So does it seem plausible seeing this product by the end of this year? Do you think it’s worth waiting for an upgrade, or a 32-core version makes more sense for a single workstation?
Also do you think RTX family cards make more sense for an upgrade for using in 3d-graphics than the previous generation?
Is 64-core Threadripper a possibility?
mister3d
#1
If you have the money, that’s up to you. But high core counts tend to have slower core speeds so anything that’s not multi threaded won’t be able to use those other cores and will be a bit slow. It’s best to check the benchmarks once it comes out.
As far as the raytracing cards go, if you want to use that feature then it’s good benefit, otherwise there isn’t as much of a speed increase otherwise over previous generations of graphics cards. But you can use the raytracing hardware now for things like Vray and some game engines if that interests you.
cookepuss
#3
The real issue here is one of price VS performance.
You can’t go in thinking that a 32-core will perform 8x faster than a 4-core. It never scales linearly. For example, one might think that an i9-9960X 16-core must be twice as fast as the 8-core i9-9900K. Nope. It’s only about 40% faster, which is great, but doesn’t nearly justify a 3x+ cost.
Ultimately, I think that the cause seems to stem from the basic architecture. Since these multiple cores are on the same die, there are various issues of bandwidth and latency since they have to share resources.
I think that you’d also be aware that more cores also require more power and produce more heat. That jump from 8 to 16 core can well increase the power consumption by an extra 75%. AMD’s 32-core is a energy hungry beast consuming a good 250W.
You also can’t assume that a 32-core AMD offering will perform the same as a comparable CPU from Intel. For example, AMD and Intel both have 16-core CPUs. AMD’s version is even clocked higher. However, AMD’s 16-core is actually about 12% slower, putting it in the same league as Intel’s 14-core…
Further proof that the number of cores doesn’t always mean anything… That 32-core 2990WX from AMD is actually about 10% SLOWER than their own 16-core 2950X. However, because they give you 2x as many cores, you graciously get to pay 2x as much. 
AMD’s single greatest advantage is price. That’s about it. Intel is frequently much more expensive, relative to core and speed.
DanHibiki
#4
yeah we got a couple threadrippers and frankly they don’t render much faster than other cpus. Multithreading is still not that great for a lot of processes, these super high core count CPUs are best used for servers and Virtual Machines.
For sheer render speed to cost ratio it still seems to make more sense to get multiple lower end systems than trying to cram it all in one case.
mister3d
#5
Thank you very much for replies and insights!
So what a single processor would you recommend for a freelancer? Would it be Intel family?
cookepuss
#6
IMO, I’d go with Intel. Yeah. AMD chips usually offer a better price to performance ratio. Intel’s packages are better engineered overall though.
As for which Intel CPU to choose, it all depends on your budget. More money generally means more power. However, you do reach a point of diminishing returns eventually. I opted for the Core i9-7920X 12-core. It’s super speedy. It performs just as fast as AMD’s 32-core Threadripper 2990WX and costs about $700 less. ($950 VS $1,600) Half the cores. Same speed. Less money. Intel wins in this particular head to head battle. Why did I opt for the 12-Core i9 instead of something beefier?
- More cores, as you can see with the AMD, doesn’t automatically mean more power. Architecture being what it is, this stuff doesn’t scale linearly. The 18-core i9-9980XE offers 50% more cores than the i9-7920X, but only brings an extra 26% more performance.
- The price doesn’t scale linearly either. The 12-core i9-7920X costs less than $1k However, the 18-core i9-9980XE costs about $2k. You end up paying 2x as much for 50% more cores and only a 26% speed boost.
Ultimately, how much value you put on that extra 26% performance boost all depends on you. If you stand to earn a lot more than that and always face tight deadlines then I can see spending the extra. However, such a CPU is really more an enthusiast thing. Such packages are only worth it if you’re a super hardcore gamer and see value in every extra frame or are into mining crypto. Beyond that, nah, I wouldn’t pay $2k for a CPU.
Which CPU to choose for YOU? I’d probably go with the 8-core i9-9900K if you’re watching your pennies. It’s only 8% slower than the 10-core i9-7900X, but costs about half the price. $500 VS $1k. It’s real steal. The 8-core also consumes about 50% less power. 95W vs 140W.
Whatever CPU you choose, just be aware of the following.
- Get a good PSU that offers your room to grow. Keep in mind that they become less efficient and end up outputting less power over time. Try not to settle for less than an 80 Plus Gold in terms of efficiency. Platinum and Titanium are, of course, much more efficient. The better the PSU, the lower your annual energy bill. Find a modular or semi-modular PSU too. Having fewer wires to connect can improve overall airflow.
- Don’t skip on the cooling solution. Stock air coolers typically stink. Go with a closed loop liquid cooler and sizeable radiator. These multi-core CPUs run hot, especially under a full load. A closed loop cooler is pretty much a set it and forget it solution. You usually don’t have to worry about leaks or having to replenish coolant.
- If you’re building the system from scratch, go with a big case. IMO, if you’re going put an i9 in your new system, you REALLY want to avoid a mid-tower ATX. You’ll have too little room to maneuver components, even less room for air to circulate., and probably almost no slots for extra fans. For a high performing CPU, don’t settle for less than a full tower ATX. If you can afford it, go with a full eATX solution. Some are gaudy as hell and heavier than a sack of rocks, but all tend to offer lots of room for extra fans, big coolers, slots for tons of HDD storage, have tons of airflow, and can fit any mobo you throw at them. Many full eATX towers are large enough to fit two full PC builds. If room for growth and cool temps are priorities, NEVER go with a mid-tower. They’re space saver boxes to be sure, but can shorten your PC’s lifespan because they’re usually hot & cramped little dust collectors.
There are various sites online where you can virtually prototype your next PC and test/price components. How much you spend is up to you and what you stand to earn as a freelancer. If you’re keeping the PC for at least 3 years, try not to spend less than $2k. The CPU and GPU alone will eat up most of that. Once you get past $7k you’re now in a whole other class of PC performers and have lots of other decisions to make.
BTW, if you can, avoid sketchy 3rd party builders like CyberPowerPC and IBuyPower. With lots of power and rock bottom prices, the deals seem too good to be true… because they are. LOTS of negative reviews. Used components. Wires not connected. Scuffed cases. Wrong components. Terrible customer support. Costly shipping costs related to repairs. NOT worth it. Either build it yourself or go with a more seasoned builder like Dell or HP.
Oh, yeah. The core-i7 line still seems to offer some nice CPUs. I would probably steer clear though. While they’re cheaper and perform nicely, the line is near its end of life. It’s become too costly to squeeze out extra performance from the i7 line now. The 10-core i7-6950X, for example, is nearly $200 more than the 14-core i9-9940X. It also performs about 28% slower. Not worth it. At this i7’s performance level, you can get a similarly performing 8-core i9-9900KF for 1/3 the cost.
Just do you research. You’ll see what I mean.
mister3d
#7
A very intelligible survey on the subject, Cookepuss! So thank you very much for your time and efforts.
I currently have 3930k, which starts showing its age in terms of speed. I was reluctant upgrading due to CPU market stagnation, which was for nearly 5 years.
I’d go with a processor you’ve suggested, as my budget for CPU+motherboard+memory is around 2k, so I should be fine with i9-7920X. I don’t want to pay over the odds for the horsepower.
7920k seems to be around 3x faster in rendering than 3930k, which isn’t bad at all.
https://benchmark.chaosgroup.com/next/cpu/details?hw=Intel(R)+Core(TM)+i7-3930K+CPU+%40+3.20GHz+x12&page=1
https://benchmark.chaosgroup.com/next/cpu/details?hw=Intel(R)+Core(TM)+i9-7920X+CPU+%40+2.90GHz+x24
I recall buying 3930k for about 600$, so it was under 1k upgrade. So yeah, it doesn’t scale so well even after so many years. I’d expect something more impressive after so many years, but i’s not too bad.
The only question if we can expect anything competitive in the foreseeable future. Even if a 64-core comes out, it will cost around 4-5k to build.
Definitely I heard some complaints about AMD components over the years, despite being cheaper for professional work.
Indeed 26% increase isn’t worth investing additional half the price increase, this is quite helpful. I’m surprised though that cores don’t scale linearly.
About the CPU cooling part, I’ve been watching LinusTechTips, and he mentioned that the most effective cooling is still Noctua air cooling. It needs some further research though.
mister3d
#8
Can someone explain to me, how can AMD Ryzen 9 3900X, having 12 cores, beat Intel monsters with 24 cores? https://www.cpubenchmark.net/high_end_cpus.html
DanHibiki
#9
because the number of cores is relevant only when comparing to cores of the same type. Different CPU and GPU cores have different speeds, which is why you have to have bench marks to confirm their relative speeds.
sentry66
#10
passmark is a bad benchmark when comparing multi-threading - it’s so far off and doesn’t scale well across multiple cores. It’s fine for other benchmarks, just not multi-thread
64 core threadripper isn’t coming guys, 64-core epyc is. It will be a $3000+ CPU. Several weeks ago, thei Threadripper update was delayed.
Threadripper will stay at 32 cores. They may do a 7nm die shrink on it from its current 12nm to lower manufacturing costs and reduce power consumption. Maybe even a price adjustment. They don’t want threadripper to cannibalize their epyc sales. 80 core Epycs with multi on-board GPU’s aren’t coming until 2021, so 64 cores wil be the top-end until then and Threadripper isn’t AMD’s top-end CPU.
polygnsyrup
#12
Love my AMD Threadripper 2990WX, Over 200% render speed increase(no OC) from my former rendering cpu; Xeon e5 2690 v3 (Cinema 4d physical render)
You want to wait for newer version of Threadripper soon with 10% speed improvement, I don’t think 64core will come out though.
3Dspline
#14
I’m using the 2970wx because the 32 core is almost an extra $1000 cad for just 8 more cores.
Its amazing, runs really great and was cheap’ish
Be careful on what you read that is against AMD or high core counts, a lot of the benchmarks and specs were made when they first came out and things like windows just didn’t support it correctly until it got patched.
But do your research regardless, personally I’m happy AMD stepped on Intels foot, they haven’t innovated anything when they had no competition and look at them now. This competition is healthy for 3d artists!
nejck
#16
No offense but I honestly don’t know how @cookepuss came to those numbers. Quite frankly I am a bit perplexed 
A 7920x is a fair bit slower than the 2990wx in V-Ray, Corona and Cinebench. Those are real, production renderers for a lot of the tasks and the numbers can easily be verified.
Anyone can check how a given CPU performs in each of the above listed renderers. An example: A 7920x is going to render an image for 1h:15min while the 2990wx will render that same image for 0h:45minutes. I’d say that is a huge difference.
The Intel platform pre Zen 2 did enjoy a fair lead in overall performance but that is mostly in apps that we, the 2d/3d people, don’t use that much / often enough - WinRAR for example, the 2990wx is barely any faster that then 9980xe despite having many more cores. Handbrake video conversion, the 9980xe is about the same speed as the 2990wx while the 2990wx still leads in some of the editing tests. Point being, not everything is for everyone.
Now, if you are mostly doing single threaded tasks (modeling, sculpting…) then yes, a 9900k or a 3900x/3950x will be faster than both the 7920x and 2990wx. There is even worse scaling in non-HEDT parts like Xeons but guess what, people still use Xeons as workstation CPUs.
Why is that? If your modeling operation takes 4 seconds to complete instead of 2 seconds but your render finishes in 1 hour instead of 6 hours, there is a lot to gain there. Conversely, if your Photoshop filters take 50 seconds to apply instead of 45 seconds and you render at twice or three times the speed… Well in that case you kind of need to decide which is more important to you.
That tradeoff right there has been in place since the good old days. The more cores you have, the more heat issues you run into hence the lower clocks.
My opinion is that generally speaking, if your CPU is at least Haswell+ / Zen 1 then anything above 3.5ghz on a single thread is going to be harder to notice - unless you do simulations bound on one core. If you are sporting a newer architecture then you are probably hitting close to 4ghz with a higher IPC anyway so its even more of a moo point - except if you do a lot of sims. Still, if 90% of the time all you do is sculpt or model then imho the Threadrippers and Xeons are probably not what you’d want. Better invest that money elsewhere - broadly speaking that is.
I’d ask everyone to use the benchmarks we have and investigate this stuff by themselves. I don’t think you can go “wrong” with any of the modern CPUs if you are mostly concerned about single threaded performance. This isn’t gaming on a 240hz screen and even there a 5% difference can sometimes be negligible.
Just to reiterate… Find your fit! If 90% of the time you do 3d modeling + 3d rendering then I personally can’t recommend the slower 9980xe over say a 2990wx. If you mostly render and can afford it then look into Higher End Xeons in 2 Socket configurations. If you do all sorts of stuff and you rarely do any rendering, look into that 9980xe / 3950x / 9900k or maybe wait for the next gen Threadripper / Intel stuff because if not anything else the prices will apparently drop even further.
Also, be mindful of the platforms. Zen 1 Threadrippers do not have motherboards that have 4+ PCI-E slots while the Intel 9980xe has a few of those. If that is of any concern that is… (It is to me actually, but I run riser cables for my GPUs).
As for the GPU vs CPU debate, I’d rather not get into that 
Both have their pros and cons imho, depends on what you do.
edit: for the record, I currently sport a Dual Xeon system myself.
polygnsyrup
#17
Actually … 64core Threadripper is likely coming by end of this year with roughly 30% faster than previous gen. wow.
mister3d
#19
AMD officially announced ROME with 64 cores, with its price perhaps under 7k. It’ s unlikely they launch Threadripper with the same amount of cores.
maytuxa
#20
single socket processors are cheaper.
64cores 7702P is just 4.5k 
rumored: for threadripper 3gen will be 2 types of motherboard. with 4 (as previous) and 8 (without overclocking possibility) channels of memory.
mister3d
#21
Still dubious, but rumors it might actually come to life
Anyway, it’s not too long to see if it’s coming to realization. I ponder how it competes with Rome server chips? Why would they make such a beefy maintream CPU?