AMD Ryzen Threadripper 5000 Series May Reintroduce 16-Core SKU

[Tsing]

Image: AMD



AMD’s current Ryzen Threadripper 3000 Series only comes in 64-, 32-, and 24-core flavors, but a 16-core SKU could be making a return for the upcoming, Zen 3-based “Genesis Peak” lineup. That’s according to DRAM Calculator for Ryzen author Yuri Bubliy (1usmus), who shared a HEX riddle that converts to an interesting line of ASCII text.



“I think some of you will want to solve this riddle,” Bubliy tweeted. “47454e4553495320313620434f524553.” Throwing that gibberish into a hexadecimal to ASCII text converter gives us the following:



HEX: 47454e4553495320313620434f524553ASCII: GENESIS 16 CORES



Specifications of what will presumably...

Continue reading...

 

[Dogsofjune]

Hopefully it's better than the 1900x was compared to it's AM4 counterpart.

Should be an entry level option

 

[LazyGamer]

Hopefully it's better than the 1900x was compared to it's AM4 counterpart.

Should be an entry level option

Entry-level for some; but if they tune the 16-core part for single-thread competitiveness with the 5950X and 5900X, they'd have a market.

You'd essentially just be paying for more PCIe lanes (and more memory channels), which makes quite a bit of sense for some.

And most don't really need the cores. It's the connectivity for more GPUs for compute, more M.2 slots on risers for high-IOPS data access, more NICs for network access and so on. You could even throw in a SATA or SAS array if you needed real mass storage on the local machine for whatever reason.

The real use for these is going to be some combination of the above. It's easy to add just a high-speed NIC, or just a second GPU, etc. to a system, but when you need two or more of those options things start to get messy, and that's where HEDT like Threadripper really comes into its own, even before you get to the high core count parts.

 

[Grimlakin]

I wonder with SAM and 4 channel memory.. could that be the gaming chip to have?

 

[Auer]

Be nice with a M-ITX motherboard.....

 

[Dogsofjune]

It would be nice in itx, but the best I have seen is a matx from Asrock.

To many features to stuff on to a tiny board

 

[LazyGamer]

It would be nice in itx, but the best I have seen is a matx from Asrock.

To many features to stuff on to a tiny board

Even then you're thin on usecases IMO. A pair of Thunderbolt ports and it'd make a bit of sense for mATX, as then you'd have a use for the extra lanes, and if they put in like a 2x2 vertical M.2... assembly... of some sort, you'd be close to having them populated.

[Of course, AMD and Nvidia just need to start putting Thunderbolt ports on GPUs at this point]

 

[Grimlakin]

Even then you're thin on usecases IMO. A pair of Thunderbolt ports and it'd make a bit of sense for mATX, as then you'd have a use for the extra lanes, and if they put in like a 2x2 vertical M.2... assembly... of some sort, you'd be close to having them populated.

[Of course, AMD and Nvidia just need to start putting Thunderbolt ports on GPUs at this point]

Hummm I understand for display but it will more likely be thunderbolt/USB combo ports using the USB 3.1 technology.

 

[Dan_D]

Be pointless with a M-ITX motherboard.....

I fixed that for you.

Everything that makes the HEDT platform compelling is lost when pairing such a CPU with mini-ITX motherboards. The only reason to use a lower count HEDT CPU over the usual mainstream CPU options in the first place would be for the extra PCIe lanes, memory channels and support for larger amounts of memory. You wouldn't get any of that with a mini-ITX based Threadripper system. It would be limited to one PCIe x16 slot and 2x DIMM slots. Of course the CPU's physical footprint itself is really too large for that form factor anyway. This would necessitate proprietary CPU coolers or something designed for 1-U servers.

We've seen this play out before with physically smaller Intel HEDT / Workstation / Server CPU's on X99 m-ITX motherboards. It really doesn't make sense. Back in the day when the X99-ITX/ac came out, I tested it and of course the lack of memory bandwidth was a factor, but you could at least make an argument for the additional core count being compelling in that form factor if the expansion and memory capacity / bandwidth weren't key to your applications. This was because you had quad-cores or less in the mainstream segment and upwards of 8-10 via LGA 2011 CPU's. More if you opted for a Xeon over a Core i7.

Today it makes less sense as you can get 16c/32t CPU's on socket AM4 which lends itself to use in SFF systems far better than sTRX4 does. Remember AMD's Threadripper 2990WX? Remember how its lack of memory bandwidth crippled it in some applications compared to its Epyc counterparts? Well, that was a thing. That CPU was knocked down to quad-channel versus 8-channel and half the CPU basically had no memory of its own and had to access it all remotely. You are talking about knocking a 16c/32t part down to dual-channel mode. Probably not an issue for some applications, but it very well could be for applications were a 16c/32t TR part would ordinarily make some sense over a conventional 3950X or 5950X CPU.

 

[Zarathustra]

Could it be?

Has AMD been reading my forum posts for the last two years?

I have long been wishing, hoping and praying for a fewer core Threadripper with higher clocks.

The reason? I don't need a massive number of cores. My workloads benefit more from high per core performance.

At the same time there is no way AM4 has enough PCIe lanes for me. I crave as many PCIe lanes as I can get.

If they launch this, I hope they don't pull an Intel, and offer fewer PCIe lanes on their fewer core models.

If:
- they launch this chip,
- it has the same number of PCIe lanes as existing Threadrippers
- it fits in my existing TRX40 motherboard; and
- clocks are binned such that they are high

...I could totally see myself replacing my 3960x with one of these on launch.

 

[Zarathustra]

Entry-level for some; but if they tune the 16-core part for single-thread competitiveness with the 5950X and 5900X, they'd have a market.

You'd essentially just be paying for more PCIe lanes (and more memory channels), which makes quite a bit of sense for some.

And most don't really need the cores. It's the connectivity for more GPUs for compute, more M.2 slots on risers for high-IOPS data access, more NICs for network access and so on. You could even throw in a SATA or SAS array if you needed real mass storage on the local machine for whatever reason.

The real use for these is going to be some combination of the above. It's easy to add just a high-speed NIC, or just a second GPU, etc. to a system, but when you need two or more of those options things start to get messy, and that's where HEDT like Threadripper really comes into its own, even before you get to the high core count parts.

Agree wholeheartedly with this!

 

[Zarathustra]

Hopefully it's better than the 1900x was compared to it's AM4 counterpart.

Should be an entry level option

Well, that was largely due to the 1900x having issues with the chip interconnect timing, right? I believe that Zen2 did away with this issue completely. At least that is what I recall.

 

[Dogsofjune]

Well, that was largely due to the 1900x having issues with the chip interconnect timing, right? I believe that Zen2 did away with this issue completely. At least that is what I recall.

I could be wrong, but I don't think the 1900x suffered from that. Not like the 1950x, 2950x and up did. And that was more of an AMD/ Microsoft issue. Linux didn't suffer with the oddity. I will have to look and revisit that topic again. It's been a bit since I read up on that.

I just remember that the 1900x I had lagged a tad behind an 1800x in most day to day things. All it had going for it was PCI lanes. quad channel ram, and the ability to get hot quick. Was great in the winter as a supplemental heater, and if you got the rotisserie option, made a fine roasted chicken.
I love my 2920x, but it too lacks for day to day stuff. It's a plow horse, not an Arabian. Great for cartoons and editing and occasionally Path of Exile.

I'm excited to see what this new version will bring to the table, but really, if one goes the TR route, you go for the cores, quad channel ram, and the lanes.
Server, render box, OnlyFans editing is what I feel this is aiming for. If more than 8/8 cores would be wanted, at least in my user case. Otherwise, wouldn't the 5900x, 5950x be the more practical route?

 

[Zarathustra]

I could be wrong, but I don't think the 1900x suffered from that. Not like the 1950x, 2950x and up did. And that was more of an AMD/ Microsoft issue. Linux didn't suffer with the oddity. I will have to look and revisit that topic again. It's been a bit since I read up on that.

I just remember that the 1900x I had lagged a tad behind an 1800x in most day to day things. All it had going for it was PCI lanes. quad channel ram, and the ability to get hot quick. Was great in the winter as a supplemental heater, and if you got the rotisserie option, made a fine roasted chicken.
I love my 2920x, but it too lacks for day to day stuff. It's a plow horse, not an Arabian. Great for cartoons and editing and occasionally Path of Exile.

I'm excited to see what this new version will bring to the table, but really, if one goes the TR route, you go for the cores, quad channel ram, and the lanes.
Server, render box, OnlyFans editing is what I feel this is aiming for. If more than 8/8 cores would be wanted, at least in my user case. Otherwise, wouldn't the 5900x, 5950x be the more practical route?

Yeah, this is one of the reasons I waited for the Zen2 Threadrippers. I wanted a "no compromises" solution. A fast arabian without having to sacrifice the PCIe lanes. I was actually disappointed they forced me to buy 24 cores, and that it didn't have higher turbo clocks than the more core versions.

 

[hubaduba]

Yeah, this is one of the reasons I waited for the Zen2 Threadrippers. I wanted a "no compromises" solution. A fast arabian without having to sacrifice the PCIe lanes. I was actually disappointed they forced me to buy 24 cores, and that it didn't have higher turbo clocks than the more core versions.

That's been a disappointment to me since Ryzen was introduced - fewer cores should be able to hit higher boost speeds than more cores in the same package. AMD makes you pay for cores you don't need to get the clock speed you want. It does seem to be less of an issue with Zen 3, since the chips appear to have similar boost behavior with the same number of cores loaded.

 

[Zarathustra]

That's been a disappointment to me since Ryzen was introduced - fewer cores should be able to hit higher boost speeds than more cores in the same package. AMD makes you pay for cores you don't need to get the clock speed you want. It does seem to be less of an issue with Zen 3, since the chips appear to have similar boost behavior with the same number of cores loaded.

I have come around on this.

Initially I felt like you do, but I think the mechanics have changed on 7nm. It used to be that adding lots of cores created lots of heat, and this the chips became heat limited and had to clock lower.

These new 7nm chips sip power to the point where it takes a lot more cores to reach the thermal limit.

At the same time smaller process nodes are more difficult to hit high clocks.

I think the many cores without much reduction in clocks is more of a symptom of this than it is an intentional strategy.

 

[Brian_B]

Seems like we are very close to it being irrelevant the total number of cores installed on a package - if you are only driving one (or a small number), PBO/Boost/etc will crank it up to where it can and you should be near the limit of the architecture and process node.

 

[Dan_D]

That's been a disappointment to me since Ryzen was introduced - fewer cores should be able to hit higher boost speeds than more cores in the same package.

This is simply untrue. Core count has no impact on boost clock behavior. At least not single-threaded boost clock behavior. There is some impact on higher core count parts as it relates to multi-threaded workloads but actual clocks come down to cooling and total package power.

AMD makes you pay for cores you don't need to get the clock speed you want. It does seem to be less of an issue with Zen 3, since the chips appear to have similar boost behavior with the same number of cores loaded.

Only in so much as AMD reserves its best binned dies for the higher end parts. There aren't enough to go around and AMD made the choice to put them in the higher end chips.