2023 by majestic

[Zoraptor]

13 hours ago, majestic said:

Either way, AMD, perhaps it is best to not make fun of melting cables next time. Almost enough to make one believe in karma.

AMD marketing has more than a bit of a habit of doing that. Next up most likely is the RX7600 having 8GB of VRAM after making fun of nVidia for skimping on their RAM.

[majestic]

TL;DR: Suspected root cause is (long term) degradation by unsafe voltages, potentially assisted by manufacturing variances and minor defects, otherwise known as silicon lottery. The lab tech identifies a whole host of other potential causes, but the issue appearing more frequently on ASUS mainboards with their excessive SoC voltage really just points to, well, failure due to unsafe voltages.

[Keyrock]

The reviews for the ASUS Ally, the first of the AMD Phoenix handhelds are coming in:

https://www.wired.com/review/asus-rog-ally/

https://www.tomsguide.com/reviews/asus-rog-ally

I'm mainly interested in the performance of the chip as I have no interest in any hardware from ASUS due to a string of bad experiences. The Phoenix Z1 Extreme (glorified 7040U APU) does seem powerful, albeit bogged down by the not optimized for handhelds Windows 11. I wonder if SteamOS could be installed on it? I don't see why not. There will be a dozen more handhelds with Phoenix APUs soon, no doubt. I would consider one with a 8" screen and better battery.

I may wait to see what Meteor Lake has to offer later this year, or Strix Point in 2024 rumored to be offered with up to 16 CUs of RDNA3+ (vs 12 CUs of RDNA3). Then again, there will always be something better on the horizon. I'll be looking to get a handheld when there is one that can play pretty much anything at 1080p. ASUS Ally kind of, sort of, almost gets there. 

[Zoraptor]

Has no reviewer tried installing a linux on one? If not I'd guess there's a contractual reason in order to receive a review unit.

Wonder how much they paid for the Win11 licenses and how much cost it adds.

[Keyrock]

42 minutes ago, Zoraptor said:

Has no reviewer tried installing a linux on one? If not I'd guess there's a contractual reason in order to receive a review unit.

There's no reason why you couldn't install Linux on the Ally unless ASUS did some sort of bollocks to the BIOS, even then someone could and would break that eventually. There's a good chance SteamOS 3.0 would run on the Ally out of the box, given that it was specifically designed for Zen2/RDNA2 and this is Zen4/RDNA3, which, while not identical, is still at its core the same architecture on both the CPU and GPU side. I'm pretty sure SteamOS 3.0 is supported on the Zen3+/RDNA2 6800U handhelds that are out now. There are, I'm sure, plenty of other Linux distros you could put on whatever handheld, though they'd be unlikely to be as optimized for that form factor and use case as SteamOS 3.0 is.

Having Windows 11 on the handheld gives it the advantage of having all the games be native*, whereas with SteamOS you have to run any non-Linux native game (which is to say the majority of games) through a compatibility layer (Proton). The downside to Windows 11 is that it wasn't made with a handheld gaming console in mind and, given that Windows is proprietary software, ASUS themselves can't alter it in any meaningful way, unless Microsoft is willing to let them look at and alter the source code. So, which one is better solution, clunky OS but with virtually all games running natively or purpose built streamlined OS but most games have to run through a compatibility layer? I suppose if Microsoft were to make a streamlined purpose built version of Windows for handhelds then you could have the best of both worlds, in a sense, but that doesn't exist (yet?).

Edited May 11, 2023 by Keyrock

[majestic]

On 4/30/2023 at 10:29 PM, Zoraptor said:

AMD marketing has more than a bit of a habit of doing that. Next up most likely is the RX7600 having 8GB of VRAM after making fun of nVidia for skimping on their RAM.

Well, all current information point to you calling that early. 8GB VRAM on a 128 bit bus, literally everything they and their fanboy brigade made fun of the 4060 series specs for.

Edited May 19, 2023 by majestic

[Zoraptor]

That wasn't exactly a risky prediction since every leak about Navi33 had a 128 bit bus.

They may well have a 16GB version like nVidia does with the 4060Ti- whether they charge an extras 100usd for it as well, who knows.

[majestic]

Well, not entirely AMD Ryzen related, but anyway:

https://cachewarpattack.com/

Quote

CacheWarp is a new software fault attack on AMD SEV-ES and SEV-SNP. It allows attackers to hijack control flow, break into encrypted VMs, and perform privilege escalation inside the VM.

[...]

In 3 case studies, we demonstrate an attack on RSA in the
Intel IPP crypto library, recovering the entire private key, log-
ging into an OpenSSH server without authentication, and
escalating privileges to root via the sudo binary. While we
implement a software-based mitigation proof-of-concept, we
argue that mitigations are difficult, as the root cause is in the
hardware.

Good news is that the fix is not expected to have any performance repercussions.

Quote

 

No mitigation is available for the first or second generations of EPYC^™ processors (“Zen 1”, formerly codenamed “Naples”, “Zen 2”, formerly codenamed “Rome”) since the SEV and SEV-ES features are not designed to protect guest VM memory integrity and the SEV-SNP is not available.

As a mitigation for the potential vulnerability, AMD has provided a hot-loadable microcode patch and updated the firmware image for AMD 3^rd generation EPYC™ processors (“Zen 3” microarchitecture, formerly codenamed “Milan”) for customers with the AMD Secure Encrypted Virtualization-Secure Nested Paging (SEV-SNP) feature enabled.  No performance impact is expected from the patch.

This issue has not been found to impact AMD 4^th generation “Genoa” EPYC™ processors (“Zen 4” microarchitecture).

 

 

[Keyrock]

tl;dw: AI, AI, AI, and also AI. Wait, did I mention AI?

They announced the very soon arriving 8040 APUs, these are functionally identical to the current 7040 APUs that are powering many handhelds sans the Steam Deck, except they also have more AI cores. I don't expect these AI cores to have any impact on gaming, not right now, anyway. I guess, in theory, they could make FSR3 frame generation or fluid frames or whatever the **** they call it better? Then we got the 8000G desktop APUs which don't interest me, but they might be of interest to somebody. I was hoping to get some news on Strix Point APUs, which is what I'm really waiting for as I suspect they will be the chips that finally deliver on my dream of a true 1080p handheld, but they were only mentioned in passing.I suppose it makes sense, given they are currently launching the Hawk Point 8040 APUs and giving details on Strix Point would be stepping on their own launch. Oh well, I'll have to wait for the next big conference.

Side-note: My definition of "true 1080p" is that I can throw pretty much any game at it and get acceptable framerates at 1080p at medium or higher settings.

Edited January 8 by Keyrock

[majestic]

TL;DR: Well, no surprise there. The iGPU is a significant gaming boost over existing ones, but still beaten by something as trashy as the RX 6500, and completely outclassed by an RX 6600 when combined with a cheap four core i3-12100F, a combination that costs about the same.

[Bartimaeus]

3 hours ago, majestic said:

TL;DR: Well, no surprise there. The iGPU is a significant gaming boost over existing ones, but still beaten by something as trashy as the RX 6500, and completely outclassed by an RX 6600 when combined with a cheap four core i3-12100F, a combination that costs about the same.

One moment, I have to go look up benchmarks for my GPU to see how it compares to the "trashy" RX 6500.

. . .

Phew, okay, I guess I'm doing alright. Yeah, as the review points out, the CPU itself is obviously much better than the i3-12100F, but...modern gaming requires a GPU, and the iGPU just isn't good enough to make the product make sense outside of maybe specialized power/battery/size-limited scenarios. If it's a gaming chip, then having a really good CPU and a bad iGPU just isn't a great pairing.

Edited January 29 by Bartimaeus

[Humanoid]

Last year I built a 5700G-based system in a SilverStone Milo 10 case, half for the novelty of it. This was after finding that my NUC8, which features Iris graphics better than the NUCs that came after it, was still incapable of running fairly light games like Rogue Legacy 2 - a side-scrolling platformer.

It was basically the best I could do in a case of that size, which has absolutely nil support for a standalone video card. Hell, in the "short" case configuration I'm using, most CPU stock coolers don't even fit. So yeah, I guess I would be the target market for something like this.

Edited January 30 by Humanoid

[Keyrock]

Seems like a good fit for a small form factor HTPC.

[majestic]

On 1/29/2024 at 7:50 PM, Bartimaeus said:

One moment, I have to go look up benchmarks for my GPU to see how it compares to the "trashy" RX 6500.

Nah, the RX 6500 XT is really just trash - when every tech youtube channel under the sun (even the most biased ones) agree that it is bad, it is really bad. Even the AMD fanboy brigade has a hard time defending the GPU, outside of saying "at least it was available during the mining craze" - forgetting that it initially cost 300$, a price point everyone would agree is too much for an infinitely more powerful RTX 4060, after two years of crazy inflation.

On 1/29/2024 at 7:50 PM, Bartimaeus said:

Phew, okay, I guess I'm doing alright. Yeah, as the review points out, the CPU itself is obviously much better than the i3-12100F, but...modern gaming requires a GPU, and the iGPU just isn't good enough to make the product make sense outside of maybe specialized power/battery/size-limited scenarios. If it's a gaming chip, then having a really good CPU and a bad iGPU just isn't a great pairing.

Indeed. You can build a really small form factor PC for some light gaming with it, but as @Humanoid said, even his older build was just more for the novelty of it and because he had a pair of RAM sticks left over. It is probably overkill for HTPC unless you need hardware AV1 support - and for that the lower tier APUs will do just as well, and if you don't, any i5 T-model from the past decade is going to be enough for less power draw than the 8700G consumes while idle. There's also a hit in available PCIE lanes in exchange for the better iGPU, making it less attractive for cases where you need the IO, so for actual work cases you're probably still better off with a 7700(X) while just using the regular iGPU.

Kinda struggling to see the use case here. I mean, outside of gaming handhelds. Well, perhaps its NPUs are really good for AI workloads.

Oh, also, classic AMD:

 

Edited January 30 by majestic

[majestic]

Here I was wondering why AMD's stock kept tumbling down in the face of Intel's fumbling, but that explains it.

[Bartimaeus]

A couple week delay to make sure everything is as it should be is better than...uh, whatever Intel's been doing for the past couple of years, at least, .

Funny comment:

"Intel: delays performance-botching patch until after AMD's launch.
AMD: delays launch until after Intel's performance-botching patch."

Heh.

[majestic]

39 minutes ago, Bartimaeus said:

A couple week delay to make sure everything is as it should be is better than...uh, whatever Intel's been doing for the past couple of years, at least, .

Can't argue with that, still a pretty bad moment when you have to recall CPUs already shipped to system integrators - systems probably all set and ready to be sold in the face of the worst possible moment for Intel since the Pentium 3 1133 debacle need to be worked on and validated again, and we're talking about a market segment where AMD is still trailing behind (x86 market shares being what they are when removing console SoCs).

Somehow I don't think we're talking about quality issues that are "doesn't reach boost spec by 50 Mhz" like AMD had in the past, but hey, could be wrong.

edit: as funny as that comment is, I think we don't need to consider the veracity of that. Ain't no way AMD would delay a product launch just to appear better than Intel when everything points to the new CPUs being better than Intel's current offerings anyway, and it would be really weird if they were not, it's a brand new generation after processors that were already competitive and in some areas much better (and more efficient) than Intel's most recent offerings.

Edited July 24 by majestic

[majestic]

This is far better than my wildest dreams. To quote Steve: "It's the Skylake stagnation all over again." Well, and the memory stability problems of early AM5 are back with a vengeance. 

Steve found two positive things in his review:

1. It is really efficient
2. It is not from Intel

Love the (admittedly few, but existing) benchmarks where the 9700X is behind the 7700X. 

Edit:

On 7/24/2024 at 11:02 PM, majestic said:

edit: as funny as that comment is, I think we don't need to consider the veracity of that. Ain't no way AMD would delay a product launch just to appear better than Intel when everything points to the new CPUs being better than Intel's current offerings anyway, and it would be really weird if they were not, it's a brand new generation after processors that were already competitive and in some areas much better (and more efficient) than Intel's most recent offerings.

Guess I was wrong, ey?  

Edited August 7 by majestic

[Bartimaeus]

Oh no, it's happening again... Intel falls off a cliff, and AMD immediately rests on its laurels.

[Humanoid]

I do like that they've reined in the power consumption of the X CPUs to be more in line with the non-X Zen 4 CPUs, even if it makes the raw performance improvement over Zen 4 look poor. A simple 8C/16T CPU at ~135W was not the direction I liked AMD going in, so I'm glad they've not done an Intel and kept pushing up the power limits.

That's not the full story of course. 7700X to 9700X, 5% improvement for around 40% less power? Great, I wouldn't complain if that was what we got compared to the previous best. The problem though is that the 7700 exists, and suddenly the 9700X is more like 8% faster for zero improvement in wattage, significantly less impressive. And also probably a sign that there won't be a cut-price 9700 in our future. Ah well.

Edited August 7 by Humanoid

[Bartimaeus]

On 8/7/2024 at 11:33 AM, Humanoid said:

I do like that they've reined in the power consumption of the X CPUs to be more in line with the non-X Zen 4 CPUs, even if it makes the raw performance improvement over Zen 4 look poor. A simple 8C/16T CPU at ~135W was not the direction I liked AMD going in, so I'm glad they've not done an Intel and kept pushing up the power limits.

That's not the full story of course. 7700X to 9700X, 5% improvement for around 40% less power? Great, I wouldn't complain if that was what we got compared to the previous best. The problem though is that the 7700 exists, and suddenly the 9700X is more like 8% faster for zero improvement in wattage, significantly less impressive. And also probably a sign that there won't be a cut-price 9700 in our future. Ah well.

Yes, if you can perform the same at significantly better power efficiency or perform significantly better at the same power efficiency, it's an upgrade...assuming it works both ways. There's always been exponentially diminishing returns as you feed these Zen CPUs more power though, so we'll see how that works out in practice in the coming weeks. AMD should've made sure its release was squeaky clean with good communication of these facts and making sure existing motherboards had stable and functional BIOS updates ready, especially after delaying it for half a month, and that doesn't seem to have happened.

Edited August 8 by Bartimaeus

[Zoraptor]

A comparison to Intel's Skylake stagnation is definite hyperbole. Krzanich would have sacrificed [a moderate number of lucrative share options] to satan to get an ~8% like to like generational improvement instead of, well, the 1% he managed.

[majestic]

The 9600X that Gamers Nexus has not tested yet because theirs was faulty showed a 1% increase in performance over the 7600X:

At a 40% premium over 7600X street prices.