The janky white fan in the second half of the video is a friendly reminder to all DIY workstation users that you absolutely need to actively cool DDR5 RDIMMs. Think of it as a return of the infernal DDR2-FB-DIMMs.
You could tell her the factual statement that computer silicon needs less power for the same work if it runs cooler, a few % for every 10 K lower temperatures :-) So by leaving the AC on you actually SAVE power*! (*Just don‘t mention the absolute increase of power consumption of the entire house)
I put my server rack in a corner, sectioned it off with a type of paneling, and added a duct with an inline exhaust fan taking all of the heat directly outside with one of those window adapters for portable ACs. Added an opening at the bottom of the enclosure on the inside for cool air. Temps in that room are now the same in the rest of the house. I could also add a small window AC where the servers are and get the server ambient temps down to 60 real fast. Without it my ambient in the area is 78, and 72 outside. Energy costs went down a little too.
Definitely have to agree on the price. I'm down for 16 cores unlocked around $750 (yes, I know that's very close to the consumer socket 16 core price, but the board is more and most importantly, the memory is outrageous). I also think that a refresh a year and a half after the initial launch means 10980XE levels of discounts, not here's two more cores for a few hundred bucks off.
Workstation powerconsumption is a big thing! Every watt put into the system translates 1:1 into heat! Depending on your (home)office-space, this can make a great impact. I remember then i switched from a 1080 to a 2080ti. The higher power consumption translated into 2-3 degrees more room temperature under load. Sure, no problem in winter or with aircon, but in regular rooms in summer, it can be awful.
What I really want is to have two 8 core rings with 40 to 48 PCIE lanes (capable of 10 to 14 four lane devices) and quad channel memory. That would be a glorious HEDT setup.
I dropped sTRx40 3960X for AM5 7900X after being left behind by AMD and being without any availability of appealing equivalents by Intel during the first Sapphire Rapids launch (there was basically 0 consumer availability in Australia). Now that I have my system I have no appetite to upgrade in the near future. I just wish these companies wouldn't fuck up their product lines and launches like they have been. It really sucks. I would've gone sTRx50, but it launched WAAAY too late for me.
@@cracklingice I really don't see Intel making a 2xxx socket for HEDT, maybe with some packaging magic we will have ring bus with HBM or eDram or both (so that we can have GREAT gaming performance with VM
Thanks Wendell. The MKL vs AOCL optimisation stuff is a kicker - packages such as Matlab, Ansys, etc still seem more optimised for intel/MKL. A while back after plenty of tyre-kicking I went with modest W7-3465X Xeon over similarly priced TR7k for just this reason. It runs a lot of benchmarks slower but my code/models faster. I'd be interested to see results for the 34xx vs 35xx when you get your hands on one.
Hi there, Jon Patton from Intel. For Xeon W, we didn't use Emerald Rapids because at a platform level, we would loose some PCIe connectivity. We didn't want someone to buy an Emerald Rapids-based Xeon W CPU, and loose PCIe slots on a W790 motherboard. Hope this makes sense.
@@jonathanpatton3411 thank you very much for the explanation. Although, I would've rather had a more performant Emerald Rapids processor and less PCI express lanes. Honestly, I really want the workstation variants for Granite Rapids to come out… But I don't think this would happen until around this time next year correct?
I really appreciate all these software optimisations like Python numpy. However, I think in reality Python devs will just pip install stuff without trying to put all these Intel libraries together. I see a lot of that in ML space. If Intel works closely with packaging teams then there is a chance that you get out of the box optimized libraries cause lots of Python devs would not know or care and system devs who use C and similar would know what to do anyway. I guess what I’m trying to say that there is a lot of performance left on the table and the whole community needs to work together to improve that. I love lots of cores and powerful machines but software optimisations are just as important. (Yes I’m looking at you pre/post processing steps written in Python taking comparable time to actual GPU processing).
This reminds me on why Clear Linux exists and what it shows to other distros what to expect with a little more optimization effort on their part. I also agree that more effort should be put into getting optimized software down to the users of these systems as easy and seamless as it can possibly be.
Sometimes I wonder why I watch these types of videos. I will never have a system like this or need a system like this? Nonetheless, I enjoy listening to Wendell talk about things he's passionate about!
If Intel manages to keep cost down they'll have a pretty good competitor to threadripper…this could be a great enthusiast platform, it's got the potential.
@@abavariannormiepleb9470 good question,that’s for our in house genius to let us know, I would assume that it would definitely improve performance of VSTi‘s, and probably normal VST‘s too, only having one AMD computer I’m not in a position too try this out, this is why I’d love Wendell to take this up as a project to see which parameters are the best for DAW‘s and Plug-in host software, I would imagine latency and consistent performance over all cores would be very important because the pc would only be as fast as the slowest core since all audio tracks must run consistently in time with each other, the software usually compensating for latency differences to keep all tracks in time
Would you be able to disable all but 8 cores, while keeping the L3 cache intact, to test what benefits in gaming performance (if any) Intel could get if they ever went the 3D-Vcache route, or is there some reason why this test wouldn't really work?
Excellent review, thank you, we are waiting for tests with the configuration of the updated platform, memory, and possibly overclocking. By the way, was the problem with 1T memory timing in the refresh solved?
12:12 I had different experience from what you mentioned for intel library(mkl). on the Ryzen 7950x3d changing the blas setting from intel mkl to the amd AOCL made something from 20 to 25% improvement in my run time on the same simulation! (I highly recommend to use 6.2 version of comsol for these kin of tests because of their lastest improvement the made for these kind of stuff)
I am really looking forward to the next video about your overclocking experiences. I hope you've reviewed the W790 Memory Overclocking thread on the L1T forums for some insights into the platform with the current 2400/3400 series cpus!
These could be amazing video editing platforms if Intel had included Quick Sync. Taking a look at Intel Ark, they haven't released a Xeon with on board graphics and QuickSync since 2021. AMD has nothing to compare with Intel's excellent codec acceleration. Even the cheapest core parts have it, yet none of the recent Xeon's do... more own goals by Intel.
It would certainly be a nice-to-have, but in a pinch a little single slot Arc A310 can serve that same QuickSync capability as well. I suspect most of these workstations will have paired GPUs that are tailored to their use case, whether that be CUDA acceleration on Nvidia, software like Siemens NX on AMD or Intel's own Arc or GPU Flex solutions for multi-stream transcoding or VDI.
@@AK-Brian Can an Arc GPU be used simultaneously with an Nvidia GPU? With a QuickSync CPU, software like DaVinci Resolve can use the CPU's on-die GPU exclusively for Quick Sync codec acceleration. While GPU intensive effects, denoising, and the rest are accelerated by a beefier Nvidia GPU.
@@TheVault1999 For straight editing, yes. But effects, de-noising, and other GPU intensive tasks work best with a fast Nvidia card. Last I checked, there was no way to use an Nvidia card and an Arc at the same time, even in systems like Threadripper that have the PCIE lanes for both. The advantage of QuickSync is that both can be used at the same time, and software like DaVinci Resolve can use the CPU's on-die GPU exclusively for codec acceleration, with all other tasks handed off the far faster Nvidia GPU.
This might just scratch my itch for more PCIE lanes on my desktop without breaking the bank, I'll have to keep an eye out for pricing for this platform when I get around to rebuilding my workstation.
QSV does work on the Arc and GPU Flex series GPUs. They're great (hardware accelerated) options for working with native 10-bit HEVC 4:2:2 video content, especially.
I have the ASRock and the Asus Ace w790 boards. The ASRock is a great board for a desktop/workstation with normal NVMe drives. The Asus board I find better for optane raid setup since you have multiple NVMe pci-e adapter cards and cram every last piece with optane in vroc. I've tried nearly all the water blocks for the 4677 socket except the alphacool block. The Iceman cooler by far works the best and using liquid metal. I can do an all core 5ghz on my 2495x with an Intel Arc a380 Air cooled and a 7900xtx (water cooled. Ram at 6600mhz 24hr karhu stable. I use a single Seasonic 1600w platinum PSU currently and I have hit 1400w watts on this setup at the wall.
@@paulblair898 Iceman cooler 4677 uses G1/4. Unless they make a laydown version that I am unaware of where the inlet and outlet is on the side of the block.
Definitely hot to have a workstation platform instead of high-end consumer under the desk, and the need for more pcie lanes is real. However, as long as even the 2595 has problems in taming a super cheap (in comparision) ryzen 9950X in more "creator" focused workloads like davinci resolve, lightroom etc. i do not see myself shelling out this kind of money. which is actually a shame.
@@Level1Techs They really did mess up the naming with this board. Actually - it started with Intel and W790 being both quad and octal channel when there's room for W795 or something to designate the difference. Or X790 and W790.
On the SR-3 with a locked CPU it won't run XMP or above the rated speed of the CPU but you can adjust timings still to improve performance so I imagine it is similar with the new platform but you never know I guess.
Could we get some comparisons with 14900k? I do a LOT of workstation work, 3d modeling, rendering, lots photo, video, AI, but... I also game and want to use it as a desktop day to day. my poor 10940x is getting a bit sad but I don't want to deal with P and E and yuck BUT.. them P cores on consumer are so fast :(
Such a cool platform. I’m gonna buy one for my gaming rig lol throw it on the 1/2HP water chiller and have a blast with unlocked Xeon 2565X 🎉 so cool man! I love this.
Wasn't some of the lack of AMD software unification addressed just a day or two ago? I believe it's called UDNA and designed to solve their issues with software forking and reliance on open source to solve their quality issues for them.
Just switch to Linux. I rode a 4th gen i5 on Ubuntu till only a few months ago. Don’t even use WSL, just skip and go to raw Linux. Trust me, the developer experience is 10x better.
@@macicoinc9363 I have been using Linux on this system but I regularly saturate the ram and CPU. I moved away from 4th gen i7 1 year ago. Both that system and this one are amazingly well built and stable. I still use the 4th gen as a proxmox server! Maybe I should just upgrade to a 9th gen i7 or i9 and max out the ram. I just love seeing new and crazy hardware!
18:46 Multi-Core Score is not so impressive in comparison with Intel i9-14900K or new Arrow Lake preliminary scores. The difference is only in the system reliability with 100% 24/365 work loads.
An interesting platform and definitely one to look at when I replace my TR3960X system. Getting burned by AMD on the TRX40 platform left a very bitter taste and MKL is very effective on my heavier compute jobs.
It seems to me that the 7970X destroys the 2595X on generic benchmarks, for about the same price. It would have been nice to see overclocked results, since you advertised it as a feature. Price on Newegg: 2595X: $2319.99 7970X: $2302.11 CPU-Z: 2595X: 20880 7970X: 29124 My overclocked 7970X: 32300 Geekbench 6.3: 2595X: 16109 7970X: 24378 My overclocked 7970X: 28000
But your benchmarks show 7970 consistently better than this Intel, excluding accelerators. TR matches this much with 24 cores 7960X and for this price offers more bandwidth on 7965WX and 8ch. I'm confused with benchmarks shown and conclusion in the video. Yes, you can overclock Intel, but you can overclock TR too...
if workstation class power consumption can be ignored on newest chip, i might as well just buy an older system. Why should i buy a new intel chip. honestly not impressed by this
The janky white fan in the second half of the video is a friendly reminder to all DIY workstation users that you absolutely need to actively cool DDR5 RDIMMs. Think of it as a return of the infernal DDR2-FB-DIMMs.
Wonder if your comment could help me sway my wife into leaving the AC on in the side of the house where my servers live.
You could tell her the factual statement that computer silicon needs less power for the same work if it runs cooler, a few % for every 10 K lower temperatures :-)
So by leaving the AC on you actually SAVE power*!
(*Just don‘t mention the absolute increase of power consumption of the entire house)
@@Darric1Actual Or you could spend $60K on a solar and battery system for the servers and AC and not have to answer to the wife.
More air is always a good idea!
I put my server rack in a corner, sectioned it off with a type of paneling, and added a duct with an inline exhaust fan taking all of the heat directly outside with one of those window adapters for portable ACs. Added an opening at the bottom of the enclosure on the inside for cool air. Temps in that room are now the same in the rest of the house. I could also add a small window AC where the servers are and get the server ambient temps down to 60 real fast. Without it my ambient in the area is 78, and 72 outside. Energy costs went down a little too.
Definitely have to agree on the price. I'm down for 16 cores unlocked around $750 (yes, I know that's very close to the consumer socket 16 core price, but the board is more and most importantly, the memory is outrageous).
I also think that a refresh a year and a half after the initial launch means 10980XE levels of discounts, not here's two more cores for a few hundred bucks off.
Workstation powerconsumption is a big thing! Every watt put into the system translates 1:1 into heat!
Depending on your (home)office-space, this can make a great impact.
I remember then i switched from a 1080 to a 2080ti. The higher power consumption translated into 2-3 degrees more room temperature under load.
Sure, no problem in winter or with aircon, but in regular rooms in summer, it can be awful.
to be absolutely pedantic, kinetic and acoustic energy come out of the input energy. also phase change. but sure, 10:9 or 100:96
damn I really want a 4 channel HEDT
or keep the socket but use ring bus up the frequency + massive eDRAM on that huge socket
What I really want is to have two 8 core rings with 40 to 48 PCIE lanes (capable of 10 to 14 four lane devices) and quad channel memory. That would be a glorious HEDT setup.
Me too! Alas, a good desktop is more than enough for what I do, realistically. I’ll have to continue living vicariously through Wendell.
I dropped sTRx40 3960X for AM5 7900X after being left behind by AMD and being without any availability of appealing equivalents by Intel during the first Sapphire Rapids launch (there was basically 0 consumer availability in Australia). Now that I have my system I have no appetite to upgrade in the near future. I just wish these companies wouldn't fuck up their product lines and launches like they have been. It really sucks.
I would've gone sTRx50, but it launched WAAAY too late for me.
@@cracklingice I really don't see Intel making a 2xxx socket for HEDT, maybe with some packaging magic we will have ring bus with HBM or eDram or both (so that we can have GREAT gaming performance with VM
Wendell really knows his shit, I trust his info completely. Curious what the difference is between air and liquid cooling.
Don’t ever trust 1 person 100%. Everyone gets something’s wrong
All hail overlord Wendell!!
@@TheaKaSaToRi Wendell is a gangsta tho
Viscosity 😁
@@justrl3klol facts!
Thanks Wendell. The MKL vs AOCL optimisation stuff is a kicker - packages such as Matlab, Ansys, etc still seem more optimised for intel/MKL. A while back after plenty of tyre-kicking I went with modest W7-3465X Xeon over similarly priced TR7k for just this reason. It runs a lot of benchmarks slower but my code/models faster. I'd be interested to see results for the 34xx vs 35xx when you get your hands on one.
Any indication why Emerald Rapids didn't make the move to workstation? Is Sapphire Rapids refresh being used to clear out existing inventory?
My guess is they ned all the emerald rapids they have (I mean can produce) for servers market.
Hi there, Jon Patton from Intel. For Xeon W, we didn't use Emerald Rapids because at a platform level, we would loose some PCIe connectivity. We didn't want someone to buy an Emerald Rapids-based Xeon W CPU, and loose PCIe slots on a W790 motherboard. Hope this makes sense.
@@jonathanpatton3411 thank you very much for the explanation. Although, I would've rather had a more performant Emerald Rapids processor and less PCI express lanes. Honestly, I really want the workstation variants for Granite Rapids to come out… But I don't think this would happen until around this time next year correct?
@danielhofmann7294 No comment on future product plans for workstation processors, but you can expect Granite Rapids on the server side coming soon.
I really appreciate all these software optimisations like Python numpy. However, I think in reality Python devs will just pip install stuff without trying to put all these Intel libraries together. I see a lot of that in ML space. If Intel works closely with packaging teams then there is a chance that you get out of the box optimized libraries cause lots of Python devs would not know or care and system devs who use C and similar would know what to do anyway. I guess what I’m trying to say that there is a lot of performance left on the table and the whole community needs to work together to improve that. I love lots of cores and powerful machines but software optimisations are just as important. (Yes I’m looking at you pre/post processing steps written in Python taking comparable time to actual GPU processing).
This reminds me on why Clear Linux exists and what it shows to other distros what to expect with a little more optimization effort on their part. I also agree that more effort should be put into getting optimized software down to the users of these systems as easy and seamless as it can possibly be.
@@seylaw that’s interesting, I’ll check Clear Linux.
@@AI-xi4jk Clear Linux is a bit hard to use as a desktop OS as it lacks a bit in ease of use. For desktop users, I'd recommend CachyOS instead.
Sometimes I wonder why I watch these types of videos. I will never have a system like this or need a system like this? Nonetheless, I enjoy listening to Wendell talk about things he's passionate about!
If Intel manages to keep cost down they'll have a pretty good competitor to threadripper…this could be a great enthusiast platform, it's got the potential.
That would make a nice audio server as plugin host, that would be a different project for given the special audio requirements
Do DAWs profit more from the higher absolute performance of individual cores or stuff like 3D-V-Cache?
@@abavariannormiepleb9470 good question,that’s for our in house genius to let us know, I would assume that it would definitely improve performance of VSTi‘s, and probably normal VST‘s too, only having one AMD computer I’m not in a position too try this out, this is why I’d love Wendell to take this up as a project to see which parameters are the best for DAW‘s and Plug-in host software, I would imagine latency and consistent performance over all cores would be very important because the pc would only be as fast as the slowest core since all audio tracks must run consistently in time with each other, the software usually compensating for latency differences to keep all tracks in time
@@abavariannormiepleb9470 From what I've seen memory bandwidth and/or more cache.
Would you be able to disable all but 8 cores, while keeping the L3 cache intact, to test what benefits in gaming performance (if any) Intel could get if they ever went the 3D-Vcache route, or is there some reason why this test wouldn't really work?
Excellent review, thank you, we are waiting for tests with the configuration of the updated platform, memory, and possibly overclocking. By the way, was the problem with 1T memory timing in the refresh solved?
12:12 I had different experience from what you mentioned for intel library(mkl). on the Ryzen 7950x3d changing the blas setting from intel mkl to the amd AOCL made something from 20 to 25% improvement in my run time on the same simulation! (I highly recommend to use 6.2 version of comsol for these kin of tests because of their lastest improvement the made for these kind of stuff)
ASWock Extweme W790 WS UwU Edition
Oh, thanks for the tip on the power states! I need to try that.
Very interesting. Thanks for the info.
whats that (printer?) at 1:44
I've been liking Sapphire Rapids so far, seems very stable on Linux and even older kernels, just hope the refresh is as good
I am really looking forward to the next video about your overclocking experiences. I hope you've reviewed the W790 Memory Overclocking thread on the L1T forums for some insights into the platform with the current 2400/3400 series cpus!
If I were allowed to study any of this or work with it professionally, this ad would be nice
i LOVE that L1T Pc Case
These could be amazing video editing platforms if Intel had included Quick Sync. Taking a look at Intel Ark, they haven't released a Xeon with on board graphics and QuickSync since 2021.
AMD has nothing to compare with Intel's excellent codec acceleration. Even the cheapest core parts have it, yet none of the recent Xeon's do... more own goals by Intel.
It would certainly be a nice-to-have, but in a pinch a little single slot Arc A310 can serve that same QuickSync capability as well. I suspect most of these workstations will have paired GPUs that are tailored to their use case, whether that be CUDA acceleration on Nvidia, software like Siemens NX on AMD or Intel's own Arc or GPU Flex solutions for multi-stream transcoding or VDI.
Go amd threadripper and intel a770..done
@@AK-Brian Can an Arc GPU be used simultaneously with an Nvidia GPU? With a QuickSync CPU, software like DaVinci Resolve can use the CPU's on-die GPU exclusively for Quick Sync codec acceleration. While GPU intensive effects, denoising, and the rest are accelerated by a beefier Nvidia GPU.
@@TheVault1999 For straight editing, yes. But effects, de-noising, and other GPU intensive tasks work best with a fast Nvidia card.
Last I checked, there was no way to use an Nvidia card and an Arc at the same time, even in systems like Threadripper that have the PCIE lanes for both.
The advantage of QuickSync is that both can be used at the same time, and software like DaVinci Resolve can use the CPU's on-die GPU exclusively for codec acceleration, with all other tasks handed off the far faster Nvidia GPU.
I'm one of the loudest and screechiest TH-cam commenters, I demand efficiency
This might just scratch my itch for more PCIE lanes on my desktop without breaking the bank, I'll have to keep an eye out for pricing for this platform when I get around to rebuilding my workstation.
Is it safe to buy i5 14600KF at the moment?
Can we get quicksync by using Intel Gpu's. ? I bet this would be a really nice video editing rig . !
QSV does work on the Arc and GPU Flex series GPUs. They're great (hardware accelerated) options for working with native 10-bit HEVC 4:2:2 video content, especially.
Intel needs to get you a 3000 series Sapphire Rapids system for testing, you're totally selling these things 👍
I have the ASRock and the Asus Ace w790 boards.
The ASRock is a great board for a desktop/workstation with normal NVMe drives.
The Asus board I find better for optane raid setup since you have multiple NVMe pci-e adapter cards and cram every last piece with optane in vroc.
I've tried nearly all the water blocks for the 4677 socket except the alphacool block. The Iceman cooler by far works the best and using liquid metal.
I can do an all core 5ghz on my 2495x with an Intel Arc a380 Air cooled and a 7900xtx (water cooled. Ram at 6600mhz 24hr karhu stable.
I use a single Seasonic 1600w platinum PSU currently and I have hit 1400w watts on this setup at the wall.
I just got the low profile LGA4677 Iceman cooler block and am having a hard time finding G1/8 fittings for it, but other than that it seems very nice.
@@paulblair898 Iceman cooler 4677 uses G1/4. Unless they make a laydown version that I am unaware of where the inlet and outlet is on the side of the block.
@@mnkydeth8189 Yeah they make a second version of the LGA4677 block with the G1/8 ports, I'll tag you when I post the build on the L1 forums.
you always show the shiny stuff, i dream of building on...
Definitely hot to have a workstation platform instead of high-end consumer under the desk, and the need for more pcie lanes is real. However, as long as even the 2595 has problems in taming a super cheap (in comparision) ryzen 9950X in more "creator" focused workloads like davinci resolve, lightroom etc. i do not see myself shelling out this kind of money. which is actually a shame.
ASRock W790 WS page shows 2 x 10 Gigabit LAN ports?
2.0 ...
@@Level1Techs They really did mess up the naming with this board. Actually - it started with Intel and W790 being both quad and octal channel when there's room for W795 or something to designate the difference. Or X790 and W790.
3:11... I'm sorry that's Cascade Lake
Referring to math library performance as python performance seems a bit too simplified.
Nice video.
On the SR-3 with a locked CPU it won't run XMP or above the rated speed of the CPU but you can adjust timings still to improve performance so I imagine it is similar with the new platform but you never know I guess.
Good Job 👍
Day this video dropped the price of that motherboard jumped to almost $900!
Could we get some comparisons with 14900k? I do a LOT of workstation work, 3d modeling, rendering, lots photo, video, AI, but... I also game and want to use it as a desktop day to day. my poor 10940x is getting a bit sad but I don't want to deal with P and E and yuck BUT.. them P cores on consumer are so fast :(
The 1/10 handling of the 13th/14th gen issues has me disregarding Intel completely for at least 10 years.
If this is anything as nice as Sapphire Rapids then I expect a good product.
From experience the server and Desktop CPUs are in different classes
Cool stuff.
Safe to run these at such power levels?
Such a cool platform. I’m gonna buy one for my gaming rig lol throw it on the 1/2HP water chiller and have a blast with unlocked Xeon 2565X 🎉 so cool man! I love this.
I want those PCIe lanes.
Wasn't Emerald Rapids already launched? Why refresh Sapphire Rapids then? I don't get the reasoning here.
Wasn't some of the lack of AMD software unification addressed just a day or two ago? I believe it's called UDNA and designed to solve their issues with software forking and reliance on open source to solve their quality issues for them.
1:50 Did I hear that right? RTX 5880 GPUs?
That's just some wierd Ada Lovelace workstation card, not Blackwell.
is your home insurance up to speed?cause testing intel nowdays....
I'm just CS student and for somthing like this but I'd kill for even half the capacity of the systems that L1 covers! My 9th Gen i5 is getting tired.
Just switch to Linux. I rode a 4th gen i5 on Ubuntu till only a few months ago. Don’t even use WSL, just skip and go to raw Linux. Trust me, the developer experience is 10x better.
@@macicoinc9363 I have been using Linux on this system but I regularly saturate the ram and CPU. I moved away from 4th gen i7 1 year ago. Both that system and this one are amazingly well built and stable. I still use the 4th gen as a proxmox server! Maybe I should just upgrade to a 9th gen i7 or i9 and max out the ram. I just love seeing new and crazy hardware!
18:46 Multi-Core Score is not so impressive in comparison with Intel i9-14900K or new Arrow Lake preliminary scores. The difference is only in the system reliability with 100% 24/365 work loads.
Re: Intel CPUs C6 states - is anyone else experiencing a weird "floaty" and inaccurate mouse-pointer in Windows 10 or Windows 11?
Thank you, Wendellman! 🙏🏼
8 channels/dimms, 16 cores, $1000 please!
This bandwidth goes up to 11
🔥
Gets me why people are so price perW
workstations are meant to get things done, not go slow and save power
An interesting platform and definitely one to look at when I replace my TR3960X system. Getting burned by AMD on the TRX40 platform left a very bitter taste and MKL is very effective on my heavier compute jobs.
i wish i could afford to play with those, but i cant even afford a whole system at that price.
RTX 5880 GPUs?
That one's going to catch people out, for sure. It's a cut down RTX 6000 Ada workstation card that was quietly launched to fill out the stack.
@@AK-Brian I still for the life of me cannot understand the retiring of the Quadro branding.
"...a whole nother level of CPU performance!"
...so, Level2Techs?
simply i want a motherboard with 7x pcie x16 slots, i don't care about how that is done as logn as i have that at consumer prices!
maybe my upgrade path from X299 with I9 7960X.
Thanks for the great content, as always, Wendell!
It seems to me that the 7970X destroys the 2595X on generic benchmarks, for about the same price.
It would have been nice to see overclocked results, since you advertised it as a feature.
Price on Newegg:
2595X: $2319.99
7970X: $2302.11
CPU-Z:
2595X: 20880
7970X: 29124
My overclocked 7970X: 32300
Geekbench 6.3:
2595X: 16109
7970X: 24378
My overclocked 7970X: 28000
"about the same relative uplift as from 13th to 14th gen" - aka piss all
$499?
Workstation class
Okay, make 4 DIMM $400 mATX board and make 18-core cost $1000-1200 and i am in for CAD workstation. Otherwise, 9950X(3D) it is.
got my 3435x to 5.2 drawing like 600w, number 1 on the 3dmark boards for that chip, but i know i can push it more im just not that skilled.
So why should anyone trust that Intel CPU won't burn out within a year?
Say what
Somebody overclock a non-X xeon please!!!
But your benchmarks show 7970 consistently better than this Intel, excluding accelerators. TR matches this much with 24 cores 7960X and for this price offers more bandwidth on 7965WX and 8ch. I'm confused with benchmarks shown and conclusion in the video. Yes, you can overclock Intel, but you can overclock TR too...
This is very cool, but why would you waste a kick-@r$e, BMF serious workstation CPU on an _interpreted_ , beginner's language like python???
Mega fail as usual
if workstation class power consumption can be ignored on newest chip, i might as well just buy an older system. Why should i buy a new intel chip. honestly not impressed by this
because your workflow uses any of the intel accelerated tasks ????
It's the cheapest current-gen workstation/server you can buy right now. PCIe lanes and memory bandwidth/capacity are important.
It's the cheapest current-gen workstation/server you can buy right now. PCIe lanes and memory bandwidth/capacity are important.
nana wants u to buy the dip .. 60 blessings
This is so new it's not on ark.
Finally! A motherboard / CPU series combo that I want. The low end of this is what the high-end zen5 should have been.
Python is literally the slowest interpreter language. Taking it and making it run 2x faster “with Intel optimizations” isn’t saying much.
If you want your code to go fast, you aren’t going to be using Python.
A lot of ML stuff uses Python for some reason
@@kendokaaa Because python is gluing a bunch of C++/CUDA together, the python overhead doesn't introduce that much extra runtime.
@@kendokaaa python is the glue code to call into the fast stuff for ML. A coordinator.
@@bryce.ferenczi I see
No one will buy lol))) Server infra will go all AMD because it's just cheaper
Asrock learned Gigabyte Rev
TB4_10G_Wifi Slot_I/O panel
...Can it run Privateer?
That is the one true test of a workstation.
Threadripper sounds cooler though so....🤷