What is your existing hardware configuration (component name - component brand and model)
NA
Which hardware will you be keeping (component name - component brand and model)
NA
Which hardware component are you looking to buy (component name). If you have already decided on a configuration then please mention the (component brand and model) as well, this will help us in fine tuning your requirement.
AMD Ryzen 3900X 3rd Gen
Gigabyte X570
Crosair Vengeance LPX 16GB 3200 MHZ *4
Zotac Geforce FTX 1660Ti ZT-T166010F-10L *2
Not sure about which HDD, SSD to go for
Is this going to be your final configuration or you would be adding/upgrading a component in near future. If yes then please mention when and which component
Will Upgrade RAM to a 128GB or more depending on motherboard
Will upgrade Storage depending on storage
Will Upgrade Graphics Card to 2080TI
Where will you buy this hardware? (Online/City/TE Dealer)
Mumbai
Open to online purchase
Would you consider buying a second hand hardware from the TE market
Yes
What is your intended use for this PC/hardware
VAPT
Virtualization
Machine Learning
Do you have any brand preference or dislike? Please name them and the reason for your preference/dislike.
Do not want to buy XYZ company product as I heard the after sales service in India is pathetic
If you will be playing games then which type of games will you be playing?
NA
What is your preferred monitor resolution for gaming and normal usage
Desktop - 1440x900
Are you looking to overclock?
Yes (If Required)
Which operating system do you intend to use with this configuration?
This might be just some random parts listing that i have given here.
Basically I am looking for a good built which can run multiple VM's at the same time (5-7), Graphics Card for Machine Learning, brute force attacks, not looking for fancy liquid cooling would upgrade to a those later if required.
I would be keeping this machine for any time around 3-5 years as my main setup. Checking for a configuration in the same manner
Regarding Monitor looking for an ultra wide monitor 27" and above
Thanks a lot for your response.
The Motherboard specified here can only take upto 64GB of RAM. is it okay for starters? Should I directly go for a 64GB RAM at first itself?
Also regarding Monitor I was looking for an Ultra Wide Screen 27inch + is there any option available with that specs?
Thanks for your reply it really helped me a lot
Amazing!!
Just one thing
Regarding Monitor I was looking for an Ultra Wide Screen 27inch or + is there any option available?
Hi ShriramIyer16,
If you don't mind me asking, what kind of AI application you do? Reason being the system RAM size, onboard RAM on GPU and even the storage space requirements will depend on that.
1. General idea is to have a slow HDD pushing data to a faster SSD which will in-turn have at least 5% of your data set ready at any given time in System RAM. But unfortunately once the data set grows you will start seeing bottlenecks in HDD <-> SSD <-> RAM so it always better to have more RAM, bigger SSD, So please aim for 64GB System RAM minimum with further expansions.
2. Though you may be aware, I would like to say that having multiple GPU's really helps, if not in parallel processing but in running multiple instances of the same data set with different parameters. This really helps in selecting better methods and running experimentation. So when selecting a motherboard please make sure you have enough GPU slots.
3. When choosing your motherboard please check the architecture layout, if the motherboard is going to share the PCI-E lanes between SSD,HDD,CPU & GPU and when all of them are running at capacity you will notice considerable performance issues. Refer Note [1]
4. Regarding GPU please have a look at Note[2], its old but has some good numbers to compare and comment section has more data for comparison.
Hi ShriramIyer16,
If you don't mind me asking, what kind of AI application you do? Reason being the system RAM size, onboard RAM on GPU and even the storage space requirements will depend on that.
1. General idea is to have a slow HDD pushing data to a faster SSD which will in-turn have at least 5% of your data set ready at any given time in System RAM. But unfortunately once the data set grows you will start seeing bottlenecks in HDD <-> SSD <-> RAM so it always better to have more RAM, bigger SSD, So please aim for 64GB System RAM minimum with further expansions.
2. Though you may be aware, I would like to say that having multiple GPU's really helps, if not in parallel processing but in running multiple instances of the same data set with different parameters. This really helps in selecting better methods and running experimentation. So when selecting a motherboard please make sure you have enough GPU slots.
3. When choosing your motherboard please check the architecture layout, if the motherboard is going to share the PCI-E lanes between SSD,HDD,CPU & GPU and when all of them are running at capacity you will notice considerable performance issues. Refer Note [1]
4. Regarding GPU please have a look at Note[2], its old but has some good numbers to compare and comment section has more data for comparison.
To answer your question. I do run LogRhythm it's a SIEM AI Based I do try it frequently. I am not into development of AI but use certain tools frequently. Moreover Kali Linux and Parrot OS are something which I use majorly. brute Force attacks are something which I do most of the times
That's just a bit of what I do..
Hi ShriramIyer16,
I checked the AI engine used by LogRhythm and sorry for saying but the "AI" label is very generous in their instances. None the less the need to have brute force functionality solves some of the issues that I mentioned in my previous post.
1. From what I know(not much though) having the entire dictionary in the GPU ram helps, and some of the dictionaries are almost 10GB in size, Will you consider GTX 1080Ti for its 11GB RAM? (I remember reading an old article on dictionary size and GPU performance based on RAM, but I couldn't find it, If I do I will post)
2. I used this site Note[1] to compute the password length and time to crack but was unable to find a definitive data on GPU performance but from what I know, it might be worth it to go for RTX 2080Ti for its RAM and Speed instead for investing twice.
3. I found a Hashcat benchmark Note[2] but not a complete one though.
Hi ShriramIyer16,
I checked the AI engine used by LogRhythm and sorry for saying but the "AI" label is very generous in their instances. None the less the need to have brute force functionality solves some of the issues that I mentioned in my previous post.
1. From what I know(not much though) having the entire dictionary in the GPU ram helps, and some of the dictionaries are almost 10GB in size, Will you consider GTX 1080Ti for its 11GB RAM? (I remember reading an old article on dictionary size and GPU performance based on RAM, but I couldn't find it, If I do I will post)
2. I used this site Note[1] to compute the password length and time to crack but was unable to find a definitive data on GPU performance but from what I know, it might be worth it to go for RTX 2080Ti for its RAM and Speed instead for investing twice.
3. I found a Hashcat benchmark Note[2] but not a complete one though.
Processor -AMD Ryzen 7
Motherboard -MSi X570-A Pro -13.9k
Graphics Card - 2080 Ti
RAM - 64/128 GB of RAM
CPU Cooler -Deepcool Castle 240RGB V2 -7.8k
SSD -Intel 660P 1TB M.2 SSD -10.5k (onlyssd.com)
Power Supply -Antec EarthWatts Gold Pro-750w -8k
Cabinet -Antec NX800 ARGB e-ATX -7.5k
Monitor -Acer Nitro VG270P 27inch 1MS 144Hz FHD IPS -17.7k (Amazon.in)
KB & Mouse -Cooler Master Masterkeys Lite L Combo -4k (mdcomputers.in)
How much you think will be the cost difference? where else I can cut the cost?
Should I not shell out money on M.2 SSD now instead go with Normal SATA 3.5 Drives upgrade them later?
Processor -AMD Ryzen 7
Motherboard -MSi X570-A Pro -13.9k
Graphics Card - 2080 Ti
RAM - 64/128 GB of RAM
CPU Cooler -Deepcool Castle 240RGB V2 -7.8k
SSD -Intel 660P 1TB M.2 SSD -10.5k (onlyssd.com)
Power Supply -Antec EarthWatts Gold Pro-750w -8k
Cabinet -Antec NX800 ARGB e-ATX -7.5k
Monitor -Acer Nitro VG270P 27inch 1MS 144Hz FHD IPS -17.7k (Amazon.in)
KB & Mouse -Cooler Master Masterkeys Lite L Combo -4k (mdcomputers.in)
How much you think will be the cost difference? where else I can cut the cost?
Should I not shell out money on M.2 SSD now instead go with Normal SATA 3.5 Drives upgrade them later?
@bssunilreddy : Can you please suggest a good config in the range of 2 Lakhs with Rzyen 9 and RTX 2080Ti with 250GB of a Nvme drive and 4TB SATA drive for other storage.
@bssunilreddy : Can you please suggest a good config in the range of 2 Lakhs with Rzyen 9 and RTX 2080Ti with 250GB of a Nvme drive and 4TB SATA drive for other storage.
I would suggest you going for a an enterprise setup with epyc and a motherboard with bmc for remote management if you want to run virtualized systems. enterprise servers are designed to run 24/7 workloads and usually have enough banks for ram. they will be loud as **** so its better to place them in the garage and connect using your laptop etc...
as an example: https://www.asus.com/Commercial-Servers-Workstations/RS500A-E10-RS12U/
things to look for: pcie-gen4, hot swappable nvme drives, bmc etc...
I would suggest you going for a an enterprise setup with epyc and a motherboard with bmc for remote management if you want to run virtualized systems. enterprise servers are designed to run 24/7 workloads and usually have enough banks for ram. they will be loud as **** so its better to place them in the garage and connect using your laptop etc...
as an example: https://www.asus.com/Commercial-Servers-Workstations/RS500A-E10-RS12U/
things to look for: pcie-gen4, hot swappable nvme drives, bmc etc...
PS: I run a lower power ESXI server for my lab use for Pentest/Malware Analysis and for other Research. I dont have any use case for AI so I am not sure on specs needed for it. But if Virtualiztion is needed, here are some pointers:
If your workflow is something like build, test, deploy, destroy, then going a hypervisor route (Proxmox/ESXI/KVM) is the best method and would give you more bang for the buck.
Amd Ryzen is not built with virtualization in mind and is not generally recommended. EPYC is built with Virtualization in mind but its expensive. Though, Level1Linux has a few videos where he has verified that Virtualization works and some of the issues with IOMMU grouping. It may or may not be relevant to you as it vastly depends on your use case.
If I were you, I would stick with an Intel CPU instead of Ryzen but thats personal choice. Few following instruction bits thats needed for Virtualization: Vt-x, VT-d(for passthrough), NX bit, EPT, AVS/SSE.x. (If you are going Ryzen, then its good to verify you have these). (Check about NIC Partitioning using SR-IOV, if you intend to simulate high network traffic in your lab.)
Option: Virtualized Build (Type 1 Hypervisor)
You could get an Intel box (used hardware) and have a compatible GPU installed on it. Then, install the hypervisor (ESXI 6.7u3/7 or KVM) on your it and create different VMs for different use cases - AI, Pentesting, Work, personal etc. For example: Create 1 VM and have the GPU passthrough to the VM and use it for doing your AI research/testing. For pentesting, create another isolated network (vSwitch) in your Hypervisor, have Kali VM and your tools in that VM have the DUT/victim PC as another VM on the same isolated network and do your testing there. This is more robust testing for long term.
Notes on getting a new GPU:
With reference to GPU, its really a bad time to buy a high end GPU. If the purchase is not immediate and could hold the GPU purchase till say September, then Nvidia is releasing the new GTX GPU with Tensor cores. Would love to see what AMD has to offer at that time.
Notes on getting a new HDD:
-You could either have the storage locally on the same box or attach it via iSCSI.
-If you can re-use your old PC as a freenas server with ZFS, it would be perfect. You could have enterprise grade gear. Watch out for "SMR" drives. WD, Toshiba, Seagate are getting sued for this. Ref:
We tested WD Red SMR v CMR drives to see if there was indeed a significant impact with the change. We found SMR can put data at risk 13-16x longer than CMR
www.servethehome.com
-You could get the older 660p 1TB SSD if you have speed in mind. Its the best bang for the buck. The 1TB used to sell for 6.5-7k few months back. If going for regular HDDs, please buy hard drives that does not use SMR. You will thank me later.
Like @booo mentioned pcie-gen4 and nvme are things worth looking out for.
Other considerations overlooked:
- If not considered already, you definitely don't want your server to run into hardware issues due to power outages. Calculate TDPs of all your comments before purchase. You would need a good UPS connected to this rig. If you live in India and have a whole house inverter with good batteries, that would work as well.
- Install Surge protectors/voltage regulators where applicable.
Bottomline:
If you have a budget for 2lacks. I am not sure if its possible, try avoiding consumer gear at all costs. Mediocre Enterprise gear with room for future expansion would be a smart move.
I can also help you with doubts. I have been working with technologies from VMware/Citrix for almost 8 to 9 years now so can share my personal experiences with you. For ESXi, you dont have have to install it in the internal SSD/HDD, you can install it on a USB/SD card too. You have share the specifics as to what you are looking for. Will it be also related to VSAN/NSX or just testing virtual machines. What i meant was, will you be doing network related analysis and that stuff, a simpler solution would also be VMware Workstation Pro but then it does not Support PCI Passthrough or GPU passthrough.
I can also help you with doubts. I have been working with technologies from VMware/Citrix for almost 8 to 9 years now so can share my personal experiences with you. For ESXi, you dont have have to install it in the internal SSD/HDD, you can install it on a USB/SD card too. You have share the specifics as to what you are looking for. Will it be also related to VSAN/NSX or just testing virtual machines. What i meant was, will you be doing network related analysis and that stuff, a simpler solution would also be VMware Workstation Pro but then it does not Support PCI Passthrough or GPU passthrough.
So you mean to say if i use kali over vm i cant use GPU to Brute Force?
Also what if I dont install VMware Instead if I go for VirtualBox will it still wont allow GPU passthrough.
do you recommend going with Ubuntu KVM solution or in worst case with Hyper-V based solution (through Win10)
I have E5 1620 with DX79TO and 32 GB DDR3 RAM with 2x1060 6GB ones - My workload (NLP) requires fewer resources than CV - if I do this, I will get an option to run the system into separate containers/docker and also ensure deployable solution tested before putting it into source control
