Hacker News Reader

Comment by gspencley | Hacker News Reader

I don't know a lot of the details, but everything I've heard over the last couple of years indicated that AMD was absolutely crushing Intel.

A recent laptop I purchased, as well as the last desktop I put together (~2 years ago) each have Ryzen chips. I forget the details but in addition to performance issues, didn't Intel CPUs also have some major security vulnerabilities? And was it that they were related to instruction-level performance optimizations that, when disabled to address the security vulnerabilities, led to even worse performance?

So if AMD isn't doing great at the moment either, I can't imagine how hard Intel has been hit. I don't know anyone who is buying or recommending Intel CPUs at the moment.

View on HN

I don't have any data, but from word of mouth and variously seeing posts and videos online, 12th gen Intel CPUs seem pretty popular for gaming builds. They're winning in benchmarks against 50 series AMD (as they should, being newer), but are also cheaper. I'll be curious to see how 13th gen Intel vs. 70 series AMD plays out. There are always complicating factors, such as motherboards for 70 series AMD being quite pricey for now.

Gamers prefer anything that has better performance for cheaper price.

Tho overall having a more stable PC that consumes less electricity is better in long term and ppl really see that.

It shows in sale numbers.

Of the last gen stuff, the 5800x3D is arguably among the best bang for buck, including for use in gaming builds. The applications where Intel's 12900K/12700K hold a significant advantage against the other chip with its 96MB of L3 cache typically aren't applications desperately in need of cpu power. IME it's poorly-optimized or hard-to-optimize software with bad cache coherency that most demands speed, and it's in those cases that the X3D delivers.

The PC gaming market is tiny in the grand scheme of things.

Intel repeated their Prescott (Pentium 4/Pentium D) strategy of completely removing power limitations on their team to beat AMD.

As a result, the Intel processors have TDPs and real world power usages >2.5x that of a comparable Ryzen. Sure, they're winning, but at what cost? The 12900K at 240 watts pulls almost the same power as a 280W 64-core Threadripper.

AMD is responding in kind, with new top-end processors pulling 170W, or higher with their built-in overclocking that pushes the chip to even higher power draws as long as cooling permits. This looks to put them back into the lead, but it's just not a sustainable strategy.

Who cares what a desktop processor plugged to a wall outlet consumes? Seriously, if you care about power consumption, don't buy a K processor.

> Who cares what a desktop processor plugged to a wall outlet consumes?

Anyone without solar panels?

Those 5 people running off-grid with solar panels only certainly do care about power consumption and wouldn't buy a K processor.

It matters for SFF builds where people often want the highest performance possible but are severely limited by cooking. Efficiency becomes extremely important.

Traditionally nobody, but the massive power consumption of the RTX 4090 combined with a power-hungry CPU might make people take notice. When your desktop needs a dedicated circuit you have a problem.

My microwave uses 1200 or 1500 watts and does not need a dedicated circuit. Sure, I don’t run it continuously for an hour, but that has nothing to do with whether or not it needs a dedicated circuit.

A 16A circuit allows for 3.8kW or something like that @ 240v.

With these ludicrous power requirements, you may in fact need to rethink how you use your power. E.g. if on same as a non-heat pump drier, you'd have to make a choice between gaming and drying clothes.

Having said that I saw a reference to a 4090 offering 95% of the perf at 60% of the power usage if undervolted, so that becomes an attractive option now.

I absolutely love my 5800X3D's insanely low power usage (insane = performance per watt for gaming in simulator titles where it runs circles around Intel).

People with good insulation or who like quiet rooms?

When a 5800X3D offers equal or better gaming performance at 1/2 to 1/3rd the power consumption (e.g. 220w vs 80w), and you pay $0.65 per kW/h during peak times, I can only imagine: "Quite a few".

50$/year vs 18$/year for a pc that's always on, not exactly life changing.

The difference is closer to $1250/year vs $450/year. I suppose you forgot to multiply by 24 hours/day.

Aw crap

That rather depends on how many hours a day you use your cpu at max performance, which is likely closer to 1 hour to the average person.

If electricity was $0.65/kWh for me, I'd move out of the country lol. Assuming a more realistic $0.40/kWh (considered very expensive in the US, where most enthusiasts live), 8 hours of gaming a day, and a 200W power limit, you're paying $19/mo. Not bad.

Don't get me wrong, the 5800x3d is a phenomenal CPU. However like many owners of that CPU, I'm also in the market for a 4090 and intend to use the full 600W power limit with my water loop and outdoor radiator. CPU power consumption is just not an issue for enthusiasts.

Nice, ha ha. I was going to say, look into Undervolting. I saw 95% perf at 60% of the power come up, ie 270w; that's actually going to be superb (very cool, quiet, still extremely performant).

So maybe configure that as an optional profile for titles not needing maximum juice?

I'm quite keen myself.

Power $ mentioned was peak times in Norway; I'm in Australia where it's not anywhere that bad yet. (0.25 AUD for me).

> I saw 95% perf at 60% of the power come up, ie 270w

That's neat, but 5% is a lot when you're spending $1600. Oh and all these measurements you see tossed around are done with air coolers and stock voltages. Of course more power doesn't help if you're limited by thermals. Specifically, you're limited by hotspots you can't see because they're between the on-die temperature sensors.

> maybe configure that as an optional profile for titles not needing maximum juice?

4K gsync monitors are typically limited to 120Hz. If my GPU is capable of running a title at 1200fps, it will idle 90% of the time and consume only ~60W regardless of the 600W power limit. So tuning this is pointless. I'd still get 1/10th the rendering latency compared to a GPU that's only capable of 120fps. Same logic applies to power limiting your GPU which will increase latency and kill frame pacing in exchange for a tiny reduction in electricity bills.

> Oh and all these measurements you see tossed around are done with air coolers and stock voltages. Of course more power doesn't help if you're limited by thermals. Specifically, you're limited by hotspots you can't see because they're between the on-die temperature sensors.

This completely misunderstands heat transfer. If the hotspot temp is 75 deg even with overclocking you're not limited by thermals: https://www.youtube.com/watch?v=zc-zwQMV8-s

>4K gsync monitors are typically limited to 120Hz. If my GPU is capable of running a title at 1200fps, it will idle 90% of the time and consume only ~60W regardless of the 600W power limit. So tuning this is pointless. I'd still get 1/10th the rendering latency compared to a GPU that's only capable of 120fps. Same logic applies to power limiting your GPU which will increase latency and kill frame pacing in exchange for a tiny reduction in electricity bills.

This completely misunderstands power consumption and the nonlinear relationship between power and clockspeed. Performing the same work in the same time in short bursts of high clocks and high voltage uses more power than constant low clocks and voltage.

> (considered very expensive in the US, where most enthusiasts live)

That's very dismissive of the entire EU, which has lots of enthusiasts, and these aforementioned expensive electricity prices.

Some electricity providers in Germany even raised their prices to 1.35€/kWh (most famously DB Netz Bahnstrom plans)

I'm not sure if the actual differences in energy usage are so clear cut. These charts [0], which account for the 12900K spending less time to accomplish tasks than the 5950X, seem to indicate the disparity isn't so terrible. You can always under-volt too; the 13th gen press release includes charts [1] showing that the 13900k at 65W matches the performance of the 12900K at 241W.

[0]: (Search for "cumulative amount of energy required to perform Blender and x264 and x265 HandBrake workloads") https://www.tomshardware.com/reviews/intel-core-i9-12900k-an...

[1]: https://arstechnica.com/gadgets/2022/09/intels-first-13th-ge...

Most people don't turn their computer on to run Handbreak and turn it off immediately afterward.

Hence power draw is meaningful for desktop computers, not just energy usage for a task.

Idle power draw is not the same as peak power draw either, these arent 'nineties chips.

I was looking for a laptop for work and decent gaming, I went with an Intel NUC. Everything pointed to Ryzen being far better in most aspects - power, weight, height, and importantly battery life - however for the price point I was looking at, the faux-no-brand Intel with the 3060 was better even if it seems the chips do lag behind for similar generations.

AMD also had alot of those major bugs, but it wasn't reported as much since the original papers all targeted intel CPU's since they are the "standard", and then a couple months later someone would do it for AMD but it was old news.

AMD's CPUs, including the latest Ryzen and Epyc processors, are immune to:

    Meltdown (Spectre v3)
    Spectre v3a
    LazyFPU
    TLBleed
    Spectre v1.2
    L1TF/Foreshadow
    SPOILER
    SpectreRSB
    MDS attacks (ZombieLoad, Fallout, RIDL)
    SWAPGS

https://www.tomshardware.com/features/intel-amd-most-secure-...

Are any of these attacks relevant unless you run a cloud service ?

All of the attacks are relevant in any environment where you execute code that you would rather not have access to any other information on your system (i.e. where you actually care about privileges).

For example, there have been demos of spectre in-browser, but applies to multi-user environments or simply apps that you didn't grant admin privs...

If you treat your computer security in a DOS/Win95 fashion, then none of them are relevant.

AMD CPUs were vulnerable to fewer of the speculative execution bugs than Intel, and the performance impact of the mitigations on AMD CPUs were far less than Intel.

> I don't know anyone who is buying or recommending Intel CPUs at the moment.

Aren’t the major cloud providers still mainly running on and buying Intel?

AWS developed two generations of an in-house ARM chip (Graviton), which I think is cheaper for an equivalent power to an x86 instance. Cloudflare has also begun switching to ARM, for 57% more performance per watt:

https://blog.cloudflare.com/designing-edge-servers-with-arm-...

IIRC GCP is using AMD EPYC.

Not sure about Azure.

Azure uses Intel, AMD and Ampere. You implicitly select your CPU when you pick the virtual machine family: https://learn.microsoft.com/en-us/azure/virtual-machines/vm-...

GCP offers instances on many processors, including AMD EPYC: https://cloud.google.com/blog/products/compute/amd-epyc-proc.... Here's another one: https://cloud.google.com/intel/

ARM is increasingly being a factor there afaik.

The past 2 companies I've been at have or had huge efforts to migrate large parts of infrastructure to AWS Graviton which are 64-bit ARM iirc.

AWS certainly offer AMD boxes as well (M5a, M6a, T3a etc), but nothing like the broad range of Intels, and not even the same kind of offerings as Graviton.

The cloud providers have been buying Intel for years and can't afford to just throw thos chips away. AMD is capturing a much larger proportion of data center CPU sales, but you should still expect to see them buying Intel as well, since AMD can't produce enough chips to satisfy all demand. Also, qualifying new platforms is very expensive so cloud providers will continue to buy already-qualified platforms for as long as the perf/W makes sense.

AMD chips had the same branch prediction side channel vulnerabilities that intel did. At this point their offerings are pretty similar with intel being a bit cheaper but using more electricity.

With the new releases, it seems that 5800x3D takes the crown for single threaded tasks and video games even in intel's 13xxx benchmarks. The 7950x is closely trailed by the 5800x3D.

Similar to response to maldev higher up - AMD has some of the same issues, but far fewer. And the mitigations had less of an impact.

Amd revenue 2021 -> $16.4 billion Intel 2021 -> $79.024 billion

But 2be fair amd is/was growing and intel is stagnant and lose market share right now but with a massive investments and going back on track with fabs process i would assume Intel soon start 2grow again (ofc after the end of the recession)

We have to remember AMD is making money out of TSMC advantage over Intel node which I assume won't last forever and if TSMC blunder even one node it can be catastrophic for AMD. Considering USA shift and focus on tech war with China Intel fabs can only grow faster or everything can crash.

We just ordered 4 new dual Xeon workstations for the office.

What kind of jobs do you run on them?

> everything I've heard over the last couple of years indicated that AMD was absolutely crushing Intel.

AMD market cap: $93B

Intel market cap: $104B

And now you will no longer be able to say that everything you've heard indicates that AMD is "absolutely crushing Intel".

Given their respective revenue, AMD's market cap does indicate they're crushing it.

Their multiplier means that the market thinks highly of their ability to continue to grow.

Given that AMD shares are down more than 61% YTD (vs 52% for INTC), I'd say neither company is setting positive expectations for the future.

*edit added YTD for clarity

I've seen another comment saying their revenue is way up YoY. What does this mean when compared with share price changes?

AMD market cap in 2015: $2B

Intel market cap in 2015: $150B

Kinda looks like crushing.

Reply on news.ycombinator.com