AMD Ryzen 9 4950X "Vermeer" Tested, the Sample Boosts to 4.8 GHz

erek

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,889
"AMD is preparing to launch its next-generation Ryzen 4000 series of desktop processors based on Zen 3 architecture, codenamed Vermeer. Thanks to the sources over at Igor's Lab, we have some new information about the clock speeds of a rumored Ryzen 9 4950X Vermeer model. Featuring 16 cores and 32 threads, the Ryzen 9 4950X is reportedly going to feature boost frequency of at least 4.8 GHz. Given that this is only an engineering sample, the final frequencies could be higher. In the report, the base frequency of the CPU is said to be 3.5 GHz. This is a very good frequency for a CPU that has that many cores. All of this information is coming from decoding the OPN code which states "100-000000059-52_ 48/35 _ Y". The 48 number indicates the boost, and 35 the base frequency. In the previous reports, we got OPN codes "100-000000059-14_46/37_Y" and "100-000000059-15_46/37_N" which suggested 4.6 GHz boost and base of 3.5 GHz, indicating that this is a new stepping."

https://www.techpowerup.com/270700/amd-ryzen-9-4950x-vermeer-tested-the-sample-boosts-to-4-8-ghz
 
I guess it really depends on how high the all / multiple cores can boost.

Most games use quite a few cores nowadays and if gaming frequency is a tad bit lower, the uplift wont be as good as it could (obviously heh)
 
I mean... my 3950 nopes right out the building the second it sees anything close to 4.5 all-core.

If these 4xxx chips are ES all-core'ing out of the box at the boost speed of the 3xxx gen chips, then rip Intel. Get those exploits dusted off or something.
 
erek said:
"AMD is preparing to launch its next-generation Ryzen 4000 series of desktop processors based on Zen 3 architecture, codenamed Vermeer. Thanks to the sources over at Igor's Lab, we have some new information about the clock speeds of a rumored Ryzen 9 4950X Vermeer model. Featuring 16 cores and 32 threads, the Ryzen 9 4950X is reportedly going to feature boost frequency of at least 4.8 GHz. Given that this is only an engineering sample, the final frequencies could be higher. In the report, the base frequency of the CPU is said to be 3.5 GHz. This is a very good frequency for a CPU that has that many cores. All of this information is coming from decoding the OPN code which states "100-000000059-52_ 48/35 _ Y". The 48 number indicates the boost, and 35 the base frequency. In the previous reports, we got OPN codes "100-000000059-14_46/37_Y" and "100-000000059-15_46/37_N" which suggested 4.6 GHz boost and base of 3.5 GHz, indicating that this is a new stepping."

https://www.techpowerup.com/270700/amd-ryzen-9-4950x-vermeer-tested-the-sample-boosts-to-4-8-ghz
Click to expand...

The Y subfix indicates this is not an engineering sample. Moreover OPN means "Orderable Part Number" and engineering samples do not have OPN. This is the final chip or something close to it (QS or a preproduction sample).
 
I was hoping for base clocks in the 4.0 GHz range...3.7 GHz is what we get now with Zen 2
 
This is certainly looking good - I wonder if there is a more performant IPC boost as well? Its especially a good sign considering that maybe AMD will launch with the high end chips available concurrently, as opposed to being delayed until some lower models are launched.

Of course we'll have to wait until launch to see for sure, but it would be nice if AMD (and Intel for that matter but especially AMD) started reporting the default turbo/boost max for multiple core setups. Single , dual and quad core, octa, and then the all core boost/turbo for the whole thing, when it comes to a 16 core like this. Given that AMD especially has been pushing core count and software is finally starting to parallelize to some degree, it would be good to be able to check for a variety of use cases considering that certainly not ALL software operates on full 16c/32t, or even half that. Some games and other software still depend on single core performance, or are limited to a lesser assortment, after all. While overclockers can manually push these CPUs with all kinds of tweaks, it would be nice if AMD would design them to (ideally in UEFI/BIOS , but also with software utils - preferably open source ones that are NOT Windows specific) have different turbo frequencies depending on the current core/thread usages

Anyway, while 4.8 is a nice jump up I (assuming single-core turbo? Would be amazing if it was all core under good cooling), I wonder if there is any more headroom to push it farther. Breaking the 5.0ghz barrier (especially with an IPC enhancement too) on their highest end Ryzen would be another feather in their cap as they continue to thrive. Hopefully we'll get more Zen3 Ryzen news soon, but I am also curious if Threadripper will be launched soon after.
 
Not alone in thinking this but I myself want to see an 8/16 CPU that hits 4.8 - 5.0 or more regularly for that sweet spot gaming/productivity. Maybe the 12/24 but even the 3900x I had for a bit was massive overkill for my day to day use.
 
RanceJustice said:
This is certainly looking good - I wonder if there is a more performant IPC boost as well?
Click to expand...
The chips should get 10% IPC boost, rumors claim 15-20%. Part of this comes from going to an 8 core cluster instead of 4, which allows the group of 8 corers to talk to each other much faster without going through L3 cache (they can talk through L2 now instead which is about 2x faster read and copy speeds and 3x faster latency). This will bring a nice bump to the multicore performance and could be the majority of what is the claimed IPC % boost. While IPC is technically instructions per cycle executed, removing a bottleneck to allow more continuous execution of instructions without stalls does bring up that IPC number.
The other main boost is to a supposedly much better floating point section that can handle bigger instructions more concurrently.
 
Last edited:
KazeoHin said:
We are all excited about this new sample... but in truth we have to temper expectations: as we know from history, when the AMD hype is real, the AMD product is not
Click to expand...

only when AMD's hyping it themselves like bulldozer and vega.. they've been pretty quiet with zen as a whole when it comes to overhyping stuff. the communities are doing all the hyping for them, lol.

definitely interested to see what they pull out of their hat with zen 3 since zen 2's going to be a hard one to beat.
 
erek said:
is it?
Click to expand...

It would pretty much have to be unless Zen3 was a step back in terms of IPC. This is highly doubtful given that the clocks are as iterative as we've always seen from all Ryzens up to this point. A Zen3 based CPU is likely faster per clock cycle than Zen2 and therefore, at similar or slightly higher boost clocks, Intel basically doesn't have a chance due to a core count deficit. Even worse, supposedly AMD is going to address the latency issues of Zen2 in the Zen3 architecture which could eliminate or drastically reduce Intel's lead in gaming. It already mitigated this some with Zen2's massive L3 cache, which is something games really seem to like and always have.

Even applications like Adobe Photoshop which feature Intel specific optimizations sometimes perform better on AMD hardware given the sheer core count. Add to that you get more cores for the money and the fact that Ryzen generally has an IPC advantage now, Intel's products are basically getting beaten at nearly every turn. AMD offers more cores at many price points which use less power and generate less heat. On the desktop, the Core i9-10900K doesn't beat AMD at very much and that starts going out the window if you really dial in the 3950X or 3900X with per CCX overclocking, auto-overclocking, or if you can manage to do anything with PBO+AutoOC. For Intel to defeat AMD's 3950X definitively requires the 10980XE, which is essentially vaporware. Even then, it must be overclocked to 4.7GHz or better, which consumes almost 500w of power on its own and requires a decent custom loop to cool. Comparatively, an overclocked 3950X doesn't even crack 400w.

It's worth noting that this gives Intel a 2c/4t advantage, which shows just how far behind Intel is. That's also comparing Intel's best HEDT offering to AMD's best mainstream processor. Going Intel in this example is far more expensive and those results I spoke of cost dearly in the form of power consumption, heat and hardware costs. The motherboards aren't necessarily worse than a high end X570 motherboard but the quad-channel RAM, custom liquid cooling loop and the CPU itself all amount to a costlier build. For what the 10980XE goes for when you can find them (typically, $400+ over MSRP), you might as well step up to a Threadripper 3960X, which utterly decimates anything Intel has on the desktop, including it's HEDT parts.

There are two things Intel has going for it right now. Gaming performance and its platform is a little easier to deal with than AMD's typically are. Drivers tend to be better from Intel and setting up things like a RAID array are usually way easier on the Intel side. Intel's storage support tends to be better regarding everything but bandwidth. Intel supports RAID 5, AMD doesn't. Intel supports a greater range of stripe sizes, AMD only supports two. When reviewing boards, I sometimes have to fight to get the OS installed on AMD based boards, though this can be mitigated by making sure you have the latest build ready to go on your flash drive. If you want to install the OS to a bootable array, you'll be in for a hell of a time on the AMD side at times. In contrast, Intel's are super simple in this regard and either just work, or are easily made to work with a simple driver which you can install during setup.
 
Dan_D said:
There are two things Intel has going for it right now. Gaming performance and its platform is a little easier to deal with than AMD's typically are. Drivers tend to be better from Intel and setting up things like a RAID array are usually way easier on the Intel side. Intel's storage support tends to be better regarding everything but bandwidth. Intel supports RAID 5, AMD doesn't. Intel supports a greater range of stripe sizes, AMD only supports two. When reviewing boards, I sometimes have to fight to get the OS installed on AMD based boards, though this can be mitigated by making sure you have the latest build ready to go on your flash drive. If you want to install the OS to a bootable array, you'll be in for a hell of a time on the AMD side at times. In contrast, Intel's are super simple in this regard and either just work, or are easily made to work with a simple driver which you can install during setup.
Click to expand...

How many people really use the Mobo's RAID option really? I never have, I just use disk management/storage spaces and freenas, and if I had a RAID card...
 
Hopefully some will move from 3950x to 4950x and sell the 3950x to me (at bargain price) for DCing ;)
 
Dan_D said:
It's probably more common than you think.
Click to expand...
Ok there is one instance that I remember back at my old job.
A client had a Dell XPS tower that had a 2TB spinner and a 32GB SSD cache card or something, and it was configured as a "RAID 0" or something like that. I figured that it was setup to use the SSD as a large write buffer and to cache the files needed to boot windows. Felt as a fast as a 10k rpm drive as far as general use. That's the only instance I came across.

Also having the BIOS/UEFI in RAID mode without any kind of RAID level active doesn't count...
 
Dan_D said:
It's probably more common than you think.
Click to expand...

If you've already got the software and hardware in operation, it's also a near zero-cost option to add for the maybe 5-10 percent of the market that uses it; no company in their right mind would blanketly cut off that percent of their business because it's a minor feature just because most people don't understand or want it.

And the percentage of people who *think* they might want to use it at some point also drives sales, so if they don't, that also adds to lost sales.

I would really like it if AMD had better RAID support. Maybe their gains in the server market will carry over to the home user side. Right now a lot of those sales are going to Intel.
 
I am overall happy with my 3900X & 2080TI for play (B450 has some minor issues), running that at 1440p I don't see a need to upgrade either part in the near future. I do need a new Threadripper for work and I am eagerly anticipating ordering that up It's going to be my main workstation and serve as an emergency backup for the EPYC Hyper-V Hosts in the event I suffer a catastrophic failure on those two systems and I need something to pick up the slack.
 
Rockenrooster said:
Ok there is one instance that I remember back at my old job.
A client had a Dell XPS tower that had a 2TB spinner and a 32GB SSD cache card or something, and it was configured as a "RAID 0" or something like that. I figured that it was setup to use the SSD as a large write buffer and to cache the files needed to boot windows. Felt as a fast as a 10k rpm drive as far as general use. That's the only instance I came across.

Also having the BIOS/UEFI in RAID mode without any kind of RAID level active doesn't count...
Click to expand...
I bet there are a number of people out there running it because they feel they needed it, but overall, on consumer systems running any form of SSD the performance increase isn't there and the complexity and DR issues don't make it worth the hassle. I do know a lot of people, myself included, who are using various forms of Windows Storage Space, as either a storage pool or raid 5 options there, not because it offers any tangible performance increase but because they either need the logical space (Media Servers) or they wanted simpler raid volume to restore in the event of a problem. The cheap MB onboard options leave too much up to chance if something goes south.
 
tangoseal said:
At 4.8 this thing is already wiping its ass with Intel toilet paper. Sorry im a little biased right now.
Click to expand...
4.8 isn't 5.0. :D

And 2 cores @ 4.8, with the rest at 800Mhz = meh

1596739703142.png
 
Lakados said:
The cheap MB onboard options leave too much up to chance if something goes south.
Click to expand...

This is something of a misconception. The only problem you have with this is making sure that when replacing a motherboard, you configure the new one correctly and that you don't jump from Intel to AMD and vice versa. Other than that, it's not a big deal. Intel's RAID volumes are backwards compatible going back to the virtual dawn of IRST.
 
Rockenrooster said:
5 is just a number.
We all learned that with bulldozer.
I'll take a faster CPU at a lower clockspeed no question...
Click to expand...
It was said tongue-in-cheek. If Ryzen 4000 improves the memory OC situation and I can re-use the DDR4-4400 I've got singing on Z390, then I'll definitely be doing a build.

Because I cannot stomach MS Word, Chrome or notepad.exe on anything slower than 35ns latency, sorry.

1596740343293.png
 
Last edited:
odditory said:
It was said tongue-in-cheek. If AMD can improve the memory OC situation then I'm definitely building Ryzen 4000.
Click to expand...

Right now, it isn't that AMD can't overclock memory well. The Ryzen platform can indeed do so. However, past 3800MHz, it's of limited value due to the connection between the Infinity fabric and memory clocks. However, speeds of 5133MHz+ have still been achieved on X570. Where AMD falls short is its ability to clock memory as high as Intel does when utilizing four DIMMs.
 
Axman said:
I would really like it if AMD had better RAID support. Maybe their gains in the server market will carry over to the home user side. Right now a lot of those sales are going to Intel.
Click to expand...
I don't think I've seen RAID set up using the chipset... ever. I know I also haven't done it personally, using a Highpoint controller back when those were a thing once for striping, and using software mirroring in Windows once. These days I just use a NAS for mass storage and local backups as SSDs have obviated any need for storage performance from the likes of RAID 0.

Everything I've seen in server rooms uses a separate controller from whichever vendor supplied the server (Dell, HP, etc).

At most, I could see the utility on workstations that actually do need more local storage performance or redundency but do not have the provision for a separate controller available, but that's a pretty narrow niche.

I think AMD has it covered. If you're using RAID 5 off of the chipset, you're doing computers mostly wrong :D
 
Dan_D said:
This is something of a misconception. The only problem you have with this is making sure that when replacing a motherboard, you configure the new one correctly and that you don't jump from Intel to AMD and vice versa. Other than that, it's not a big deal. Intel's RAID volumes are backwards compatible going back to the virtual dawn of IRST.
Click to expand...
Maybe but I have been hired to try and fix too many MB based raid 5 failures to ever want to inflict that on somebody else. I have just seen too many go seriously south when rebuilding the dead drive and the very limited interface and overall lack of support options makes it something I just can't recommend.
 
IdiotInCharge said:
If you're using RAID 5 off of the chipset, you're doing computers mostly wrong :D
Click to expand...

Yeah, that's mostly the case today, but for a long time in the home theater PC market, it was standard. It's still common with video editing rigs.

Definitely a no-no on the business side of things unless you're a small business on a razor's edge of a budget.

So there's your target market: startup video editing companies. That and people who just want one machine with all the stuff in the same box.
 
IdiotInCharge said:
I don't think I've seen RAID set up using the chipset... ever. I know I also haven't done it personally, using a Highpoint controller back when those were a thing once for striping, and using software mirroring in Windows once. These days I just use a NAS for mass storage and local backups as SSDs have obviated any need for storage performance from the likes of RAID 0.

Everything I've seen in server rooms uses a separate controller from whichever vendor supplied the server (Dell, HP, etc).

At most, I could see the utility on workstations that actually do need more local storage performance or redundency but do not have the provision for a separate controller available, but that's a pretty narrow niche.

I think AMD has it covered. If you're using RAID 5 off of the chipset, you're doing computers mostly wrong :D
Click to expand...
Given AMD's PCIE lane design running raid there is just going to give you a bad time, I do have one small AMD EPYC 3000 series server that is running Raid 1 for the boot drive using the MB LSI based raid controller, but that then uses windows server to use 3 of the remaining drives as a raid 5 volume that the VM's it runs actually runs on but they are low demand VM's. I really do like the Dell PERC raid controllers though, great documentation and awesome support options and solid performance. Hard to go wrong with them, HP burned me a few years ago and I haven't touched their servers in years so I don't know what their raid cards currently look like.
 
Lakados said:
Maybe but I have been hired to try and fix too many MB based raid 5 failures to ever want to inflict that on somebody else. I have just seen too many go seriously south when rebuilding the dead drive and the very limited interface and overall lack of support options makes it something I just can't recommend.
Click to expand...

RAID 5 is something of a different animal using a motherboard controller. I wouldn't recommend using it, despite Intel supporting it. While its been sometime since I've dealt with it, the biggest problem was that onboard RAID 5 used the CPU for all parity calculation and rebuilds. While this seems like it would be OK, the performance difference between a real, hardware based controller with dedicated cache and a cache battery, to onboard RAID 5 is night and day. In fact it takes fewer nights and days to rebuild a RAID 5 array on a real controller than an onboard one. The way people tend to operate systems with a proper controller is probably different as well. IE, not shutting them down in the middle of the rebuild and so on.
 
Last edited:
Lakados said:
Given AMD's PCIE lane design running raid there is just going to give you a bad time, I do have one small AMD EPYC 3000 series server that is running Raid 1 for the boot drive using the MB LSI based raid controller, but that then uses windows server to use 3 of the remaining drives as a raid 5 volume that the VM's it runs actually runs on but they are low demand VM's. I really do like the Dell PERC raid controllers though, great documentation and awesome support options and solid performance. Hard to go wrong with them, HP burned me a few years ago and I haven't touched their servers in years so I don't know what their raid cards currently look like.
Click to expand...

HP's controllers are basically shit. The P82x controllers in particular suck due to the fact that their heat sinks are insufficient for cooling the controller's processor and often the retention mechanisms would fail over time. I've had to diagnose and replace literally hundreds of these in my career. This is a problem I've seen all the way up to the Gen9's. The current Gen10's have been OK, but I've still seen failures, though not necessarily for the same reason.
 
Dan_D said:
RAID 5 is something of a different animal using a motherboard controller. I wouldn't recommend using it, despite Intel supporting it. While its been sometime since I've dealt with it, the biggest problem was that onboard RAID 5 used the CPU for all parity calculation and rebuilds. While this seems like it would be OK, the performance difference between a real, hardware based controller with dedicated cache and a cache battery, to onboard RAID 5 is night and day. In fact it takes fewer nights and days to rebuild a RAID 5 array on a real controller than an onboard one. The way people tend to operate systems with a proper controller is probably different as well. IE, not shutting them down in the middle of the rebuild and so on.
Click to expand...
Generally if you're actually needing the benefits of parity calculation, you're either a) dealing with large datasets approaching or including 'big data' where throwing a whole server away isn't a big deal, or b) you're using a decent full-stack storage solution like ZFS and whatever Red Hat is cooking up for RHEL 8 that isn't ZFS, if you're not using some custom and / or less common vendor-produced job.

Dan_D said:
HP's controllers are basically shit. The P82x controllers in particular suck due to the fact that their heat sinks are insufficient for cooling the controller's processor and often the retention mechanisms would fail over time. I've had to diagnose and replace literally hundreds of these in my career. This is a problem I've seen all the way up to the Gen9's. The current Gen10's have been OK, but I've still seen failures, though not necessarily for the same reason.
Click to expand...
We've kept plenty of spares for Gen7+ HP RAID controllers. I'm still wondering who received the bribe to force us to use HP equipment (and Cisco, for strictly layer 2 work) for our main program. I much prefer the Dell stuff that we used on another. Mostly prefer Dell in general...
 
You must log in or register to reply here.
Back
Top