• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

CPU review process

Solaris17

Super Dainty Moderator
Staff member
Joined
Aug 16, 2005
Messages
27,792 (3.84/day)
Location
Alabama
System Name RogueOne
Processor Xeon W9-3495x
Motherboard ASUS w790E Sage SE
Cooling SilverStone XE360-4677
Memory 128gb Gskill Zeta R5 DDR5 RDIMMs
Video Card(s) MSI SUPRIM Liquid 5090
Storage 1x 2TB WD SN850X | 2x 8TB GAMMIX S70
Display(s) 49" Philips Evnia OLED (49M2C8900)
Case Thermaltake Core P3 Pro Snow
Audio Device(s) Moondrop S8's on Schitt Gunnr
Power Supply Seasonic Prime TX-1600
Mouse Razer Viper mini signature edition (mercury white)
Keyboard Wooting 80 HE White, Gateron Jades
VR HMD Quest 3
Software Windows 11 Pro Workstation
Benchmark Scores I dont have time for that.
Hey,

I am building a fun system; more of a project then something that will replace my main system and was shopping around for a CPU for the motherboard I chose. Now given the nature of this machine the CPU itself didn't matter as much, but my curiosity lead me to the 9950X3D review (which is what I decided to go with,.)


I have to say I REALLY appreciate this format. There is a lot of data and extra testing in written form like this, or at least from you @W1zzard specifically then I can really find online for consumer products. Generally I am working with server or workstation which understandably aren't generally benchmarked and put on tech sites. However; the cross over for people like me matters.

Specifically:

Starting with probably the most common of the uncommon tests due to rising popularity is the AI tests.


Classification, upscaling and the training portion being what I care the most about. If I make one recommendation, if possible I would look into updating and moving to a new GPT model or one of the other new models available. GPT-2 is great for consistency, but at least imo not so much for relevance.

The next would be the AV page:

Not a lot on this page, but I appreciate the Altium benchmark (I also think you have done pcbway before?) I admit that for the most part anything modern leaves little benefit to smaller projects like the ones I do (badges and cards) I do pay attention to it and information like this is almost impossible to get elsewhere.

Finally the workstation tests:


All are valuable to me in some regard. I don't use vbox professionally of course. The systems I generally touch are the big 3, Hyper-V, VMWare, Nutanix. With a ton of experience in Proxmox and XCP-NG. But I can do my own tests and see if I am in the ballpark of your numbers, and while I think its a concession to even get this test, and this CPU is not going to be anywhere near the production OSs listed above it might be nice to one day see one tested instead. It would also be neat to maybe benchmark a local docker install, be it windows or otherwise. (since I see a lot of devs spinning local sessions for testing before commit.)

Databases are really important, this tells me a lot about the architecture in general, and while this data is more readily available there are some cases in which enterprise trails consumer; specifically with refreshes, so its nice to get this data ahead of time. I would appreciate a Postgres addition though.

Web hosting is important to me personally, and I always look at it. ASP isn't something I generally deal with but SharePoint and other apps are built on it so it matters. It would be nice to see something like nginx one day though. Even if its less app and more server.

Honorable mentions:


These tests don't concern me much at all, since the clock rate of consumer CPUs is too high compared to what I will be dealing with. While the throughput of multi threaded apps is generally hundreds of times higher on the CPUs I'm used to dealing with, (which makes these comparisons almost impossible because I cant really project performance based on these clock rates) I always find myself looking at the cryptographic portion. A lot of specialty data is handled on ASICs and other custom boxes where I work and with what I deal with, but I still use a ton of COTS parts and machines to serve VPNs encrypt data in other ways etc and its nice to see these results.

and:


Thanks for the C++ compile times.

bpvo5w9e040e1.jpeg


I think there is likely a surprising amount of readership that looks at these pages, even if the comments in most of your reviews for both CPU and GPUs center more around the gaming or more consumer workloads, and that's understandable given TPUs target audience, but I am sure I speak for at least a handful of us that do pay attention to pages like the above where things like that can impact us professionally, and you are one of the few people that actually do it. So thanks a ton for that.
 
Last edited:
I actually fully agree - I pay attention to professional workloads even when it’s not something that I use. Simply for the reason that it allows me as a curious enthusiast to gauge the performance of an architecture and efficiency of changes done to it more readily since those ARE the workloads that those architectures are actually being built for. I have been saying it for years, but while consumer and gaming benchmarks are very relevant to the mass audience and undoubtedly important, reality is that for games your GPU will be magnitudes more valuable and for general consumer tasks like Office and Web Browsing ANY reasonably modern CPU is absolutely sufficient - we’ve hit the point where CPU performance barely matters for those usecases. Same reason I usually skim with interest the Phoronix data - you can see clearly there, for example, that Zen 5 is absolutely hitting what AMD actually needs from the arch, all jokes about “Zen 5%”’on the consumer side notwithstanding.
 
professionally
I appreciate it, and yes, my benchmarks are geared not only towards gamers.

Altium benchmark (I also think you have done pcbway before?)
Only Altium, it's industry standard, no plans to replace with anything else. Also took me long enough to come up with a good test sequence

I don't use vbox professionally of course. The systems I generally touch are the big 3, Hyper-V, VMWare, Nutanix
Everybody hates VMWare and VMWare Workstation is kinda EOL. Not even sure what Broadcom plans with the rest of their offerings, except for extort more money, so won't touch.
Hyper-V has too much OS integration. Nutanix really has no Windows offerings?

Not sure if keep on VBox or just drop, really dont want Hyper-V, which seems to be the new standard though

What do you have in mind? Just serve a gazillion static pages per second?

Generally I'm always open to explore adding new workloads, but they have to be somewhat real-life, run the real app
 
Not even sure what Broadcom plans with the rest of their offerings

Not sure either, and given the pricing increases I don't really care. Everyone I know is trying to get off. Maybe Proxmox w/ LXC or using a terraform provider. Vbox can really stay, it was just an idea, not to mention its on windows with exception to hyper-v is pretty much the only realistic option.

What do you have in mind? Just serve a gazillion static pages per second?

Generally I'm always open to explore adding new workloads, but they have to be somewhat real-life, run the real app

You are probably right. Nginx is too fast to get any real benefit from benchmarking it specifically, I was hoping there might be a use case for more real world uses like reverse proxying to other apps and caching static content, but at that point you are basically doing a memory or disk subsystem benchmark.

Thanks again for the format its really great.
 
like reverse proxying to other apps and caching static content, but at that point you are basically doing a memory or disk subsystem benchmark.
Reverse proxy is limited by the upstream app. and for static you're basically testing disk cache (= memory), or memory + disk if working set > disk cache size. I guess that could be a reasonable test for a CPU review .. nginx + 10,000 x 1 MB = 10 GB of static files, random url selection, a ton of workers fetching and measure throughput
 
Back
Top