Hacker News Reader

Comment by soygul | Hacker News Reader

It is absolutely on top of the Performance/$ chart on PassMark: https://www.cpubenchmark.net/cpu_value_available.html

It offers exceptional value. Pair that with a GeForce GTX 1650 SUPER which is at the top of Performance/$ chart for GPUs: https://www.videocardbenchmark.net/gpu_value.html

And you have a champion of a workstation right there.

View on HN

And this is hardware you can order today and have running under your desk by the end of the week.

This is such a welcome change from the crypto days with GPUs in short supplies and at very high prices, and stagnant Intel CPU performance!

I’ve been trying to upgrade my desktop since the motherboard is dying and I’ve found that small factor pc (SFFPC) components are rarer and at even more of a premium than usual. Most commenters in various forums chalk it up to the plague, so now I’m hoping once the next generation of CPU, motherboard, and GPUs come out things won’t be as scarce.

Interesting you mention this, because I noticed a surprising surge out of nowhere among my friends, who suddenly all decided to build a SFPC. Despite none of them ever being interested in SFPC and all of them running mid/full-towers prior.

None of them are first-time PC builders tho, so it might have something to do with lockdown-induced boredom, and my friends deciding to challenge themselves to build a SFPC. As it sounds like a major pain and challenge compared to even a mid-tower (according to those of them who have already finished).

I won't be building a new computer for a few more years, as I currently can't justify the cost of replacing my build, but when I do it'll almost certainly be a SFF build.

I've seen a number of people who have built tiny custom cases, and the challenge of both building something like that and getting everything to fit in it is super enticing. Definitely seems like the challenge and cool factor of fitting tons of power into something tiny is making a lot of people consider such builds over traditional mid or full tower machines.

For me it was because I’m moving soon and wanted to have a smaller computer to move. The shortages put a stop to that though.

> I’m hoping once the next generation of CPU, motherboard, and GPUs come out things won’t be as scarce

New desktop Intel 10th Gen CPU are actually coming out pretty soon. It's got a new socket (LGA 1200), so new motherboards are coming out for it as well.

It's only marginally faster on single thread benchmarks than the Ryzen 3000s (ie Zen 2) chips, and consumes a lot energy.

Also, Intel is still using their circa-2104/ancient 14 nm process for these desktop 10th Gen chips, which is quite disappointing. No Ice Lake 10 nm desktop chips yet.

I was most likely going to stick with and AMD CPU, the intel power consumption and price per performance is a bit high for me. I’m mostly developing javascript and gaming so ultra high cpu performance isn’t as necessary.

> javascript and gaming

Both of which are heavily impacted by single-thread performance. So getting a high single-thread perf CPIU would be a good thing to do.

The whole crypto thing is such a fascinating experiment of human behavior and side effects and so on.

Glad I bought mine before that craze. Has that opened a cheap used GPU market or did they burn them out?

Yep, rx 570/580s are selling for great used prices on eBay. I pretty much recommend them to anyone going a budget build.

I'd definitely recommend against buying second hand GPU from crypto miners. They have been worn out running 24/7 for years.

If it's selling for $30 I don't say, but I've seen them going for a fair bit of the original price.

The thing is I've never worn out a piece of equipment in 25 years of computer enthusiasm. I've had some DOAs and infamous equipment (say Deathstar) but never failures even second hand equipment. Is that different for 24/7 max power draw or is it just YMMV (and usually will still work)?

I've seen every single piece of equipment worn out over the past decades except the CPU.

Hard disk and GPU are the worst in my experience. They have the shortest lifespan and they're the most impactful when they die. Hard disk die suddenly destroying all data, self explanatory. GPU die slowly, progressively having rendering error then crashing the system erratically.

I've taken 3 GPU to their grave between my personal computers and my family's.

To limit my spending I have a personal policy where I don't upgrade my GPU until it dies (I don't generally play the latest AAA games on my PC) and with just one outlier since 2004 I've had a new one about every 4-5 years. I've actually never had a CPU fail but I've had several HDDs die.

Nah if your HDD is dying it’s either your task really demands it OR you’re not cooling them enough.

If your GPU is dying it’s VRAM or VRM is failing under high heat. Manufacturers cut fan speed to cut fan noise until warranty claims increase substantially so they never provide sufficient cooling at stock settings.

Rackmount servers don’t sound like jet engines for no reason. Things fail if you don’t do that.

I've had hard disks fail in a PVR setup (heavy load), I've had memory fail, I've had motherboards fail, I've had power supplies fail, and I've literally typed laptop keyboards to death in ~2 years. I'd be utterly unsurprised about a GPU failure.

The best case scenario is "selling it because I'm buying a new one", the worst case scenario is "selling it because it's starting to flake out", and "selling it because the warranty ran out" is not much better.

Storage fails all the time. Cables, power supplies, UPS, monitors, keyboards, mice - I have a graveyard in a cabinet.

Also motherboards who's capacitors dry up; connectors that develop faults.

This is of course leaving out devices who's firmware get exploited and become poison (routers, USB devices with malware)

It's both.

The ymmv is always big with electronics. I.e. my second last graphics card (gtx970) died within 2 yrs. But the aging effect from running 24/7 under high load is very real as well.

The high temperatures of permanently running on full power just wears it down

Besides mechanical parts you can still see failure from weary chemical components (capacitors, easily fixed), solder joints (fix depends on the joint. BGA chips aren't easy to fix) failing due to cycles of thermal expension and retraction or water damage, and lastly electromigration (the real unfixable damage, it usually happens from overvoltage/overheat).

At worse you might need to spend 20 bucks to replace tired fans.

Mining GPUs run undervolted without powercycling. They're fine.

Eh, I mined on some GPUs for ~a year. They're all still running today in various family members' PCs. Admittedly I was mining Ethereum where the main goal was to undervolt and underclock as much as possible while maximizing the memory clock.

How did you cool them? I mean, didn’t you find stock fans way too weak especially had it been in a “quiet” PC?

All the cards I had were "aftermarket" cards with better coolers (7x dual fan GTX 1070s & Pentium G3258 running Gentoo). It wasn't silent but it really wasn't that noisy. After unvolting and underclocking the GPU core, and overclocking the memory, they ran at ~60C under constant load.

I gave three of them to family members and sold the other 4. I was absolutely upfront in my eBay listings of their history.

I picked up a refurbished Vega 56 last year for a great deal. I suspect it was a mining card. Seems to work fine.

They are fine but you will need to replace thermal compound and reflash bios. The miners usually under clock the GPU and overclock the memory.

It's remarkable how not only the first but also #3, #5, #6 positions are some Ryzen 5 ?600? CPU. The 2600X is the only that is currently lower at #12. Also remarkable how utterly absent Intel is from this chart. The i3 9100F managed to break into #2 and then a Pentium Gold G5400 at #15 but both of these are sub-$75 parts while most AMD parts are significantly more expensive likely bringing more profit to AMD. Usually price per performance goes sharply down with performance, it doesn't scale linearly but remarkably a $405 processor from AMD is in the top 10, the 3900X. Power-performance wise also AMD is the killer, https://www.cpubenchmark.net/power_performance.html the new 15W TDP AMD 5/7 4xxxU CPUs are the first four while most Intel chips at the top of the chart are in 4.5-7W territory. If you look for ordinary socketed chips, you will find the Ryzen 3900, the Intel 9900T (which is a special 35W part), the 3950X and astonishingly the EPYC 7702 at 200W beating the Intel Core i3-1005G1 which is a 15W part on the latest Intel node at 10nm ... that's just embarassing.

I mean #3 is the same CPU with a slightly higher factory clock for $20 more

Ps. The $405 is an Amazon price which is not in stock but bhphotovideo has it in stock for $410.

Some days ago I checked https://opendata.blender.org/ and noticed that the RTX 2060, RTX 2060 Super and RTX 2070 are all equal when it comes to performance in the Blender benchmark.

Also the RTX 2070 Super might be the cheapest performance card at the moment.

Charts like these can save you a lot of money. But it will depend on the usage of course.

> Charts like these can save you a lot of money. But it will depend on the usage of course.

You have to be careful though. For example, the Radeon RX 570 is right next to the GeForce GTX 1650 SUPER at the top of the chart, but the GeForce costs almost 40% more. If the GPU isn't your bottleneck you may be better off with the less expensive one and save money to buy a faster CPU or more memory. Whereas if it is, you may be better off spending more for a faster GPU than either of those.

What these charts are great for is to look at the top fifteen or so as the set of candidates, which will all have good performance/$, and then if you prioritize better performance look at the fastest ones and if not then look at the least expensive ones.

But even then you have to be careful. For example, on the CPU chart the Core i3-9100F is one of the best values. The Ryzen 3 3200G is only slightly faster for $17 more (i.e. almost 25% more). But the 3200G has an iGPU, which is worth a lot more than $17 if it means you don't have to buy a discrete GPU.

this is why I favor looking at the XY scatter charts rather than a made up heuristic like G3D/$. for instance those 2 gpus in you post will almost lie on a straight line through 0 without anything else on lower slope side. but we dont actually have to get the absolute lowest perf/$ & looking at the scatter chart we can bring in those other factors into play too.

In addition the most cost efficient alternative may not be good enough for your particular use-case and expectations, and then it wasn't a saving at all.

Also consider power consumption. Polaris RADEON is cheaper but power hungry compared to Turing GeForce.

The power consumption is only significantly different under load. If you're in the corner case where Polaris Radeon is just fast enough even though you're constantly pushing its limits then this matters, but generally if you're running it at full load often enough for power consumption to be meaningful then you should be looking at the faster GPUs to begin with.

Yes, 1080p at 60fps which is still the most popular resolution out there according to the Steam Hardware Survey.

https://store.steampowered.com/hwsurvey/Steam-Hardware-Softw...

I wonder why the GTX 1080 isn't in the videobenchmark all-time value chart.

Because their chart is determined by benchmark score / current sale price. The 1080 wasn't heavily discounted when the 2xxx series was launched, and high end GPUs are rarely value for money winners anyway as cost scales more than linearly with performance. Double the $ might get you 25% more frames.

For context, their page on the 1080 (https://www.videocardbenchmark.net/gpu.php?gpu=GeForce+GTX+1...) gives it a score / $ ratio of 20, while the lowest item on that page has a score of 30.

You could argue they should have some minimum threshold for performance before including them on the chart, certainly no one should be considering a r7 260 or gtx 770 for their new system, however cheap they've gotten, but they don't.

probably has more to do with their price scraping system than it does with the card's performance

I'm surprised the 1650 super is ~ 30% cheaper than the 1660 super but it looks like current prices pretty accurately reflect original MSRPs

* https://www.tomshardware.com/reviews/evga-nvidia-geforce-gtx...

vs.

* https://www.tomshardware.com/reviews/nvidia-gtx_1650-super-t...

Does anyone know if the GeForce GTX 1650 SUPER, combined with a 3600x, is able to power a 4k monitor at 60hz?

Note: this is NOT for gaming. Just everyday workflow and dev environments in 4k, 60hz.

If you go on NVIDIA's website you can find the specifications of the card where it states the maximum resolution and refresh rate is as follows:

7680x4320@120Hz [1]

[1] https://www.nvidia.com/en-us/geforce/graphics-cards/gtx-1650...

You don't even need a discrete GPU for that.

Without a doubt. For comparison a 13” MacBook Pro can drive two 4K displays plus the internal display at 2880x1800, all at 60Hz, from an integrated GPU.

You can probably run as many 4K monitors as the 1650 has video ports.

Yes, easily. Laptop 1050s could render your desktop in 4k60.

I've had a 2600 for a year or two. I was looking to see what I'd gain from an upgrade to the 3600..Interesting to see the new 2600s still compete in the value department...now that their prices has been reduced by about $50.

When I was building a workstation in late 2018, it was a close match for price to performance, even after getting a 50% discount on Intel HDET processors from a friend! I'm happy to see AMD killing it in the CPU space.

With the proviso that the perf/$ approach works less well on GPUs cause you need to match that with what type of screen res & refresh you want to drive (if gaming I guess).

..which might easily land you in 2060/2070 territory.

Wow, looking at the XY scatter graph there shows a pretty stark contrast.

It's neat to see that the 3900X is on the zen 2 threadripper price/performance line, but is available at mortal consumer prices and on the low end platform (B450). The numbering scheme makes sense in the context of this plot, even though the 3900X is not threadripper while all other 39X0X parts are.

The 3950X is also an AM4 part (non-threadripper).

Interesting. AMD dominates the entire top 10 of that list.

Indeed! I knew that AMD was thoroughly on top, but that list really shows by how much. The first Intel in that list which I'd consider for a mid-range workstation (the i5-9600FK) is way down at #22.

i wish there was also a column for Watts/CPU Mark, this matters to me and i'm sure AMD kills it here, too.

It matters to me too.

Sorta here: https://www.cpubenchmark.net/power_performance.html

Unfortunately it's hard to tell which are embedded versus traditional sockets.

And limited to x86.

I've heard that Intels idle at lower power. Ryzens do not go below 18W, but Intels, even ancient Sandy Bridges idle at 5W. Not sure if it is true though.

That's roughly equivalent to Cinebench points per dollar, which is unrelated to gaming performance, if you want that. Doesn't mean the 3600 is bad, but it does mean that these kinds of synthetic benchmarks tell you very little about gaming performance, which is less about the speed of the cores themselves but rather memory latency, something all current Ryzen CPUs are architecturally very bad at (20-50 % slower than Intel).

I was just reading through the Tom's Hardware review of the Ryzen 3600 and I can't see what you're talking about. The gaming performance is within 3% or so of the i5-9600K in basically every game. Granted, these tests were conducted at only 1920x1080, but still, if there were such a dramatic performance difference, it should be noticeable at lower resolutions too.

https://www.tomshardware.com/reviews/amd-ryzen-5-3600-review...

Native gaming isn't particularly cpu-bound these days. I wish I could take the overall performance of a 3700 and divide it between four cores instead of eight. IPC is king when it comes to game console emulation.

That same review has an overclocked i5-9600K vs a stock one and the performance difference is pretty large. An i5-9600K @ 5.0 is almost as significant of an improvement over a stock one as a stock i5-9600K is over the lowest performing CPU of the bunch.

What percentage of 9600Ks can hit that OC and keep stable under load?

~90 %

Bulldozer series were that - 4,6,8 normal cores which could have been glued in pairs to make double wide cores (this also produced a metric ton of comments and articles how FX series "weren't really 8 (6,4) cores"). CPUs were good enough but apparently this architecture was hard to utilize fully (and had other issues, e.g. with IPC).

Memory latency is 20-50 % worse, but that doesn't mean 20-50 % lower FPS. What it does mean is in CPU limited scenarios Ryzen CPUs currently cannot touch Intel's CPUs. Even several year old high end Intel parts outperform Ryzens there, even if the Ryzen part in question has better single threaded throughput (i.e. Cinebench).

At the 3600 price point this is largely irrelevant though, since you will be GPU limited by way of budget anyway.

A 3600 may have much worse memory latency, but it's also got more than twice the cache of a 9900k. At least according to the linus tech tips review, the 3700x beats the 9900k in CSGO, a horrendously single-threaded, cpu limited game.

I don't think CSGO provides much insight, considering that its engine was already outdated for many years on release (2012) and has received few significant updates since. The 3rd party benchmark map that is typically used for these tests does not reflect actual gameplay well. You'll notice that in the benchmark the game is running quite smoothly, meanwhile in actual games CSGO has been plagued by horrid microstuttering and poor framepacing for years (largely caused by the ever-increasing amount of skins and swappable models).

Games like SOTTR provide much better insight into the performance characteristics of modern engines instead of 2004 tech, both in CPU and GPU limited scenarios (depending on settings).

You specifically said:

> What it does mean is in CPU limited scenarios Ryzen CPUs currently cannot touch Intel's CPUs.

CSGO, regardless of how old the engine is, is very much a CPU limited title, and not only do Ryzen CPUs "touch" Intel in CSGO, they beat it.

In SOTTR the 3600 is about on par with a 9600k: https://www.gamersnexus.net/hwreviews/3489-amd-ryzen-5-3600-...

This is true, I’d say the build GP describes would probably be good for someone who is short on cash and wants to build a machine that is OK/Good for gaming and Good for things like programming or photo/audio/video editing. It’s true that that build won’t knock anyone’s socks off and isn’t optimized for anything in particular, but the fact that you can make a build like that with such good parts at that price is a net positive for consumers.

Weird thing about benchmark is that Cinebench score is treated as most important metric. It's very informative score for CG rendering use cases but not for all use cases like gaming, video editing, etc. I suspect that vendors/benchmarkers use it because of Cinebench multithread is scales very well by increasing threads (so easier to advertise higher core count model).

Reply on news.ycombinator.com