AMD Ryzen 7 1800X (14nm Zen)

[CAT-THE-FIFTH]

Well, I think that's taking it a bit far but I somewhat agree.

The Hexus review is the best I've seen thus far. What I do find interesting is the fact that AMD falls behind in 1080p but locksteps at higher resolutions.

Yes,I do tend to use colourful metaphors at times! The issue why I am so annoyed it is a solid CPU,but AMD generally launches solid enough products....but with a tendency to be somewhat slightly undercooked at launch and I would think that after so many launches where the same thing happened again and again,that they would learn by now.

I am really worried about Vega now - I really hope AMD launches the RX Vega in a polish manner,but I am just getting flashbacks to Fiji and Hawaii where they made 99% of the effort and it was that final 1% which did them in. Its what I feel about this launch.

[QuorTek]

It is a bargain that gives intel a run for their money

[imadaily]

@CAT-THE-FIFTH
Polishing products takes time and money, the further you want something polished the harder it is to polish it more.
AMD are cash and resource limited, they can't afford to spend too long doing it before release.

As for the SMT reducing performance I think that AMD telling me that the 8C/16T processor I just (Hypothetically) purchased will work best if I reduce it down to a 8C/8T processor would be a massive slap in the face.

As far as I know, enabling and disabling SMT is not a per application can be changed on the fly setting it's a 'reboot and change it in BIOS' setting.
I think it's much better for people to just get on with using it as a 8C/16T CPU and then down the line when Microsoft/AMD/Game devs work things out then you get a nice extra boost.

[CAT-THE-FIFTH]

@CAT-THE-FIFTH
Polishing products takes time and money, the further you want something polished the harder it is to polish it more.
AMD are cash and resource limited, they can't afford to spend too long doing it before release.

As for the SMT reducing performance I think that AMD telling me that the 8C/16T processor I just (Hypothetically) purchased will work best if I reduce it down to a 8C/8T processor would be a massive slap in the face.

As far as I know, enabling and disabling SMT is not a per application can be changed on the fly setting it's a 'reboot and change it in BIOS' setting.
I think it's much better for people to just get on with using it as a 8C/16T CPU and then down the line when Microsoft/AMD/Game devs work things out then you get a nice extra boost.

Its the first impressions which are important and so many major sites like Ars Technica are testing with SMT enabled which reduces performance - they literally are saying buy a Core i7 7700K. If those scores were between 5% to 15% higher it might change the tone of some reviews.

First impressions do count and the issue is AMD does want to convert all the less techy types who would only buy Intel to buy AMD. They are unlikely to be on forums or read comments,etc.

The problem is with SMT on,the performance drops are enough to crash performance down by upto 15% in games and you are still getting an 8 core CPU for between £320 to £500. The whole issue is we on forums might look at this think,yeah it will get better.

OTH,plenty of reviews don't mention this or test with SMT off,which makes performance look a bit meh,and it also means people might think the 4C/6T ones will be the same and just ignore Ryzen in its first iteration.

Its important that it is put out there,that the Ryzen SMT ability is not fully supported by games,and look at the 8C/8T performance - its quite reasonable. Games also generally tend to scale with cores than with virtual threads.

Review sites need to be aware of this for 4C/8T and 6C/12T parts before they are reviewed if there are not any fixes in the next few months for the performance regressions in many games.

But for me it really means I am not sure I am going to get Ryzen.

I have a Xeon E3 1230 V2/Core i7 3770 which I want to upgrade at some point. The 4C/8T Ryzen is meant to be between £175 to £200 and the 6C/12T one is £260.

The issue is that the 4C/8T one is now going to have to go core for core with a Core i5 7500 or 7600 and the latter has a clockspeed and IPC advantage.

AMD is a bit screwed in that area now - the Core i5 7600>>>Ryzen R5 1400X since the SMT on Ryzen does nothing for gaming.

It makes my plan of probably getting a 4C/8T Ryzen pointless now,since it will be a downgrade over my 4C/8T IB Core i7 since the SMT is close to useless in gaming. It will have to be the 6C one,but then I can get a Core i7 7700 for not much more and since I am a mini-ITX fan,Intel is better served in that area generally.

At this rate I might as well see how Coffee Lake fares.

[malfunction]

It looks like another scheduler type issue similar to bulldozer - windows / apps and games will need to know how to schedule / which core(s) to pick (affinity). Some theory here (if google translate is working):

http://www.hardware.fr/articles/956-...e-memoire.html

[Phage]

Hi Cat
I'm not so budget limited and rip a lot of discs to play when I'm on the road. I'm going to wait a month and see if things improve in the next 6 weeks, as that's when I'm expecting my tax refund.

[CAT-THE-FIFTH]

It looks like another scheduler type issue similar to bulldozer - windows / apps and games will need to know how to schedule / which core(s) to pick (affinity). Some theory here (if google translate is working):

http://www.hardware.fr/articles/956-...e-memoire.html

Will need to see if the first set of windows drivers/patches AMD said will drop in a few weeks will make a difference - I really hope by the time the 4C/8T models drop we still don't have negative SMT scaling in games,otherwise the only way they will compete is to have the max 4C/8T SKU at £175,as the Core i5 7500 will be quicker in games. Thank goodness for AMD,Intel blocked BCLK overclocking with Kaby Lake.

[scaryjim]

... We've not quashed anything and I gotta say mate you have a real crappy attitude over this ...

Hm, maybe that did come over as a bit angsty. OTOH

... I'm saying it because for a long while yet games are going to be limited to 4 cores being used ...

We're already seeing a number of games benefit from more than 4 cores - or at least more than 4 threads. That number will continue to increase.

... quite a few games will benefit from extra speed. ...

Hmmm, here's where your assertion gets tricky.

Firstly, there's the assumption that Intel *has* extra speed when you look at price equivalent chips. The vast majority of gamers won't buy Ryzen 7 or a Core i7 7700k. Look to the mainstream chips - Ryzen quads top out at $199, according to the so-far-very-accurate rumour mill, and that's for a 4C/8T chip with 3.9GHz turbo and XFR. At that price point Intel has multiplier-locked Core i5s with much lower clock speeds. Based on the comparison between the Ryzen 7 chips and the 7350k/7600k, they're close enough that a 5% - 10% clock speed advantage could actually make them faster in a lot of tests.

Secondly, there's the assumption that the CPU is a meaningful bottleneck. For benchmarking at 1080p with a GTX 1080 strapped in, sure the CPU is going to make a measurable difference. But a) that isn't what most people will be running, and b) a measurable difference isn't necessarily a noticable difference. The most powerful GPU used in my family is an RX 460, and we all play games - a faster CPU would do little to change our 1080p gaming experience.

So, we're looking at a situation where for most people their GPU is more likely to be a bottleneck than their CPU, and where at a given price point - unless Intel start hacking their list prices - it looks likely that AMD will provide roughly similar single threaded performance with either more cores or more threads. And that's without even considering overclocking - which AMD will be offering on all SKUs while Intel reserve it for only the most expensive chip in each range.

That's why I don't get why everyone is still touting the "Intel's the only choice for gamers" line. Everything that's been published today points to the sub £200 market being VERY competitive once Ryzen 3 & 5 launch.

[CAT-THE-FIFTH]

SJ,the problem is the 4C/8T SKUs are utterly screwed now - they need to get SMT sorted out in the next few months to work reliably with gaming,otherwise at £175 to £200 its going to be literally 4C Ryzen against 4C KL,and Intel will win that one.

If they can't get SMT to work reliably in games in the next few months,I think they should not bother with the 4C/8T version and just launch 4C Ryzen CPUs(it should be easier to get more to qualify as a 4C/4T SKU) and the 6C/12T at £220 to £260.

At least the people buying a 6C CPU will probably also want to do stuff other than gaming and it will be Intel 4C vs 6C just like the good old Core i5 750 vs the Phenom II X6 1055T.

Hi Cat
I'm not so budget limited and rip a lot of discs to play when I'm on the road. I'm going to wait a month and see if things improve in the next 6 weeks, as that's when I'm expecting my tax refund.

Thats the thing for that kind of stuff,Ryzen 7 looks pretty good TBH. Its where Intel will find it much harder to compete in with their quad cores.

[scaryjim]

SJ,the problem is the 4C/8T SKUs are utterly screwed now - they need to get SMT sorted out in the next few months to work reliably with gaming ....

The SMT thing needs more investigation, but there is at least one bit of good news for 4C Ryzen - have you read malfunction's link?

It looks like another scheduler type issue similar to bulldozer - windows / apps and games will need to know how to schedule / which core(s) to pick (affinity). Some theory here (if google translate is working):

http://www.hardware.fr/articles/956-...e-memoire.html

According to that, one of the biggest issues for Ryzen in gaming (and some other workoads) is actual extremely high latency and low bandwidth between the CCXes, which is exacerbated in moderately threaded situations by Windows 10 regularly moving threads between cores. If a thread gets moved and its data is now in the other CCX's L3, it'll end up with a cache miss and a huge latency penalty getting that data back in.

Assuming 4C Ryzen works by completely deactivating one CCX (which seems logical given the halving of L3 cache as well) that won't be a problem for it - there won't be another CCX for threads to get migrated to. So part of the problem may be mitigated inherently by the method of harvesting dies...!

EDIT: looking at the SMT scaling you posted here, it looks like Civ and GTA V are least affected, which I believe are the most CPU intensive games in that list? That would make sense if Windows 10 only moves threads in situations where cores are lightly loaded - put lots of load on the cores and no thread movement so no cache misses; lightly load the cores, more thread movement, more cache misses. That'd be easily fixable in driver or scheduler - simply tell the scheduler not to move active threads...!

[CAT-THE-FIFTH]

The SMT thing needs more investigation, but there is at least one bit of good news for 4C Ryzen - have you read malfunction's link?



According to that, one of the biggest issues for Ryzen in gaming (and some other workoads) is actual extremely high latency and low bandwidth between the CCXes, which is exacerbated in moderately threaded situations by Windows 10 regularly moving threads between cores. If a thread gets moved and its data is now in the other CCX's L3, it'll end up with a cache miss and a huge latency penalty getting that data back in.

Assuming 4C Ryzen works by completely deactivating one CCX (which seems logical given the halving of L3 cache as well) that won't be a problem for it - there won't be another CCX for threads to get migrated to. So part of the problem may be mitigated inherently by the method of harvesting dies...!

EDIT: looking at the SMT scaling you posted here, it looks like Civ and GTA V are least affected, which I believe are the most CPU intensive games in that list? That would make sense if Windows 10 only moves threads in situations where cores are lightly loaded - put lots of load on the cores and no thread movement so no cache misses; lightly load the cores, more thread movement, more cache misses. That'd be easily fixable in driver or scheduler - simply tell the scheduler not to move active threads...!

I hope that is the case TBH which is why I am still perplexed they couldn't wait another month until the windows patches dropped as they will most likely have some improvements to scheduling.

Motherboards are also a big issue currently - not only are motherboard companies scrambling to get the BIOSes upto scratch,the OEMs apparently have underestimated the number of CPUs AMD have made for sale too.

This has been needlessly rushed out.

[scaryjim]

... This has been needlessly rushed out.

I know you keep saying that, but the announcements and planning has been going on for months. I don't think it's been rushed as much as bungled - there was plenty of time to get everything lined up, motherboards were announced - what, last October when Bristol Ridge launched to OEMs? So why the BIOSes weren't ready is beyond me. AMD must've known how many chips they'd be putting into the channel weeks ago ... it's almost as if they've planned the CPU/engineering side of the launch, then suddenly realised at the last minute that they've forgotten and platform and software ecosystem to go with it...

[imadaily]

....That would make sense if Windows 10 only moves threads in situations where cores are lightly loaded - put lots of load on the cores and no thread movement so no cache misses; lightly load the cores, more thread movement, more cache misses...

If it's just lightly loaded cores that get shifted would there be some kind of benefit to artificially loading cores unused by the game just so that windows can't mess things around?
It'd be a super bodge job but possibly worth a go.

The ideal would be to find something that loads the required number of unused cores while using as little of the other CPU resources and memory bandwidth as possible...
Could make for an interesting benchmark: running a couple of instances of super pi in the background in an attempt to increase the in game performance!

I'm fairly sure there are reason that would mean that this wouldn't work but it would be interesting!

[CAT-THE-FIFTH]

I know you keep saying that, but the announcements and planning has been going on for months. I don't think it's been rushed as much as bungled - there was plenty of time to get everything lined up, motherboards were announced - what, last October when Bristol Ridge launched to OEMs? So why the BIOSes weren't ready is beyond me. AMD must've known how many chips they'd be putting into the channel weeks ago ... it's almost as if they've planned the CPU/engineering side of the launch, then suddenly realised at the last minute that they've forgotten and platform and software ecosystem to go with it...

What is another month when AMD already pushed it back another three months? I understand AMD needs to get a return on sales,but pushing out the platform before its ready is pointless.

With at least some scheduling patches,more mature motherboard BIOSes,and maybe even an extensive test to see if SMT impacted games,this would have been a better launch.

I mean there are not enough motherboards even available yet. You are more likely to buy a CPU than the motherboard to put it in.

AMD did this with the Fury X - launched it with poor QC and drivers which were not quite there yet. The RX480 was a success but the poor cooler did lose them sales to Nvidia.

Its probably why I am so irritated - its a continual thing we see with AMD.

Please,pretty,please AMD don't do this with Vega.

[CAT-THE-FIFTH]

A side note some of the AMD benchmarkers on XS,have look at overclocking in Ryzen. They have concluded that the R7 1700 is a low leakage part and the R7 1700X/1800X are high leakage. This is confirmed by reviews which show the R7 1700 at the same voltage consuming less power.

[kompukare]

If they can't get SMT to work reliably in games in the next few months,I think they should not bother with the 4C/8T version and just launch 4C Ryzen CPUs(it should be easier to get more to qualify as a 4C/4T SKU) and the 6C/12T at £220 to £260.

HT still has issues in gaming according to the hardware.fr numbers:
Average SMT/HT on versus off in games is -1.9% for Intel (i7-6800K) versus -8.9% for AMD.
The converse of that is that the average in apps is a gain in both cases with +22.5% for Intel and +25.6% for AMD.
So a really impressive showing for AMD's first try at SMT vs HT which Intel have been using for years (2008 in Nehalem if we ignore the P4 HT from 2002).

Speaking of Hardware.fr, it's nice that did a comparison at a constant 3.0GHz (pity they didn't include Skylake or Kabylake though although they might have to have forced them all to run as quad core for that to be meaningful).
http://www.hardware.fr/articles/956-...l-e-3-ghz.html


On average, for applications the uplift vs Piledriver is +131.9% which is impressive but Broadwell-E is +159.5%. The biggest outliner there seems to archiving where the quad channel makes such a big difference.

Still, gamers a picky bunch (who constantly seem to ignore that reviewers generally benchmark without any background applications), so it might make sense for AMD to release 5-1450. That is a 6C/6T part with 4.0GHz or more Turbo. The 5-1500 is rumoured to cost $230 (c. £220) and the 4C/4T parts (3-1200X and 3-1100) are cheaper than the 4C/8T parts by about $50, so such a 5-1450 part might be around £200.