New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
65W vs 175W Ryzen 9950X?
I recently make a post for a suitable VPS to host a minecraft server. and learn about advinservers.com. but there's a different plan for the location I need which is only 65W. How's the performance compare to the full 165W?
Comments
All-core frequency drops from 5 GHz to 3 GHz when you switch from 175W to 65W.
In some or even many use cases, you might never notice the difference, but it uses significantly less power and generates much less heat.
I guess a lot of providers could save a good amount of money by switching to 65W, but the LET drama queens will probably go ballistic when they find out.
See: https://www.club386.com/heres-what-happens-when-you-run-an-amd-ryzen-9-9950x-at-65w/
Could save even more by not buying that good a CPU in the first place.
This is first time I see this feature being used for server usage
huge drop in multicore performance (around 50% less than full tdp)
same single core burst
however most providers won't give u an unlocked chip (due to thermal constraints and power consumption)
@advinservers, how big is the performance difference between the two setups you are selling?
Oh yeah, I'm wondering this as well. At least, I felt a noticeable difference once I migrated one of my client site from an EPYC 7000 series to a Ryzen 9 9950X VPS, and it made the site run smoother.
I’ll respond to this thread with benchmarks in a little bit.
The AMD EPYC 4545P is essentially an AMD Ryzen 9950X that has been set to use 65W. Instead of that, we prefer purchasing 9950X and using the Eco mode preset in the BIOS to set it to 65W.
They perform the same as the EPYC 4545P with similar power usage. We generally prefer 9950X because availability is much better.
I prefer when my 9950x3d consumes 200w.
you guys should've seen this man boasting about the 4545p before release then being let down.
4545P???!!!!??!?
What’s the point of getting the 9950X then? Those benchmarks can be pretty misleading. There’s a noticeable dip in single-core performance, and the multi-core results are underwhelming. Plus, try running benchmarks on a fully loaded system — I’m sure the single-core performance will tank even more.
And As a user, why should I care whether it’s using 65W or 170W, unless those power savings actually benefit me in some way? I am paying for 9950x, and I expect it to perform like it should.
From the YABS above, the performance seems comparable to the Genoa 9454. That 65W mode might look good in benchmarks, but it feels more like a gimmick. You are saving power, sure but the performance is taking a hit at the same time.
The multi-core score of 3658 could probably be matched by just one core of the Ryzen 9950X. 2 cores should do atleast 6000 gb6 so there is roughly 50% drop in performance than normal 9950x.
I'm not sure which node the OP has their VM on. It would be incorrect to assume that the hypervisor it's being ran on is empty, given that we are sold out on VMs in APAC. I'm not even sure if their VM is running on a 65W TDP hypervisor.
It is benefiting you with a lower price. The problem is that if we were to keep it at the normal TDP in Asia-Pacific, we would need to charge a lot more to make it a viable product.
For example, instead of $10/month for 8GB RAM, 256GB NVMe in APAC, we would need to charge $16/month.
Some might pay that, sure, but that is not the market we are targeting with this lineup. I believe that the price we are offering is pretty fair for the resources, in an underserved market like Asia-Pacific.
In fact, $16/month for 8GB RAM was the old price when we used to sell VMs on a 9950X 170W hypervisors. The cost savings are directly passed onto you, we aren't trying to be greedy.
That is why it is very clearly marked as a 9950X 65W on our website, and you are paying for 9950X 65W worth of performance.
We are not hiding that in any way, it's not something that is hidden in our knowledgebase at any point of time. It is quite clearly on the order page, hence this thread.
If you are ordering this product, you should expect it to perform like an AMD EPYC 4545P. This is only a feature in our Asia-Pacific location. The rest of our locations run at their normal TDP. This product lineup is still experimental and in its early days, so we are still making adjustments and taking in feedback from our customers. We could add a selector in the future.
We are completely open to suggestions/feedback and we really appreciate the community input on this.
So let me recap @advinservers. You are aiming APAC interested customers with cheaper 9950x with lower performance and this is mentioned as 65w mark.
Maybe it would be better to show the performance difference between 65w vs 170w on your page. Saying like 65w would perform x% percent slower compared to 170w. Or add info about 65w 9950x will actually performs like 4545P.
But probably you wouldn't do that as you would lose some customers.
who on LET offers 2 cores on a 9950X at full power cap in APAC for $10?
I mean, honestly, that's what I thought as well, like wouldn't that be kind of misleading advertisement? Then I realized, people buying @advinservers have enough IT knowledge to know the implications of running a CPU at -62% TDP.
It is already somewhat in our knowledgebase:
https://docs.advinservers.com/information/hardware#tdp-limitations-kvm-frequency-vps-johor
The problem is that there is not really a particular x% percent difference in performance. The performance difference will depend on the load of the hypervisor. The main difference is really the all-core clock speeds.
For bare metal:
Passmark shows a 16% difference in multicore, 3.8% difference in singlecore: https://www.cpubenchmark.net/compare/6758vs6211/AMD-EPYC-4545P-vs-AMD-Ryzen-9-9950X
Geekbench5 shows a 15% difference in multicore, 0.3% difference in singlecore: https://browser.geekbench.com/v5/cpu/compare/23637938?baseline=23637930
But this will not really apply to VMs. Let's say, 12 cores are loaded out of the 16 cores available. The clock speeds would decrease and the multi core/single core performance will be vastly different compared to, let's say, a situation where 4 cores are loaded out of the 16 cores.
We would definitely like to do something like that, but it is really hard to contextualize the performance difference.
If any customer tries out a VM and has performance troubles, we have a 72 hour money back guarantee. We refund all payment fees and there are no questions asked as long as it's not through cryptocurrency.
We care about transparency. This does not particularly matter to us.
Will there be more options in the future?
I imagine that you're putting these on 1U's with a $10 cooler and that makes a substantial difference in cost vs just the electricity. What's a 175W 1U or 2U cooler go for?
really?
$10/month for 8GB RAM, 256GB NVMe in APAC = 65W
$16/month for 8GB RAM, 256GB NVMe in APAC = 170W
There is not really an upfront cost difference. A liquid cooler is about $30 extra. On a hypervisor that costs almost $3,000, it is not a huge difference.
An air cooler is still around $50.
Maybe, we’re still looking into it.
It still 10% more performance than AdvinServers' EPYC 9654 APAC plans, sure it has two less core but at the same cost. It still benefit people who need high single core performance. My only worry is that it has small plans and with the number of threads 9950x has, it feel like the CPU can easily oversaturated if the node has many small plans users.