News AMD Announces 16-Core 32-Thread Ryzen 9 3950X for $749, 4.7 GHz boost, Launches in September

Page 4 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Yup. I am very content with my 480 right now. No reason to upgrade if they don't give any cost incentives.

If your happy then there really isn't a reason right now. I was really expecting to see the 5700XT launch with an MSRP of ~$400, not $450. Slightly better (according to AMD's benchmarks only ~5%) performance, not as efficient (higher TDP- higher power usage, higher heat), and no ray tracing. Unless independent testing shows the 5700XT to be better than AMD's claims I just don't see it as being a compelling product for $450 even if Nvidia doesn't lower the 2070 and keeps it a $500. People will be torn by is it worth $50 to have ray tracing and better efficiency given that overall performance is almost equal. If Nvidia lowers their MSRP by only $25 I can see a lot of people opting for the 2070 to have ray tracing and better efficiency, which will put AMD in a bind. I think if anything we won't see the RTX 2070 drop much in price unless AMD is forced to lower the 5700XT first. The RTX 2060 vs 5700 is a little more compelling as its supposed to be about a 10% performance advantage for the 5700. What we may see is Nvidia dropping the cost of the 2060 to slightly lower than the 5700 but keeping the 2070 at $450 - $470 and charging a premium for their "Super" line.
 
I simply can't understand AMD releasing the 5700XT for $450 while, by their own internal benchmarking, has only a ~5% performance advantage (while having no ray tracing and less efficiency) over the $500 RTX 2070 yet they are releasing the R9 3950X for $750. The R9 3950X, by AMDs internal benchmarking, has nearly equal performance to the i9 9900K in gaming and can flat out dominate the $1,200 i9 9920X in content creation (while having better efficiency). Using this as an example the pricing of the 5700XT is set too high, $400 would have made more sense.
 

mihen

Honorable
Oct 11, 2017
466
54
10,890
AMD isn't going to price their products low if they don't have to. AMD needs CPU market share more than they need GPU market share, and for the first time in 50 years they have a clear advantage over their competitor. So they are probably focusing most of their effort on increasing their consumer and server market share. With GPUs, selling a cheaper to produce card for $50 less than your competitor is probably a good starting price point if your goal is to earn capital. This is also the first generation of a new architecture. They need to test the waters.
 
Nvidia and Intel are two companies with very similar approaches to doing business. They very simply hate to lower the costs of their products especially if they still hold an edge somewhere no matter how small of an edge that may be.
Well, Nvidia's launch pricing does tend to be quite high, but I wouldn't say they don't do price drops. When the GTX 1080 launched, it was $700. Nine months later, the much faster 1080 Ti was launched at that same price point, and the 1080 dropped to $500. And Vega wasn't even coming out for another five months, so it's not like they had any real competition for the 1080 at the time. I suspect they wanted to head off AMD's new cards before they had a chance to make headlines though. Currently, we're nearing nine months since the 20-series launched, so it wouldn't be surprising to see somewhat faster cards take existing price points, and existing cards see their prices slashed to some extent. 20-series pricing was widely considered to be underwhelming relative to what the 10-series had already been offering, so I'm sure a lot of people had been holding off on upgrades due to the pricing of these cards. Reduced prices could help Nvidia tap that market, while disrupting the launch of AMD's new cards.

Intel, on the other hand, seems much less likely to do price adjustments. When Ryzen came out, AMD was offering significantly higher core counts at any given price point, along with similar efficiency and relatively competitive performance per core, but rather than drop prices of Kaby Lake, Intel maintained their existing prices, holding out until Coffee Lake was ready near the end of the year. And even then, they didn't drop prices for Kaby Lake. A new 7700K still costs over $350, and quad-core i5s are still around $200 or more, despite Intel's own newer offerings providing similar performance at significantly lower price tiers. They instead arbitrarily blocked the new processors from running on the older motherboards, forcing anyone upgrading an older system to either pay the unattractive prices for Kaby Lake, or buy a new motherboard.

...lets hope we never go back to overpriced processors offering a whopping 4, 5% IPC gain over the previous generation with the same core count as the previous generation...
It will undoubtedly happen eventually, and you may even see lower performance gains than that, or longer time between generations. Each manufacturing node becomes harder to achieve than the last, and eventually we will get to the point where it becomes impractical to gain performance by shrinking transistors. There might be major breakthroughs in processor design that work around that, but for the most part it seems likely that CPU performance will continue leveling off in the long term. That will also likely affect performance gains for other components as well.

What we may see is Nvidia dropping the cost of the 2060 to slightly lower than the 5700 but keeping the 2070 at $450 - $470 and charging a premium for their "Super" line.
The 2060 already is priced slightly lower than a 5700. At least going by US pricing, the 2060 has an MSRP of $350, and it's possible to find some single-fan models on sale for even less than that. AMD is claiming that the 5700 will be around 10% faster, but it has an MSRP of $380.
 
Undercut, yes. Undercut by less than half? No. There is no way they would undercut Intel by that much.
I'm chiming in late here and I know the conversation has moved on, but I just wanted to respond to this...

When first gen Ryzen was released, the high end SKUs were competitive in heavily threaded productivity tasks against Intel's 8 core 6900K; then priced at nearly $1100. The 1800X was released at less than half that and you could get the 1700 for less than a third of that price @ $330.

Profit depends on sales volume as well as profit margins. The Ryzen 5 3600 gives us pretty solid evidence that AMD expects to make a profit on a 1 chiplet + IO die package priced at $200. Considering that, the margins on the 3950X @ $750 are likely fantastic. After all, compared to the 3600, it's the same package with extra Infinity Fabric traces, an extra chiplet die and better binning, and a better cooler. For that, AMD have decided to charge an extra $550 - nearly four times as much. I can't see too many shareholders complaining.

If we look at the mainstream desktop market, at present enthusiasts wanting the premium CPU on are already expecting to pay around $500 for the 9900K. It looks to me as though AMD have decided to tempt those buyers by offering double the cores at 50% higher cost. Someone who has already decided to spend $1500-2000 on a high end gaming computer might not need much convincing to find an extra $250 for the ludicrous, top of the line, never-before-seen-on-the-mainstream-platform 16 core beast; even if they have little real need for those cores. This is vintage upselling - a tried and true sales and marketing strategy. If you charge $1000 for the CPU though, double the price for double the cores, my guess is that you lose most of those 9900K/$500 CPU budget buyers.

IF the 3950X performs as AMD are touting (and that is still very much an if at this stage), I agree with @jimmySmittyas that as a competitor against Intel's HEDT, the $750 price tag seems too generous on AMD's part. IMHO however, $750 makes perfect sense as a new premium offering on the mainstream platform. Given the much higher volumes on the mainstream platform, it wouldn't surprise me if AMD could justify some potential lost revenue from HEDT customers in an effort to attract high end mainstream buyers to a higher priced and higher profit SKU.

TLDR: I would not assume that the 3950X performance must be lower than teased purely on the basis of its low price in comparison to Intel's HEDT offerings.

Having said all that, we still don't know the impact of multi-die chiplets on gaming. To my knowledge (correct me if I'm wrong), all AMD's gaming demos to date have been on the single chiplet CPUs. That's still a big question mark for me.
 
Last edited:

Ninjawithagun

Distinguished
Aug 28, 2007
747
16
19,165
I'm not ok with the 3600 clock speeds when comparing to AMD's own existing 1600 and 2600 parts, you're not getting much other then IPC unless this thing overclocks well. Personally I'll go to AMD's new mid-range during the holiday season and get a 3800 or 3800X. I don't need the 3900X or 3950X at this point.

I am curious about the 3000-3500X line up and what that will bring to the truly budget constrained.

The 3600 clocks at 4.2Ghz boost (all cores), and the 3600X boosts at 4.4Ghz (all cores). Be careful comparing single core boost clocks from some of the Intel CPUs with all core boost clocks from AMD. My 2700X boosts to 4Ghz all cores regularly with no issues whatsoever. I plan to upgrade to the 3950X, which boosts to 4.7Ghz. For me, the 3950X is a bargain compared to anything Intel makes today. $750 vs. $1500 is all I need to see. Even if Intel's upcoming 10nm CPUs have better IPC and higher clocks, it's not worth double or even triple the price. Intel will never learn.
 
The 3600 clocks at 4.2Ghz boost (all cores), and the 3600X boosts at 4.4Ghz (all cores). Be careful comparing single core boost clocks from some of the Intel CPUs with all core boost clocks from AMD. My 2700X boosts to 4Ghz all cores regularly with no issues whatsoever. I plan to upgrade to the 3950X, which boosts to 4.7Ghz. For me, the 3950X is a bargain compared to anything Intel makes today. $750 vs. $1500 is all I need to see. Even if Intel's upcoming 10nm CPUs have better IPC and higher clocks, it's not worth double or even triple the price. Intel will never learn.

Except they have had better value many times. Its all dependent. he Q6600 was $300 when the QuadFX was well more than double the price and was vastly and inferior product (double the cost, way more power draw, less performance)

If AMD gets to a certain point or equivalent performance with they will price accordingly.

And until the reviews come out I don't think the 3950X is 4.7GHz on all cores. Even with AMD all their boost has been certain cores just like Intel with the highest clocks going to single core clocks. When you set the boost yourself its different. Hell most overclocking since then has been setting boost to all cores at a certain clock which is what most people do. Although Intel is killing even that with the 9900KS which is just 5GHz no matter what.
 
Except they have had better value many times. Its all dependent. he Q6600 was $300 when the QuadFX was well more than double the price and was vastly and inferior product (double the cost, way more power draw, less performance)

If AMD gets to a certain point or equivalent performance with they will price accordingly.

And until the reviews come out I don't think the 3950X is 4.7GHz on all cores. Even with AMD all their boost has been certain cores just like Intel with the highest clocks going to single core clocks. When you set the boost yourself its different. Hell most overclocking since then has been setting boost to all cores at a certain clock which is what most people do. Although Intel is killing even that with the 9900KS which is just 5GHz no matter what.

All core boost of Ryzen depends greatly on what supporting hardware you have. Ryzen is very temperature sensitive, it will only boost as long as there is cooling overhead, and that is "out of the box" settings. In my testing to get the 2700X to boost on its own to an all core 4.2 - 4.25Ghz you really need to be under water cooling and keep the temps ~55C as at 60C the processor will start to automatically downclock itself. With certain motherboards and overclocking this can be overcome. I have an Asus Strix X470-F board that has a "slimmed down" version of Hero VII bios and has PBO overclocking through the use of Performance Enhancer. At level 3 and level 4 it will override AMD's caps most importantly (as far as my testing has shown) EDC. At PE level 4 I can set the EDC as high as 168A and this will allow the processor to easily boost all cores to 4.35Ghz for as long as is needed. These options aren't enabled by default, however it is still using the smart technology to overclock itself. All core boost limitations with Ryzen depend greatly on the motherboard and system cooling.
 

TJ Hooker

Titan
Ambassador
@Redneck5439 I think Ryzen's multi core boosts are still restricted by the CPU power limit, regardless of temp. You can use PBO or whatever to remove/adjust those limits, but I would still consider that to be non-stock operation the same as if you enable MCE on Intel platforms.
 
They didnt call it xfr either.
If the turbo speed is like the ryzen 3000 series, it will be a 1 thread turbo.
Even though the 2600s turbo is 3.9ghz, it usually is around 3.7/3.8 on all cores.

As stated above with Ryzen it depends on your system cooling, what motherboard and bios options you have, and if you are using PBO (Precision Boost Overdrive) which will boost the system higher (all core boost). I have never fully "manually" overclocked my processor, I use PBO and have all core boost of 4.35Ghz.
 
@Redneck5439 I think Ryzen's multi core boosts are still restricted by the CPU power limit, regardless of temp. You can use PBO or whatever to remove/adjust those limits, but I would still consider that to be non-stock operation the same as if you enable MCE on Intel platforms.

I wouldn't call it strictly "stock" however PBO is a simple bios option as is Performance Enhancer. All you have to do is enable those options and give it a bump in offset voltage and Ryzen is off to the races. I've helped several people who have never "overclocked" before and it is extremely easy compared to the "old school" multiplier overclocking.
 
  • Like
Reactions: TJ Hooker

TJ Hooker

Titan
Ambassador
Oh, don't get me wrong, it sounds like a great feature from everything I've heard. I'm just saying that, IMO, when somebody talks about Ryzen multi-/all-core boost clocks that should be without PBO enabled. Unless one is specifically talking about tweaked/OC'd performance. In the same way that I would never claim that an 8700K has a 4.7 GHz all core turbo, even though all that may take is flipping on a single BIOS setting (which some boards may even enable by default).
 
Oh, don't get me wrong, it sounds like a great feature from everything I've heard. I'm just saying that, IMO, when somebody talks about Ryzen multi-/all-core boost clocks that should be without PBO enabled. Unless one is specifically talking about tweaked/OC'd performance. In the same way that I would never claim that an 8700K has a 4.7 GHz all core turbo, even though all that may take is flipping on a single BIOS setting (which some boards may even enable by default).

I see exactly what you are saying, and you are correct. The two technologies are very similar. PBO is a great option and a simple click to enable in bios, however not every user can take full advantage of it due to motherboard VRMs, bios options, and of course the need for premium aftermarket cooling solutions. I myself am using a NH-U14S with two 140mm fans in push pull and have 6 140mm case fans, I'm sure someone on stock cooling won't be able to achieve 4.35Ghz all core with PBO. The Prism cooler is very good, especially for a stock cooler, however before upgrading to the NH-U14S I was limited to using PE level 2 and my PBO all core boost was limited to ~4.2Ghz all core, even with all the case fans. For users with cooling overhead and a capable motherboard, PBO is an amazing option and shows what the processor is actually capable of.
 
  • Like
Reactions: TJ Hooker

MasterMadBones

Distinguished
Except Lisa said they should be on level IPC wise. And even if its only 12.5% more cores it still is an entire core plus the SMT threads.
Lisa Su said they should be on level in single core performance. That makes sense, given Zen's IPC deficit was about 5%. A 15% increase should put it around 8-10% ahead. In terms of clock speed, Ryzen 3000 is only about 5-10% behind as well, depending on the price point. That would put them right on par with Intel, before security patches are applied.

The gaming performance is more interesting, because the additional L3 cache will not benefit each game equally. The Windows 1903 scheduler update may make up for the difference, and it hadn't been installed yet in AMD's slides. From my personal experience, some games really did improve 10-15% on my 2600X, which doesn't even have the chiplet problem.

All in all, although the gaming results may be cherry-picked to a degree, AMD seems to have a lot of performance in hand over Intel as well. They seem to be looking to exceed their pitched expectations this time, which is always good marketing.
 
Lisa Su said they should be on level in single core performance. That makes sense, given Zen's IPC deficit was about 5%. A 15% increase should put it around 8-10% ahead. In terms of clock speed, Ryzen 3000 is only about 5-10% behind as well, depending on the price point. That would put them right on par with Intel, before security patches are applied.

The gaming performance is more interesting, because the additional L3 cache will not benefit each game equally. The Windows 1903 scheduler update may make up for the difference, and it hadn't been installed yet in AMD's slides. From my personal experience, some games really did improve 10-15% on my 2600X, which doesn't even have the chiplet problem.

All in all, although the gaming results may be cherry-picked to a degree, AMD seems to have a lot of performance in hand over Intel as well. They seem to be looking to exceed their pitched expectations this time, which is always good marketing.

Every single benchmark that comes from the manufacture is cherry picked. Its why they compare only certain CPUs in certain cases or sections. Its why they can never be trusted. AMD did this with Bulldozer. Every single slide made it look like it was better than Intel. Then the third party sites got a hold of the chips and did their reviews and the truth as not as clear as AMD made it out to be.

I still don't see a 16 core 32 threaded CPU beating an 18 core 36 threaded CPU by 30% in a high thread count optimized scenario much like I wouldn't see Intels 12 core beating AMDs 16 core either.

The chiplet design will present some issues. Mainly communications will be constraint to probably the same speeds as memory since the interconnect does seem to show reliance on that. We shall see when reviews hit.

I am interested to see Intels Forevos design compared to AMDs chiplet design as I would imagine Intels Forevos would have a speed advantage based on the information out there but it will depend on the reviews in a few years when we finally get one.
 

MasterMadBones

Distinguished
Every single benchmark that comes from the manufacture is cherry picked. Its why they compare only certain CPUs in certain cases or sections. Its why they can never be trusted. AMD did this with Bulldozer. Every single slide made it look like it was better than Intel. Then the third party sites got a hold of the chips and did their reviews and the truth as not as clear as AMD made it out to be.
It's unfair to compare Zen 2 to Bulldozer. Zen 1 did live up to AMD's claims, even if those claims were only for it to be equal to Broadwell. Also Lisa Su said that AMD actually expected a full Ice Lake launch this summer, which hasn't happened.
I still don't see a 16 core 32 threaded CPU beating an 18 core 36 threaded CPU by 30% in a high thread count optimized scenario much like I wouldn't see Intels 12 core beating AMDs 16 core either.
As someone else already mentioned, that's only a 12.5% difference in the number of threads. If the 3950X has the same all-core boost behavior under strong cooling we've seen from Ryzen 1000 and 2000, I see no reason why an upset couldn't be true, although 30% is another story. But if we look at AMD's own slides for the 3900X vs the 9920X, I wouldn't hold my breath for this so quickly. Regardless, the difference will be minimal. Maybe Geekbench benefits massively from the larger L3 cache for some reason, which would make it an outlier.
The chiplet design will present some issues. Mainly communications will be constraint to probably the same speeds as memory since the interconnect does seem to show reliance on that. We shall see when reviews hit.
This is mostly true, which is why the L3 cache is so huge. AMD is trying market it as some revolutionary gaming feature, but the truth is that they can't do without it.
The IF's clock is still tied to the memory clock, but the link width has doubled to 512 bits, which should improve throughput tremendously, even at the same transfer rate. The real problem is the the extra latency because the I/O die is an extra 'hop'. The new Windows scheduler will alleviate much of this problem, until threads start to spill over to the other die.
 
It's unfair to compare Zen 2 to Bulldozer. Zen 1 did live up to AMD's claims, even if those claims were only for it to be equal to Broadwell. Also Lisa Su said that AMD actually expected a full Ice Lake launch this summer, which hasn't happened.

As someone else already mentioned, that's only a 12.5% difference in the number of threads. If the 3950X has the same all-core boost behavior under strong cooling we've seen from Ryzen 1000 and 2000, I see no reason why an upset couldn't be true, although 30% is another story. But if we look at AMD's own slides for the 3900X vs the 9920X, I wouldn't hold my breath for this so quickly. Regardless, the difference will be minimal. Maybe Geekbench benefits massively from the larger L3 cache for some reason, which would make it an outlier.

This is mostly true, which is why the L3 cache is so huge. AMD is trying market it as some revolutionary gaming feature, but the truth is that they can't do without it.
The IF's clock is still tied to the memory clock, but the link width has doubled to 512 bits, which should improve throughput tremendously, even at the same transfer rate. The real problem is the the extra latency because the I/O die is an extra 'hop'. The new Windows scheduler will alleviate much of this problem, until threads start to spill over to the other die.

I am not comparing Zen 2 to Bulldozer. Just the marketing. Its the same tactic that was used. Compare specific areas only to CPUs that it wins against. Considering the 3950X is a mainstream processor I wonder why it is not being compared even in power to its main competition, the 9900K.

Its a big reason why I don't trust OEM slides with performance comparison. The only ones I do trust are when they compare to their own previous products. Its why everything performance wise I wait until TH, Anand etc do a proper review.
 
  • Like
Reactions: NightHawkRMX
I am not comparing Zen 2 to Bulldozer. Just the marketing. Its the same tactic that was used. Compare specific areas only to CPUs that it wins against. Considering the 3950X is a mainstream processor I wonder why it is not being compared even in power to its main competition, the 9900K.

Its a big reason why I don't trust OEM slides with performance comparison. The only ones I do trust are when they compare to their own previous products. Its why everything performance wise I wait until TH, Anand etc do a proper review.

I don't see the R9 3950X as as the i9 9900Ks main competition. I don't see how it could be... The 9900K is a ~$500 CPU and the 3950X is a $750 CPU. The 9900K is 8 cores and 16 threads, the 3950X is 16 cores and 32 threads. The two processors aren't in the same competition space no matter how you look at it. The 3900X at $500 is at the same price point as the 9900K, so the argument can be made that it is the 9900K's main competition, however even with this we are comparing a 12 core processor to an 8 core processor. You could also compare the 3800X as it is also 8 cores and 16 threads, however it isn't in the same price bracket as it is only $400.

Really the i9 9900K's main competition would have to be the R9 3900X as they are both in the $500 range. The 3950X really doesn't have a direct competitor as it is priced $250 more than the 9900K and ~$250 less than the next closest Intel processor the 10 core i9 9900X ($1000) and $450 less than the 12 core i9 9920X ($1200). If AMD comes anywhere close to the performance most think the 3000 series will have they are priced to downright dominate the market and Intel will be forced to lower their extreme overpricing.
 
I don't see the R9 3950X as as the i9 9900Ks main competition. I don't see how it could be... The 9900K is a ~$500 CPU and the 3950X is a $750 CPU. The 9900K is 8 cores and 16 threads, the 3950X is 16 cores and 32 threads. The two processors aren't in the same competition space no matter how you look at it. The 3900X at $500 is at the same price point as the 9900K, so the argument can be made that it is the 9900K's main competition, however even with this we are comparing a 12 core processor to an 8 core processor. You could also compare the 3800X as it is also 8 cores and 16 threads, however it isn't in the same price bracket as it is only $400.

Really the i9 9900K's main competition would have to be the R9 3900X as they are both in the $500 range. The 3950X really doesn't have a direct competitor as it is priced $250 more than the 9900K and ~$250 less than the next closest Intel processor the 10 core i9 9900X ($1000) and $450 less than the 12 core i9 9920X ($1200). If AMD comes anywhere close to the performance most think the 3000 series will have they are priced to downright dominate the market and Intel will be forced to lower their extreme overpricing.

I compare CPU's on the value they give me at a certain price. If Intel'$ CPU cost $500.00 then I will compare it to AMD's closest priced CPU. If AMD can give me 4 extra cores then I'm still fine making that comparison because the prices are the same. I also look at the cost of the Motherboard, RAM, CPU Cooler and what ever else I need to make a complete system for a given CPU.

So AMD's Ryzen 3900X and the Intel 9900K to me are what I would compare. The proper Intel motherboard's are pricey for the Intel 9900K and it's looking like I would need one of the better X570 or possibly the rumored X590 motherboard to safely run and possibly overclock the 3900X so in this can lets just assume both Intel and AMD motherboards are going to be about the same price.

AMD ships with a respectable CPU Cooler although I doubt it will be of much use for overclocking or sustained work loads. Intel requires a ultra high end CPU Cooler like a Noctua NH-14 or better a high end AIO CPU Cooler or better to overclock. So in this case I am going to give the win to AMD because at least initially I can hold off on buying a CPU Cooler and can upgrade later.

AMD looses in RAM price. AMD Ryzen CPU's simply love lower timing, high speed memory so here we are looking at a minimum of DDR4-3200 CAS 15 or better from a reputable brand where the Intel 9900k can get away with lesser RAM as it is less dependent on memory speed and timings.

INTEL 9900K
============
CPU = ~$500
CPU Cooler = $90 (Noctua - NH-D14 64.95 CFM CPU Cooler) * May need to upgrade if overclocking is desired.
Motherboard = $250 (ASRock - Z390 Taichi Ultimate ATX LGA1151 Motherboard) ($120-$500+)
DDR4 2666 C15 = $170 (Corsair - Dominator Platinum 32 GB (2 x 16 GB) DDR4-2666 Memory)
===============================================================================================================
TOTAL = $1010 USD

AMD 3900X
============
CPU = ~$500
CPU Cooler = Included * May need to upgrade for best performance and if overclocking is desired.
Motherboard = $190-$500+, I'm going to assume a solid motherboard will start around $250 like the ASRock Z390 Taichi Ultimate listed above.
DDR4 3200 C14 = $250 (G.Skill - Trident Z 32 GB (2 x 16 GB) DDR4-3200 Memory)
===============================================================================================================
TOTAL = $1000 USD