TR Forums

Needless to say, I've been reading as many reviews and watching as many reviews of Ryzen as possible.

TR, ComputerBase.de, TechSpot, HardwareUnboxed, Guru3D and HardwareCanucks show 1800X 5-7% behind 5960X on average in terms of 1080P game performance, even using a Titan X (P) in certain reviews. In tests with SMT disabled, Ryzen gains another 2-3% over all. I'm impressed as:

1) Windows 10 obviously doesn't play well with Ryzen's SMT implementation at this point.

2) Ryzen is at a HUGE handicap in terms of memory bandwidth AND latency.

3) Even with a Titan X (P) at 1080P which makes the CPU bottleneck that much worse... it still kept up with the world class Haswell CPUs.

I have no doubt at this point when the Windows thread scheduling, UEFI microcode or apps get optimized that Ryzen will bridge the gap with the 5960X and maybe even be able to close in on the 6900K.

Of course the Kaby Lake i5 7600K and i7 7700K are still better choices for strictly gaming, however they're only a few percent better and if one really does need the cores/threads for productivity and can't afford to splurge on Haswell-E / Broadwell-E... AMD has found a relevant place in the market!

I'm even more excited now to see what Ryzen 5 will bring!

Given the SMT weirdnesses (which I think Intel had to deal with themselves back in the P4 days), I'm considering getting an R7 after all... and disabling SMT, leaving me with a octocore system.

This might end up helping the memory bandwidth problem (assuming that it is a problem) by reducing the (virtual) core count.

That should be both a better and cheaper option than the Haswell-E system I was considering as an alternative.

Well there are some similarities. K7 was only competitive with the P3 at the same clock some of the time, and there was some weirdness with the Irongate AMD 750 chipset. Today, Zen is sort of competitive in certain workloads, and there are boards waiting for the BIOS kinks to get worked out.

K7+ (or Athlon XP) with SSE support and a faster FSB was a real thorn in Intel's side. Maybe Ryzen+ will be the same?

derFunkenstein wrote:

Well there are some similarities. K7 was only competitive with the P3 at the same clock some of the time, and there was some weirdness with the Irongate AMD 750 chipset. Today, Zen is sort of competitive in certain workloads, and there are boards waiting for the BIOS kinks to get worked out.

K7+ (or Athlon XP) with SSE support and a faster FSB was a real thorn in Intel's side. Maybe Ryzen+ will be the same?

I'm sure that the Zen2 core is going to be where it's at. We just need to buy Zen1 so that AMD survives that long.

Vhalidictes wrote:

I'm sure that the Zen2 core is going to be where it's at. We just need to buy Zen1 so that AMD survives that long.

I'd say they're not on the way out yet. I'm not saying it's all golden or whatever, but considering a few years ago we didn't even know if they'd last to this point...

- Ryzen, Vega, and the 500 series(?) all launching now/soon, giving them a lot of $ now and down the road.
- OEM sales from where Kaveri was went up a hell of a lot, so their overall sales went from damn near 0 to respectable, if not great. (Might be slowing down until Zen APU hits, but still.)
- Obviously, the console sales from being in two of the largest sellers have helped, though it's not likely a huge amount of money it is solid business.
- Despite their debt, they've managed to pay off significant amounts of the short term debt with fairly small losses every year comparatively.

I'd say even if they don't suddenly start turning gigantic profits, they've got pretty solid footing right now.

I think when we see the full Ryzen lineup, including true 'desktop' chips meant to compete in gaming, we'll see ... better competition in gaming. As much as everyone wants them to, the R7 just doesn't focus on the 7600K/7700K.

MOSFET wrote:

I think when we see the full Ryzen lineup, including true 'desktop' chips meant to compete in gaming, we'll see ... better competition in gaming. As much as everyone wants them to, the R7 just doesn't focus on the 7600K/7700K.

I agree. The R3 should clock the best, having fewer cores and possibly no HT. In fact, Zen1 and Zen2 might have the same clocks, assuming that Zen2 is an APU and will therefore have the extra heat/power to deal with.

Not even slightly the case.
The processor needs to be significantly cheaper to offer what the K7 did.


When I purchased a K7, I did so because it was 80% the performance of an Intel P4 setup at only 50% of the cost, maybe 60% at most.
/Furthermore/ back in those days, it was $1200 AUD to build the Pentium 4 system vs the cost of about $600? for the K7, maybe $700?
I saved $500 by avoiding the P4 and got 80% of the speed.
Nowadays the price difference isn't significant enough and the TCO isn't as high. It's not $700 vs $1200, it might end up being $550 vs $700. Only $250 difference, not the $500 difference of old, due to the cost, /overall/ of components dropping.


You could argue the R3 and R5 Ryzens are more in line in competing with Kaby Lake, however they aren't out yet to discuss AND even if they were,.........
Kaby Lake systems come with an iGPU. Many of you here at Techreport couldn't care less about this. *I* do and MILLIONS of others do, infact I'd argue the majority of Kaby Lake purchases will infact be using the iGPU. That raises the cost of the AMD system even more for me, by say $50 at the bare minimum.

So, believe it or not, I firmly believe the AMD CPUs need to cost EVEN LESS for me to go "right, ok this is just ridiculously great value and I see a logical reason to switch here"

MOSFET wrote:

I think when we see the full Ryzen lineup, including true 'desktop' chips meant to compete in gaming, we'll see ... better competition in gaming. As much as everyone wants them to, the R7 just doesn't focus on the 7600K/7700K.

Yup. There are so many claiming "foul" on reviewers for showing poor gaming performance when compared to the 76-7700k, but that's true for 6 and 8 core parts from Intel as well.

I don't blame the reviewers, I blame AMD for an absurd marketing campaign. AMD should have first introduced a 4 core 8 thread CPU with 95% of the performance of the 7700k when both are overclocked - yet with a price tag that's $199. THAT would wake-up the market.


..I should also note that AMD should have had board-partners for their m-boards producing ecc capable boards when the latest 8 core parts were presented to market. (..and they should be capable of utilizing registered DIMM's. EDIT: did a web search, no registered DIMM, also while most boards are capable of using unbuffered ECC memory, NONE currently support using it AS ECC memory.)

Intel could just convince Microsoft and other software developers to require a feature. After all, even the earliest Socket 423 Pentium 4 has SSE2 so can run today's Firefox, Chrome, Flash, and Office 2013, which is something not even the newest Socket A Athlons like Barton can do.

Then there's the CMPXCHG16b instruction suddenly required to run 64-bit Win8.1 or Win10, just to make it so no S754 or 939 A64 can run 64-bit software except in Win7.

I think it has more in common with Phenom II launch. It is a compelling option for those who do more than just causal stuff with their computer. Intel has the edge on their mainstream and portable platforms due to lower power consumption (not as large as of a gap as Bulldozer versus Sandy Bridge days) and product availability.

Speaking of production and many-core workloads, it's kind of funny that AMD has managed to make a better "Bulldozer" than Bulldozer, while primarily aiming for something else.

Meadows wrote:

Speaking of production and many-core workloads, it's kind of funny that AMD has managed to make a better "Bulldozer" than Bulldozer, while primarily aiming for something else.

Good thing this Bulldozer Mk. II is getting 85-90% of the single threaded performance of the recent Intel CPUs; FX 8150 could barely managed 2/3's of Sandy Bridge in single threaded loads. The only area where it was competitive was 7-Zip.

AbRASiON wrote:

Not even slightly the case.
The processor needs to be significantly cheaper to offer what the K7 did.

Kaby Lake systems come with an iGPU. Many of you here at Techreport couldn't care less about this. *I* do and MILLIONS of others do, infact I'd argue the majority of Kaby Lake purchases will infact be using the iGPU. That raises the cost of the AMD system even more for me, by say $50 at the bare minimum.

Assuming Ryzen 5 1600X is able to score similar in 99th percentile framerate as the 1800X (i.e. 76 for 1800X versus 86 for the 7700K): That's roughly 88.4% of the performance for $80 less (assuming the rumored $270 price point is true). Then I'd also have to wonder how well the quad core, octo thread 1400X would hold up. If either of those perform on par with 1800X in gaming it will be quite disruptive indeed.

As for iGPU users, those people are probably confined to laptops. I doubt any serious desktop user would be using iGPU outside of HTPC use. However there will be Raven Ridge, which is essentially Summit Ridge + iGPU coming within the next two quarters or so.

ultima_trev wrote:

Meadows wrote:

Speaking of production and many-core workloads, it's kind of funny that AMD has managed to make a better "Bulldozer" than Bulldozer, while primarily aiming for something else.

Good thing this Bulldozer Mk. II is getting 85-90% of the single threaded performance of the recent Intel CPUs; FX 8150 could barely managed 2/3's of Sandy Bridge in single threaded loads. The only area where it was competitive was 7-Zip.

AbRASiON wrote:

Not even slightly the case.
The processor needs to be significantly cheaper to offer what the K7 did.

Kaby Lake systems come with an iGPU. Many of you here at Techreport couldn't care less about this. *I* do and MILLIONS of others do, infact I'd argue the majority of Kaby Lake purchases will infact be using the iGPU. That raises the cost of the AMD system even more for me, by say $50 at the bare minimum.

Assuming Ryzen 5 1600X is able to score similar in 99th percentile framerate as the 1800X (i.e. 76 for 1800X versus 86 for the 7700K): That's roughly 88.4% of the performance for $80 less (assuming the rumored $270 price point is true). Then I'd also have to wonder how well the quad core, octo thread 1400X would hold up. If either of those perform on par with 1800X in gaming it will be quite disruptive indeed.

As for iGPU users, those people are probably confined to laptops. I doubt any serious desktop user would be using iGPU outside of HTPC use. However there will be Raven Ridge, which is essentially Summit Ridge + iGPU coming within the next two quarters or so.

88.4% of the performance for $80 less, but missing my point, that in the old days, it was $500 less...........the TCO was significantly higher.
Now, why not just pay the $80 to be sure?

Like I said, it needs to be /even cheaper/ but ideally, it should be faster. Maybe the 6 core will do 4.3ghz? (I doubt it) but if it did, maybe that with a $30 price drop, that might make for a reasonable bit of hardware.

As for the iGPU stuff, I think you're wildly estimating incorrectly. Wildly. Firstly, business desktops, we had thousands at my work, Intel CPUs with an iGPU and not just 45W stuff, some of the 65 and even the occassional 91 / 95W stuff. (Mostly 65W)
Plus PC users who just don't care for games, PLUS on a modern iGPU, really basic gaming is actually viable.

That lack of a GPU is a shortfall, not a massive one, but when it adds, rather than subtracts to the price, it's an issue.

What your company does may differ from many others.

I work as a sys admin in a Fortune 500 company and the vast majority of our desktops have been phased out in favor of laptops + docks. The previous thin clients we did use ironcially used AMD APUs, although you probably don't need a Kaby Lake i3 let alone i5 for spread sheets and word docs, or in my case, SSH terminal/RDC for remoting to servers.

Meadows wrote:

Speaking of production and many-core workloads, it's kind of funny that AMD has managed to make a better "Bulldozer" than Bulldozer, while primarily aiming for something else.

I wouldn't be so sure they were aiming for "something else". 8C/16T is overkill for HEDT/gaming use case, and Bulldozer would've been an OK server CPU if the performance/watt had been better.

So...

AMD is still smarting from losing all the gains they'd made in server market share with Opteron. They also see that a lot of consumers are ditching traditional desktops for mobile devices, and deployment of back-end infrastructure ("The Cloud") is surging. They initially bet the farm on their enterprise ARM initiative, but it became clear that that wasn't going to save them; so really the only choice they had (aside from exiting the broader CPU market and focusing entirely on console SoCs) was to pivot back towards making an x86 CPU which would make them competitive in the datacenter again.

So why would they release the desktop part first, then? That makes a certain amount of sense too. In effect, they're using the HEDT/gaming market as a beta test for their next wave of server CPUs.

ultima_trev wrote:

What your company does may differ from many others.

I work as a sys admin in a Fortune 500 company and the vast majority of our desktops have been phased out in favor of laptops + docks.

It is actually difficult to even get a desktop at my current job. Everyone uses laptops. The group I am in has a slightly different use case than most of the other developers ("closer to the metal"), and because of this it is useful to have systems we can plug PCIe cards and HDDs into. We ended up scrounging parts left over from a datacenter upgrade and built ourselves a few "Frankenservers" which now live under a table in our area of the office.

ultima_trev wrote:

I doubt any serious desktop user would be using iGPU outside of HTPC use.

I DO! Corei7 4790K with iGPU. Why? I don't play games but I still want a fast CPU. Rather than waste my money on a graphics card, went for higher CPU clock speed and 2 SSDs.

just brew it! wrote:

It is actually difficult to even get a desktop at my current job. Everyone uses laptops. The group I am in has a slightly different use case than most of the other developers ("closer to the metal"), and because of this it is useful to have systems we can plug PCIe cards and HDDs into. We ended up scrounging parts left over from a datacenter upgrade and built ourselves a few "Frankenservers" which now live under a table in our area of the office.

At our work, it's probably 95% desktops here. Only people who need to travel use laptops. Also only if we need more monitors that the iGPU can handle (Ivy Bridge users wanting more than 2 monitors) have graphics cards. We're also close to the metal.

just brew it! wrote:

It is actually difficult to even get a desktop at my current job. Everyone uses laptops. The group I am in has a slightly different use case than most of the other developers ("closer to the metal"), and because of this it is useful to have systems we can plug PCIe cards and HDDs into. We ended up scrounging parts left over from a datacenter upgrade and built ourselves a few "Frankenservers" which now live under a table in our area of the office.

Nice.

Alot of our decommed kit is pre-Oracle SPARC or POWER5 based rack servers so lot of the legacy hardware is pretty much useless even for DYI server as the power consumption would cause our office's power bill to skyrocket (hence we're slowly phasing out SPARC and POWER in favor of x86).

srg86 wrote:

ultima_trev wrote:

I doubt any serious desktop user would be using iGPU outside of HTPC use.

I DO! Corei7 4790K with iGPU. Why? I don't play games but I still want a fast CPU. Rather than waste my money on a graphics card, went for higher CPU clock speed and 2 SSDs.

Pretty good odds I would be using an iGPU right now if there had been any AM3+ boards with decent iGPUs on them. The passively cooled GT 720 I'm using might even be slower than the current crop of iGPUs (I honestly don't know and don't care).

ultima_trev wrote:

As for iGPU users, those people are probably confined to laptops. I doubt any serious desktop user would be using iGPU outside of HTPC use. However there will be Raven Ridge, which is essentially Summit Ridge + iGPU coming within the next two quarters or so.

AMD may have a surprise hit on their hands with Raven Ridge. AMD video card drivers have always been CPU-performance sensitive, and it was clear from the benchmarks that Bulldozer was actually holding the 512-shader (and possible even the 384 shader) iGPU back.

Vhalidictes wrote:

AMD may have a surprise hit on their hands with Raven Ridge. AMD video card drivers have always been CPU-performance sensitive, and it was clear from the benchmarks that Bulldozer was actually holding the iGPU back.

There was also a significant lack of memory bandwidth to consider, at least on DDR3 parts.

LostCat wrote:

Vhalidictes wrote:

AMD may have a surprise hit on their hands with Raven Ridge. AMD video card drivers have always been CPU-performance sensitive, and it was clear from the benchmarks that Bulldozer was actually holding the iGPU back.

There was also a significant lack of memory bandwidth to consider, at least on DDR3 parts.

There sure was! But the scary part was that those APU under-performed even considering the limited bandwidth they had. DDR4 can only help as well.

I'll have to look and see if there are any DDR4 (AM4) APU benchmarks, but those products were DOA so I doubt it.

ultima_trev wrote:

I'm even more excited now to see what Ryzen 5 will bring!

Pretty much the same thing, clockspeeds won't go up much for the overclocks.

Will still offer more cores and threads for the cost than Intel, and all unlocked.

Will still be having overclocks top out around 3.9-4.1ghz. Maybe that moves up a few 100mhz, but the power increase will be drastic.

bfg-9000 wrote:

Intel could just convince Microsoft and other software developers to require a feature. After all, even the earliest Socket 423 Pentium 4 has SSE2 so can run today's Firefox, Chrome, Flash, and Office 2013, which is something not even the newest Socket A Athlons like Barton can do.

Then there's the CMPXCHG16b instruction suddenly required to run 64-bit Win8.1 or Win10, just to make it so no S754 or 939 A64 can run 64-bit software except in Win7.

Socket 939 wasn't the only platform that has ever been made obsolete by required support for new processor instructions, just the best known since it occurred in the middle of Microsoft's stumblings and bunglings with Windows 8. And the platform was ten years old at that point, with the newest CPUs being seven years old.

I'm not terribly excited by the current Ryzen chips. They don't make sense unless your tasks are exceedingly parallel, and that's a tiny niche of the desktop market.

ludi wrote:

And the platform was ten years old at that point, with the newest CPUs being seven years old.

There was also a solid technical reason. Having an atomic like that is a godsend.

Maybe picking Win 8.1 to require it was unwise, but that's really the only reason it stands out like you said.

Requiring it isn't a jerk move, it's incredibly convenient and useful and burning address bits etc... as a work-around is fugly.

ludi wrote:

bfg-9000 wrote:

Intel could just convince Microsoft and other software developers to require a feature. After all, even the earliest Socket 423 Pentium 4 has SSE2 so can run today's Firefox, Chrome, Flash, and Office 2013, which is something not even the newest Socket A Athlons like Barton can do.

Then there's the CMPXCHG16b instruction suddenly required to run 64-bit Win8.1 or Win10, just to make it so no S754 or 939 A64 can run 64-bit software except in Win7.

Socket 939 wasn't the only platform that has ever been made obsolete by required support for new processor instructions, just the best known since it occurred in the middle of Microsoft's stumblings and bunglings with Windows 8. And the platform was ten years old at that point, with the newest CPUs being seven years old.

And MS has a 32-bit version of Win10. MS is trying to have its cake and eat it too.

Sure, I can agree that the processor requirements will change over time. But all that does it change the question to "WTH is there a 32-bit version of Windows 10 in the first place?".

Vhalidictes wrote:

And MS has a 32-bit version of Win10. MS is trying to have its cake and eat it too.

Hmm. Interesting. cmpxchg8b works fine for 32-bits, but it doesn't for 64-bits.

Does 32-bit windows still work on S939?

Glorious wrote:

Vhalidictes wrote:

And MS has a 32-bit version of Win10. MS is trying to have its cake and eat it too.

Hmm. Interesting. cmpxchg8b works fine for 32-bits, but it doesn't for 64-bits.

Does 32-bit windows still work on S939?

You know, I'm morbidly curious, but I don't have a S939 based system to try it on, and haven't for over ten years now.

EDIT: The Interwebs report that Win10 x86 works fine on older Athlons. Score?